Loin Pain Haematuria Syndrome Complicating Nehroptosis: Case Report

Loin Pain Haematuria Syndrome Complicating Nehroptosis: Case Report

Case Report

A 25 years old Saudi single female school teacher presented 9 years ago to the haematologist with iron deficiency anaemia (Haemoglobin was 8.3 gr/l. During her stay in hospital, she suffered from bilateral epistaxis. Mucoperiostal elevation cured her epistaxis after 7 bilateral cauterizations had failed to control her nasal bleeding. No bleeding or coagulation abnormalities were detected. She was also reported to have dysmenorrhea, dysurea and right loin pain. Attending physicians excluded genitourinary and other system abnormalities, Loin pain and microscopic haematuria reoccurred. Urological investigations including Urine analysis and culture, full blood count, renal function tests, grayscale ultrasound and intravenous urography (IVU) were repeatedly normal. No cause for her painful haematuria was found for 5 years but it was confirmed to originate from her right kidney on cystoscopy when bleeding was seen spyrting from the right ureteric orifice. Retrograde pyelography (RGP) was initially normal. She had repeated investigations done at various hospitals for her loin pain and haematuria.

These included computer axial tomography (CAT) scan, magnetic resonance imaging (MRI), digital subtraction arteriography (DSA) and isotope renography which were all normal in supine posture. Psychological disorders, opiate dependency and/or imaginary pain were thought on repeated admissions to be the cause of her undiagnosed and treated suffering. Psychiatric assessment excluded personality disorders and she was denied opiate therapy for years. Five years ago, right nephroptosis was suspected after palpating a mass at the right iliac foss on erect examination which disappeared on lying supine- unless the kidney mass was held down by gripping the loin above it. Nephroptosis was demonstrated on IVU with erect film. The kidney dropped >3.5 vertebrae from its normal position to the pelvis. A year later here haematuria progressed to frequent gross episodes of frank bleeding with intractable renal pain requiring frequent hospitalizations. During these attacks she was treated conservatively by fluid therapy, blood transfusions, antibiotics, opiates and bed rest. Thorough gynecological, nephrological, neurospinal, gastrointestinal and haematological investigations were repeatedly normal at various centres.

Two years ago, she underwent right nephropexy elsewhere but ptosis and haemauuria reoccurred. Even now the right kidney remains posted on erect IVU (Figure 1, upper segment). Repeated RGP showed atony and dilatation of right renal pelvis (Figure 1, middle segment- right). Bladder biopsy showed interstitial cystitis. Repeated CAT showed upper pole cyst and normal otherwise (Figure 1, lower segment). MRI was normal. DSA was normal apart from migration of the renal artery towards upper pole (Figure 1, middle segment- left)- being feasible only on supine posture. Right renal biopsy showed mesangial proliferative glomerulonephritis. Left renal biopsy was refused. DTPA isotope renographic study was also normal on supine posture. On repeating this study on sitting up posture, the right kidney showed impaired perfusion and filtration: GFR of 57.5 in the left kidney and 40.5 in the right kidney were demonstrated on supine posture. At sitting up posture, GFR of 63.8 in the left kidney and 22.2 in the right posted kidney were demonstrated. The normal expected value is 104 ml/minute. Such big deficit in differential renal function of the right kidney at posted position occurred in the abscence of organic ureteric stenosis or obstruction.

Figure 1: It shows 3 segments of upper, middle and lower.

Note: Upper segment shows IVU with supine film on the left and erect film on the right side. Right nephroptosis of 3 vertebrae is shown despite previous nephropexy. Middle segment shows DSA on the left that normal apart from migration of the renal artery towards upper pole. The RGP is shown on the right side and demonstrates pelvis dilatation. The lower segment shows CAT scan which is normal apart from upper pole cyst.

These findings suggest diminished right arterial blood flow and GFR on upright posture, possibly due stretch narrowing (uniform stenosia) of the renal artery. Diminished vasculature of the right kidney, as compared to the left, may be observed on DSA- despite being done at supine posture (Figure 3). Frequent crisis of the loin pain haematuria syndrome (LPHS) have continued during the last 4 years, requiring countless hospital admissions. She has been so crippled by her LPHS that she lost her job as school teacher. She carries a bag full of useless medications and a list of prohibited nephrotoxic drugs. She refused surgical treatment of renal sympathetic denervation and nephropexy surgery which has proved successful in other cases.

Discussion

The natural history of this case was followed up from the initial onset of loin pain to occurrence of recurrent microscopic haematuria, to diagnosis of symptomatic nephroptosis (SN), to the development of LPHS [1]. To my knowledge this is the first case report of LPHS to complicate SN. It is certainly not the last [2]. It demonstrates many of the problems encountered in the diagnosis and therapy of both SN and LPHS. All textbooks do not index SN as it was disparaged long ago [3] but mention LPHS. Chance diagnosis of SN on supine imaging is unlikely. So, SN has become a universally forgotten diagnosis [3]. Vascular anomalies, ischaemic renal scarring and messengial proliferative glomerulonephritid are features of LPHS [1]. Most of these renovascular complications of LPHS are also documented in SN [4-7]. Despite known overinvestigations, no erect imaging was ever done in LPHS. Previous reports on erect artieriography [5-7] and isotope renography [7] demonstrated these complications in SN. Recurrent stretch of renal vessels in erect posture causes renal artery elongation and narrowing in the presence or absence of stenostic lesions [5].

Transient ischaemic attacks (TIA) of renal pain or “renal angina” occur in the absence of organic lesions for years. Chronicity establishes fibro-muscular dysplasia [5,6] manifesting with the organic ischaemic renal complications of LPHS [1,2] and pelviureteric atony- as shown in the case reported here. To salvage a kidney with such irreversible damage by sympathectomy and/ or auto renal transplantation is extremely hard, and 75% of cases end up with nephrectomy [8]. Our patient and I are resenting nephrectomy which though cures pain and haematuria and saves what is left of her miserable life is hazadous. This is because of a possible involvement of the left kidney too. Conservative treatment is neither effective nor safe. My patient, being a Muslim, has found relief and comfort in her prayers, particularly at so good position or the knee-chest position. The spiritual and gravitational benefits of this position are obvious. This is an example of how the weak and helpless find power to resist an unbearable situation. Other patients like her become depressed by the prolonged undiagnosed and untreated suffering [1,4] and may even commit suicide [3].

Her painful episodes of haeematuria are also known “Dietl crisis” of SN, who also advised knee-chest position for temporary relief. This tragic case represents the tip of an iceberg. Experience at King Khaled Hospital in Najran Saudi Arabia with 190 cases suffering from SN, of whom LPHS complicated SN in 36 (18.9%) of patients [8,9]. This figure represents cases with gross haematuria only. It is higher when cases with microscopic haematuria are included. Making an early diagnosis of SN by erect IVU may upright issues and brings relief to the unfortunate sufferers. Based on recent reports, SN has proved to be a preventable cause of LPHS when the correct surgical procedure is done timely. The surgery of renal sympathetic denervation and nephropexy has proved curable for Both SN and LPHS. Both nephropexy for SN and renal sympathetic denervation of the LPHS were reported separately [10,11]. Modern favourable results of renal sympathetic denervation and nephropexy (RSD&N) have proved 100% success rate for both SN and LPHS [8,9]. Nephroptosis was disparaged and nephropexy was abandoned >70 years ago, both have been deleted from surgical and urological textbooks and have become universally forgotten. This is because of the many problems in diagnosis and therapy that have been recently addressed and resolved [10,11].

More BJSTR Articles: https://biomedres01.blogspot.com/

Emerging Evidence of Mast Cell Involvement in Oral Squamous Cell Carcinoma

Emerging Evidence of Mast Cell Involvement in Oral Squamous Cell Carcinoma

Mast Cells – as Sentinel Cells

Mast Cells (MC) are tissue-resident, bone-marrow derived, granular sentinel cells. They enter circulation from the bone marrow as committed Mast Cell Progenitors (MCP) and undergo terminal differentiation in response to a stimulus. MC and their progenitors are found throughout connective tissues and the mucosae, though the highest MC densities are found in the skin, airways and digestive tract, that is, host-environment interfacescongruous with their role as sentinel cells [1].Two distinct subpopulations of MC have been identified in human tissues, classified according to the protease contents of their secretory granules-tryptase-only containing type, referred to as MCT, and chymase- and tryptase-containing type, referred to as MCTC [2]. The latter also contain other proteases, including cathepsin G and carboxypeptidase A. In addition to these proteases, MC possess an arsenal of effector molecules. These include molecules stored in secretory granules, such as heparin, serotonin, and histamine, and those which are synthesised de novo, such as leukotriene C4, prostaglandin D2, platelet activating factor, and an assortment of cytokines [3]. MC activation may precipitate degranulation, or only the release of select effector molecules, and may occur in response to IgE crosslinking, complement activation or certain toxins [4]. MC effectors may contribute to both physiological and pathological events. While these cells are well known as orchestrators of hypersensitivities and immune reactions, more recently the role of MC in modulating tumour growth-both positively and negatively – has become a focus.

Tissue Distribution, Activation and Migration of Mast Cells in Oral Cancer

The altered tumour microenvironment induces changes in mast cell migration, activation and tissue distribution. Mast Cell Progenitors in peripheral tissue are under the influence of tumour-derived cytokines and chemoattractants controlling their maturation and migration. Stem Cell Factor (SCF), derived from the tumour cells, and its receptor c-Kit represent possibly the most significant and potent, and best characterised, mast cell chemotaxis and migration pathway. An increase in mast cell infiltration in areas of Oral Squamous Cell Carcinoma (OSCC) development has been demonstrated in a number of studies [5-12], and correlation between Mast Cell Density (MCD) and disease progression has also been reported [13-15]. In a slightly different vein Aromando, et al. [16] noted no change in total mast cell numbers in a hamster cheek pouch tissue during experimental carcinogenesis, but rather a decrease in MC in the adventitious tissue and an accumulation of MC in peritumoural and intratumoural stroma, as well as a reversion of the ratios of active/inactive MC in favour of the former [16]. Furthermore, differences in MCD between intratumoural and peritumoural stroma were evaluated, showing significantly higher MCD values in the peritumoural stroma than intratumoural, probably reflecting the migration of the cells from adventitious tissue as well as functional roles in Extracellular Matrix (ECM) degradation and induction of cell proliferation [16].

Further characterisation of MC subpopulations within the Intratumoural (IT) and Peritumoural (PT) stromal regions has shown that, while both MCTC and MCT counts are significantly increased throughout the tumoural stroma, MCTC type predominates in the PT stroma, while MCT subtype predominates in the intratumoural stroma [17]. The authors hypothesise that the distribution of subpopulations reflects functional requirements: MCTC contain chymase, which plays a role in activation of Pro-Matrix Metalloproteinase-2 (MMP-2) and Pro-Matrix Metalloproteinase-9 (MMP-9) to their active MMP-2 and MMP-9 forms, respectively [18]. Both MMP-2 and MMP-9 possess the capacity to degrade type IV collagen [19], a significant component of the basement membrane and barrier to tumour invasion. Hence, the localisation of MCTC at tumour peripheries suggests an ECM remodelling role for these cells. Similarly, MCT predominance in the IT stroma suggests a role of these cells and their potent angiogenic mediator, tryptase [20], in neovascularization of the tumour. In contrast, some studies of OSCC and other carcinomas have failed to demonstrate a statistically significant increase in MCD in tumour regions, or a decrease in MCD as the degree of differentiation decreases [21-23]. These results have on occasion been attributed to massive degranulation of the MC compromising visualization, or shortcomings in specificity of the toluidine blue staining protocol compared with antitryptase immunohistochemistry [23-,25]. Other studies have described a significant reduction in MC in cancers compared with controls [12,25].

Putative causes of this decrease are tobacco exposure, shown in an experimental carcinogenesis model to accentuate a decrease in MC infiltration in tumours caused by 4-NQO [26], or a failure in migration of MC, indicated by the significantly decreased c-kit+ MC/ tryptase+ MC ratio, compared with control, attributable to changes within the tumour microenvironment [25] (the c-kit receptor on MC, together with its ligand SCF, are responsible for the migration, activation and maturation of MC [27]). However other authors have found no significant difference in c-kit+ /tryptase+ ratio [25], suggesting no migration failure in such a case. Additionally, MC are present in neoplasms irrespective of the presence of inflammatory infiltrate, suggesting chemotactic pathways are selective for MC. Transforming growth factor-beta (TGF-β) is synthesised and released by MC, and is increased in OSCC [5]. Its local roles are pleiotropic, including: its initially cytotoxic, but progression of the pathology. There lacks, however, correlative data in the literature between mast cells and clinicopathological features such as tumour size, regional nodal involvement, or metastasis. Eventually cytokinetic role in tumourigenesis [25] its action as a potent chemotactic factor for MC [28]; its role in angiogenesis; and its supposed part role in mediating a phenotypical change in tumours from CD34+ fibrocytes to alpha-smooth muscle antigen+ (α-SMA+ ) myofibroblasts [5].

Mangia et al. [29] report that tryptase can also induce phenotypic shift from CD34+ /α-SMA- fibroblasts to CD34- /α-SMA+ myofibroblasts, lending further credence to the role of MC in this context. CD34+ fibrocytes express Granulocyte Macrophage ColonyStimulating Factor (GM-CSF) [30], which downregulates CD117 (c-Kit) expression in mast cells [31]; hence a phenotypic shift away from CD34+ fibrocytes as they differentiate to alpha-SMA myofibroblasts, decreases repression of CD117 expression and consequently, allows MC migration and infiltration [5].

Mast Cells in Oral Cancer and Angiogenesis

Angiogenesis and neoangiogenesis are the processes of formation of new blood vessels from pre-existing blood vessels, and formation de novo, respectively. Tumour proliferation is limited by oxygen perfusion, and tissue oxygen perfusion greater than 2mm has been reported to be prohibitive of tumour growth [32]. Neovascularisation, therefore, is a process central to tumour growth and development, and has been implicated in dissemination and metastasis. Mast cells store and have the capacity to synthetise a number of angiogenic and neoangiogenic mediators, including angiopoietin-1, FGF-2, VEGF, IL-8, TGF-β, TNF-α, histamine, heparin, tryptase and chymase, among others [33]. These mast cell mediators can act at various stages of angiogenesis including degradation of the ECM, migration and proliferation of endothelial cells, formation and distribution of new vessels, synthesis of ECM and pericyte mobilization [34,35]. It has been shown that during the initiation of angiogenesis, mast cell tryptase promotes ECM degradation through the activation of MMPs and plasminogen activator [36]. One quantification of the degree of tumour vascularisation is microvessel density. A number of studies have correlated mast cell densities with Microvessel Densities (MVD) in oral cancers [10,32,37- 40].

Moreover, mast cell densities have been shown to increase with MVD as disease progresses, or degree of tumour differentiation decreases [41]. The distribution of MC within tissue is also indicative of functional roles. As described in section 2, Rojas et al. [42] characterized MC subpopulations in OSCC and determined that MCT were the predominant subtype in the intratumoural stroma, while MCTC were in the peritumoural stroma of OSCC. Separate studies have shown increases in MVD in the intratumoural stroma, while no significant increase was observed in peritumoural regions. This supports the hypothesis of Rojas et al. [42], that MCT are so localised for an angiogenic role via the known potent angiogenic factor, tryptase [41]. It has been shown that during the initiation of angiogenesis, mast cell tryptase can also promote ECM degradation through the activation of MMPs and plasminogen activator [36]. Data reporting the colocalisation of MC and blood vessels in oral cancer also suggest an intimate relationship between the pair and role in angiogenesis for MC in tumourigenesis [10]. While MC are implicated in neovascularization and are known to contain angiogenic factors, mechanisms are uncertain. The role of the potent angiogenic cytokine Vascular Endothelial Growth Factor (VEGF) contained in MC and released from tumour cells is not straightforward.

Several studies correlate MC with VEGF expression. Release of VEGF from MC is mediated in part by the interleukin-33/ST2 signalling axis [44]. IL-33 is upregulated in OSCC, and a correlation between IL-33 and MVD, as well as IL-33 and MCD has been reported, suggesting VEGF may be the intermediary [39]. However, the specific role of VEGF in MC-mediated angiogenesis is not clear. Artese et al. [46] showed that, in OSCC, while MVD was significantly increased in tumours and correlated with tumour grading, VEGF expression did not vary between OSCC and tumour-free controls, and no significant correlation between VEGF expression and MVD was observed. Carlile et al. [47] produced similar results in a similar study, noting that vascularity increased in OSCC compared with control, though this increase did not correlate with VEGF expression. Other studies have found significant VEGF expression increase in OSCC vs control, although VEGF expression did not directly correlate to MVD [32]. MCD did, however, correlate with MVD. A single-linkage cluster analysis on these three variables (MCD, MVD, VEGF expression) grouped VEGF expression and MVD, implying an indirect link between the two variables. Whereby, VEGF may recruit mast cells [48] which contain angiogenic factors [32,49]. Hence, VEGF indirectly induces angiogenesis. A more prominent and directly acting role for tryptase is therefore suggested [50,51], seems to be confirmed by the results of Rojas et al. [42]. Yet interrelationship between MC and angiogenesis is not however a universal finding and we must state while the evidence is significant there remain many inconsistencies. Some authors, while finding a significant increase in MVD in SCC vs control, did not observe a correlation between MVD and MCD [52]. Other data as much as show the converse of the above – that is, increasing MVD inversely and significantly correlated with MCD [52].

Mast Cells in Extracellular Matrix Remodelling in OSCC

An important feature of cancer progression is the ability to degrade the Extracellular Matrix (ECM), and consequently permit proliferation and migration of cells, invasion of surrounding tissues and dissemination. Matrix Metalloproteinases (MMP) are endopeptidases responsible for degradation of the ECM. The MMPs can be categorized according to their preferred substrates, i.e. the collagenases (MMP-1, -8, -13), gelatinases (MMP-2, MMP9), stromelysins (MMP-3, MMP-10, MMP-11). Human mast cells interact with several MMP, both stimulatory and inhibitively, in ECM homeostasis [53]. Gelatinases A and B (MMP-2 and MMP-9, respectively) are secreted by MC [54,55], or indirectly activated by MC-secreted chymase [54]. MC tryptase itself has also been shown to directly exert gelatinase-like activity [56], and tryptase is also involved in the processing and activation of MMP-3 and MMP-1, the latter being dependent on the activation of the former [57,58]. Chymase is also capable of directly activating MMP-1 and MMP-3 [59]. Further MC chymase, but not tryptase, may directly cleave procollagen to fibril-forming collagen [60]. Hence MC contribute both directly and indirectly to processes which degrade the ECM.

In the context of oral cancer, MMP-9 expression has been shown to be upregulated in OSCC compared with healthy tissues, and significantly correlated with MCD [61]. Another study showed lip SCC samples that expressed higher MC counts also showed increased collagen degradation, assayed by picro-sirius staining [7]. MMP-9 has been associated with aggressive tumour growth, proteolytic processing of the ECM and activation of cytokines (such as TGF-β) [10]. MMP-9 is capable of processing type IV collagen of the basement membrane [62] and other ECM components, which are key events in tumour invasion and metastasis (see Fig. 1). However, evidence supports a fluctuating role of MMP-9 in OSCC. High MMP-9 expression has been shown to correlate with nodal involvement and metastasis, and poor prognosis in OSCC [63]. Meanwhile, Guttman et al. [64] reported no correlation between MMP-9 and tumour size or nodal involvement. Similarly, other authors reported that MMP-9 expression was not associated with clinical variables, such as tumour stage, recurrence rate, etc. [65]. Other data suggest that MMP-2 and MMP-2 expression significantly correlates with collagen degradation and local invasiveness, though this was not related to metastatic potential of the disease [66]. Meanwhile, it has been suggested that although MMP-2 and MMP-9 expression is high in OSCC, the ratio of active/inactive MMP-9 is low, suggesting MMP-2 is the gelatinase of greater importance in OSCC [67]. Conversely, MCs have also been implicated in collagen deposition. Vidal et al. [10] observed the accumulation of MC in areas of fibrosis surrounding malignant minor salivary gland tumours and proposed the hypothesis that ECM remodelling, specifically collagen synthesis, may be mediated by MC.

A similar hypotheses have been made regarding odontogenic tumours [68] and breast cancers, in which it was suggested that tryptase played a role in collagen deposition [69]. Additionally, an association between MC and fibroblasts in the potentially malignant condition, oral submucous fibrosis, has been inferred [70,71]. Supporting these observations, MC products TGF-β and tryptase have been shown to stimulate collagen deposition, fibroblast migration and fibroblast proliferation [20,72-77]. Hence, ECM homeostasis is a delicate balance between factors, which promote degradation, and proliferation. The tumour microenvironment may lead to unpredictable disruption of this balance.

Mast Cells and Tumour Proliferation, Invasion and Dissemination

Mast cells can precipitate mitogenicity in tumour cells directly through mediators, and indirectly through microenvironment modulation. The most abundant mast cell protease, tryptase, is implicated in promoting neoplastic cell proliferation via the Receptor Tyrosine Kinase (RTK) protease activated receptor 2 (PAR2), expressed on the surface of neoplastic cells. Tryptase cleaves and activates PAR-2, stimulating proliferation of receptor-bearing cells, as well as inducing the expression of cyclooxygenase-2 (COX-2) [16,77]. The proliferative consequence of tryptase-mediated PAR-2 activation has been reported in lung tissue, colon cancer and breast cancer [76,78], but few studies exist correlating MC with tumour cell proliferation in OSCC, and those that fail do to demonstrate a significant correlation [7]. A study pertaining to the potentially malignant oral condition actinic cheilits has, however, quantified COX-2, PAR-2, MC and tryptase in human actinic cheilitis tissues. COX-2 is responsible for eicosanoid biosynthesis from arachidonic acid, and among the metabolites is Prostaglandin E2 (PGE2), which is also capable of promoting tumour proliferation [79]. The authors reported a significant correlation between tryptase-positive MC and PAR-2 expression, as well as COX-2 overexpression, inferring a role for tryptase in PAR-2 activation and COX-2 overexpression. Increased MC counts have also been associated with higher levels of DNA synthesis in an experimental hamster oral carcinogenesis model, again implicating tryptase-mediated PAR-2 activation [16].

Conclusion

Mast cells are influenced by, and influence, malignant tumours. They commonly accumulate in tumour-associated stroma and can promote tumour proliferation and aggressiveness via a plethora of secreted molecules. However, the literature is divided on the clinical significance of these local effects, and if or how they may represent or open up therapeutic possibilities. This variability in evidence suggests that a more complex set of interactions exists and overbears the net outcome and effects of mast cells in tumours. That is, while the mechanistic role of mast cells in cancer is becoming clearer and is seemingly becoming important, it is still incompletely understood. As such, conclusions about the role of mast cells in oral cancer would be fallacious to draw despite evidence correlating their accumulation with factors correlated with tumour progression.

Acknowledgement

a) Conflict of Interest: The authors declare that there are no conflicts of interest related to this letter. No external funding, apart from the support of the authors’ institution, was available for this study.

Author Contributions: Basha S and Hassan NMM study concept, design, and preparation of the manuscript. Akhter R, Ibaragi S, Cox S and Sasaki A: help in manuscript preparation.

More BJSTR Articles: https://biomedres01.blogspot.com/

Does Technique of Mesenchymal Stem Cells Perineural Migration Seem Promising in Technologies of Combined Therapy with Stem Cells?

Does Technique of Mesenchymal Stem Cells Perineural Migration Seem Promising in Technologies of Combined Therapy with Stem Cells?

Introduction

The progress of modern medicine is also associated with highcost technologies, expensive creation of medicines, and unique ways of transplantations of inner organs, which are available only for selected people [1]. However, statistics even in advanced countries shows low success in treatment of socially important diseases [1,2]. Mortality due to cardiovascular diseases and oncological pathology remains high. So, classis and even innovative therapeutic and surgical techniques of fatal pathological processes treatment appeared to be low-effective. All realistic people think that such situation cannot be accepted. It is fortunate that there are modern Don Quixote’s who try to struggle against certain diseases using alternative ways of treatment but not against windmills. It is important that smart part of modern Don Quixote’s does not reject existing protocols of treatment. They try to broaden possibilities of these methods by new ideas. Researchers create such innovative techniques which complement classic ones [3-5]. What did we really achieve till nowadays?

We will focus attention on treatment techniques using stem cells (SCs) in this article. Discovery of SCs perplexed scientists. It was found that organs and tissues have potential to recover damaged tissue areas. But why such potential appeared to be so weak? We always stare in envy at scientific films about recovery of damaged tails in lizards. But reparative processes involving endogenous SCs are low-effective in humans. Researchers tried to combine cell therapy with standard protocols of treatment-and the hope appeared on the horizon. Unfortunately, all good hopes were destroyed by data on follow-up results. First positive outcomes of such combined therapy were of short duration. SCs did not want to survive in tissues. That is why there were weak positive effects of cell therapy noticed in long follow-up periods. Researchers tried to improve the situation by using scaffolds, exosomes and techniques of directed SCs movement caused by magnet and ultrasound [3-5]. Also new methods of combined therapy using stem cells and macrophages were developed [1]. It was based on join role of both stem cells and macrophages in the processes of immunosuppression, tissue remodelling, angiogenesis [1].

Advantages of Perineural Way of Autologous MSCs Injection for Recovery of Impaired Functions of Human Tissues and Organs

Authors enthusiastically began experimental analysis of MSCs functions in the recovery of impaired brain functions [6-10]. Majority of scientists apply systemic ways of MSCs administration, but this leads to inconsistent results and does not satisfy neither patients, nor doctors. The problem of blood-brain barrier overcome is still unresolved. Its protective function is amazing – take at least observation when post mortal injection of trypan blue into brain major vessels does not allow identifying this stain in brain tissue within several hours [11]. Administration of MSCs via lumbar puncture does not consider rostro-caudal flow of cerebrospinal fluid. Direct injection of MSCs into damaged brain region followed by trepanation is frequently unreasonable due to additional surgical interventions. Authors chose technique of MSCs perineural migration in combined therapy of brain injuries and strokes with SCs [6-10]. Cranial nerves were chosen as the way for MSCs migration. Injection of MSCs into cranial nerves’ endings in facial area is quite simple surgical procedure which guarantees MSCs migration to damaged brain regions [6-10].

Such manipulations were performed in line with approved by the Ministry of Health of the Republic of Belarus treatment protocols in 40 patients with ischemic and hemorrhagic strokes [12,13]. Only autologous MSCs were used [12,13]. Authors considered well known fact that MSCs actively move in the living organism to damaged tissue or organ. Such migration is triggered by multiple signaling molecules of protein and other nature which are expressed in damaged tissues. This observation is very promising for bioprinting technology [14]. Authors experimentally proved effectiveness of such technique on laboratory animals (Figure 1) when MSCs were injected into perineural space of vagus nerve at the level of thyroid cartilage leading to recovery of impaired heart functions [15,16]. The technique (Figure 1) was further developed and used for MSCs delivery from perineural space of vagus nerve to damaged areas in liver, pancreas, stomach and small intestine [17]. Such method of cell therapy guarantees targeted perineural migration of MSCs to certain damaged organ. Therefore, there is no more need of using major surgery to implement technique of MSCs migration.

Figure 1: Scheme of perineural MSCs migration after unilateral vagotomy and perineural implantation of MSCs into intact vagus nerve on the other side. Arrows indicate MSCs migration.

Conclusion

Obtained experimental data and clinical outcomes speak for technique of MSCs migration as promising method for activation of reparative processes in damaged areas of inner organs. Future development of combined cell therapy with standard treatment ways should consider using only autologous MSCs together with individual characteristics of patients and optimal way of MSCs perineural delivery to damaged area.

Acknowledgement

This pooled analysis was funded by OOO “Synergy”, and by SSTP “New methods of medical care”, section “Transplantation of cells, tissues and organs” (2016-2020).

Conflict of Interest

All listed authors concur with the submission of the manuscript; all authors have approved the final version. The authors have no financial or personal conflicts of interest.

More BJSTR Articles: https://biomedres01.blogspot.com/

The Introduction of Designing a Hybrid Brain Computer Interface System

The Introduction of Designing a Hybrid Brain Computer Interface System

Introduction

A Brain-Computer Interface (BCI) system can communicate without movement based on brain signals measured with Electroencephalography (EEG). BCIs usually rely on one of three types of control signals: The P300 components of the Event-Related Potential (ERP), Steady State Visual Evoked Potential (SSVEP), or Event Related Desynchronization (ERD). Research about BCI has been widely developed over the past few decades. BCI systems are used in various areas. However, different BCIs have their own advantages and disadvantages. In order to improve the performance of BCIs, Pfurtscheller et al. [1] proposed the hybrid BCI system increasing advantages and reducing disadvantages from different BCIs. A hybrid BCI system is composed of two different BCIs, or at least one BCI and another system. A hybrid BCI must fulfill the following four criteria like any BCI:

a) The device must rely on signals recorded directly from the brain;

b) There must be at least one recordable brain signal that the user can intentionally modulate to effect goal-directed behavior;

c) Real time processing; and

d) The user must obtain feedback [1].

A hybrid BCI can process different inputs simultaneously, either sequentially or in parallel. In the past decades, different kinds of BCI Speller systems have been researched in order to increase communication of disabled people with their peripheral environment and typing text rapidly. As a result, the conventional single modality of BCI speller systems have progressed in many ways while there are several problems as follows:

a) Most of the developed BCI speller systems are singlemodality.

b) Most of the developed BCI speller systems are experimented with expensive EEG devices such as BrainAmp DC amplifier.

c) There are limitations for people who have to use BCI speller because they use a special sentences just in some places like their home or their room in hospital while they need convenient and portable device for all places to spell everything to short time. In recent years, with the advancement of Hybrid BCI systems, many previous problems have been solved and also instead of using expensive devices, inexpensive and convenient devices have been replaced such as EMOTIV Epoc.

History and Literature Review

Before proceeding with any project, knowledge of the related work done in the past is very important. In this chapter, we review the related work of hybrid BCI systems and focus on a specific type of hybrid BCI systems that is related to the work in this thesis, i.e. the P300-SSVEP hybrid BCI systems.

P300-MI Based BCI

A possible combination for a hybrid BCI is P300 and motor imagery-based BCI. The basic concept in this type of hybrid is based on the features of P300 and ERD/ERS in control applications. P300 is a reliable BCI type for selecting one item out of several items and can be used for discrete control commands. On the other hand, due to the low degree of freedom presented by MI-based BCI, this type of BCI is more efficient for continuous control commands. These two types of BCIs can be joined to present more complicated control commands in one task. In [2], for controlling a wheelchair in a home environment, several approaches using different BCI techniques were introduced. The wheelchair control commands were divided into three steps; destination selection, navigation and stopping command. In first step, the user should select the destination of the wheelchair motion by selecting one of the items through a list of destinations by a P300 BCI presented at a screen.

The experiments on healthy subjects showed a response time of about 20 seconds, the false acceptance rate 2.5% and the error less than 3%. The results of this task showed that P300 was a suitable item for the interface. In Second step, the destination was selected and the wheelchair started its motion toward the destination following virtual guiding ways. A proximity sensor was considered for stopping the wheelchair facing obstacles. In third step, two methods were presented. The first method was the fast P300, in which, on the screen, there is only one item “The Stop” and the task is the detection of user’s intention. Experimental results showed reduction in response time. The second method was to use a MIbased BCI. The position of cursor was considered for presenting the visual feedback for the MI-based BCI system and control of the cursor was based on an arm movement imagination. Experimental results showed approximately the same response time as the fast P300 method, but for false acceptance, a rate of zero was achieved. Since the low false acceptance rate and fast response are the most important needs for this type of BCI, it seems that MI-based BCI is a more reliable system for this application.

In [3], P300 and MI were introduced to be components of the hybrid BCI in robotic control decision applications. Parallel and asynchronous classifications were introduced. The system task was to detect the intended pattern. Classification accuracy of hybrid model was evaluated during the experiment. Sixty trials were presented to four subjects: thirty trials for P300 presentation and thirty trials for MI. During the second thirty trials, the P300 stimuli were also presented but the subjects were not supposed to pay any attention to the stimuli. Empirical results indicated that subjects could achieve good control over the hybrid BCI. In particular, subjects could switch spontaneously and reliably between the two brain activity patterns. The hybrid classification reached an average P300 classification accuracy of 82% and the hybrid system reached an average MI classification accuracy of 71%. In conclusion, the performance of the hybrid system allows for the reliable control of devices such as robots.

SSVEP-Mi Based BCI

In [66], the proposed hybrid BCI was evaluated during the task and was applied under three conditions: MI-based BCI, SSVEPbased BCI, and MI-SSVEP based BCI. During the MI-based BCI task, two arrows appeared on the screen. When the left arrow appeared, subjects were instructed to imagine opening and closing their left hand. For the right arrow, subjects imagined opening and closing the corresponding hand. In the SSVEP task, subjects were instructed to gaze at either left (8Hz) or right (13Hz) LED depending on which cue appeared. In the hybrid task, when the left arrow was showed, subjects were imagining the left hand opening and closing while gazing at the left LED simultaneously. The task was similar to the right arrow (Figure 1). Results show the average accuracy of 74.8% for MI, 76.9% for SSVEP, and 81.0% for hybrid. The number of subjects who achieved less than 70% accuracy [4], reduced to zero from five using the hybrid approach.

Figure 1: Example of a 6 × 6 User Display in P300 Speller [12].

Materials and Methods

The objectives of this research consists of two parts: The first objective of this BCI research is to design a simple hybrid BCI platform that translate disabled people’s intentions into a control signal for an external device such as a computer (Figures 2 & 3). With this platform, a patient can select items from a (6×6) matrix on the screen in order to spell characters or phrases. To achieve this, we use two-modality (Steady state visual evoked potential & P300 Potential) sequentially for detecting desired character and we expect to improve the performance of the single-modality (only SSVEP or P300) BCI systems. The second objective of this BCI research is to use a low cost device for recording brain signals instead of expensive device that gives us good performance relatively.

Figure 2: An Illustration Graph of SVM that Fined the Optimal Hyperplane (solid line) to Separate Two Classes by Maximizing the Margin (ᵞ) [13].

Figure 3: General Diagram of Project Implementation.

Experimental Setup for Dataset II of Berlin BCI Competition III P300 Speller

We performed initial experiments with the P300 datasets from the BCI competition 2004 [5] to test and compare the efficiency and reliability of our P300 speller-based BCI approach. First, the competition dataset description is presented, and then is discussed; the experiment step followed by data preprocessing and feature extraction step, and classification training and validation step.

The Competition Dataset Description

This competition dataset indicates a complete record of P300 evoked potentials recorded with 64 channel EEG according to 10- 20 system by BCI2000 software. The user was presented with a 6 × 6 matrix of characters as shown in Figure 4. The user should focus attention on characters in a word (for example one character at a time). All rows and columns of this matrix randomly intensified at a rate of 5.7Hz. Two out of twelve intensifications of rows or columns contained the desired character (i.e., one particular row and one particular column). The responses evoked by these infrequent stimuli (i.e., the 2 out of 12 stimuli that did contain the desired character) are different from those evoked by the stimuli that did not contain the desired character and they are similar to the P300 responses reported originally by Farwell et al. [6]. The signals are collected from two subjects (A and B) in five sessions and each session includes a number of runs. In each run, the subject focused attention on a series of characters. Afterward, each row or column in the matrix is intensified randomly for 100ms. After intensification of a row/column, the matrix was blank for 75ms. Row or column intensifications were block randomized in blocks of 12. The sets of 12 intensifications repeated 15 times for each character epoch. For example, any specific row/column is intensified 15 times, thus there were 12×15 =180 intensifications totally for each character epoch that 2×15 = 30 of them is caused P300. Each character epoch was followed a 2.5 s period and during this time, the matrix was blank. This period informed the user that this character was completed and the user could focus on the next character in the word that displayed on the top of the screen. As mentioned above, the goal in the competition is to use the training data with label of both subjects A and B to predict the character sequences in the test set of each subject. Finally, the recorded data converted into four MATLAB. Mat files, one training with 85 characters and one test with 100 characters for each of the two subjects A and B.

Figure 4: First Proposed SSVEP Visual Stimuli Platform.

Data Processing and Feature Extraction

Before processing the signals, we should preprocess them because we only interest in part of the EEG signals occurring after each intensification and we want to build some features that can be fed to a classifier. At first, since the evoked potentials appear about 300ms after the stimulus, we understand that this window is large enough to get all needed time features for a classification. Initial sampling frequency was at 240Hz. We filtered with an 8-order band-pass filter (Chebyshev Type I) which cut-off frequencies are 0.1 and 10Hz and we down sampled to 24 samples. After filtering, we normalized the data to an interval of [-1, 1]. We used only one channel (Cz electrode) through the 64 channels to extracting the stimuli for plots and analysis. To perform the classification, we averaged all 15 intensifications and showed averaged P300 responses over Cz channel among all training samples. The data is labeled as a target or non-target set corresponding to the matrices including the event cue onset such as Stimulus-Code, Stimulus-Type and Flashing prepared with the datasets. If the averaging is applied, datasets are averaged based on target and non-target bases. Due to facilitating and avoiding the errors, the signals should have divide into the sections that start to intensify in one row/column and to be continue for 1 sec (equivalent to 240 samples).

Data was recorded by 64-channel EEG with 10-20 system, but it is necessary to choose appropriate channels because of the number of channels and the calculation’s volume are high. In feature extraction step, we use only seven channels through 64 channels consist of Cz, Pz, C4, C3, Fz, Po8 and Po7 [7]. The 240 time samples are used as features that we down sample them to 24 samples (feature dimensions are 7 (channels) ×24 (samples)). The extracted features scaled to zero mean and unit variance to feed to the classifier. Scaling constants computed for each channel from all trials in the training set and then applied to the testing data. It is important since scaling each trial individually could destroy important amplitude information characterizing the P300.

Machine Learning

After preprocessing the training data sets of both subjects (A and B) and extracting distinct P300 wave features, the classifier is trained. The svmtrain function is used for training the classifier and the svmpredict function is used for predicting the character in MATLAB. We used Support Vector Machine (SVM) as classifier. Support vector machine is a practical learning method based on statistical learning theory and have been successfully applied for classification and regression in various domains of pattern recognition. SVM which has been introduced by [8] aims to separate data into two classes using the best hyperplane to solve classification problems. In this research, we used SVM for classifying EEG signals to detect presence or absence of the P300 component in EEG ERP that is very important for the P300 speller paradigm in field of BCI. For the support vector machine, the Libsvm open source is used. Main features of Libsvm include different SVM formulations, efficient multi-class classification, Cross validation for model selection, probability estimates, various kernels and Weighted SVM for unbalanced data. A kernel function with a second order radial basis (Gaussian) and C = 0 are used. Different values of hyperparameter C has been tried as mentioned in [9]. The study proposed in the competition winner, introduced the use of the ensemble classifiers method and showed the best results for the BCI competition.

Process of Characters Prediction in the Test Set

Test sets are processed similarly to the training set and then fed to a classifier. For the competition, performances is evaluated based on the correctness of predicted characters in the test sets. We showed target character sequences as announced in the competition and predicted the character in each epoch then the correct characters from labeled data are obtained in percentage whether the resulting character is or is not correctly predicted. We used the labeled training data to predict the character sequences in the test sets.

The Design of the Proposed Visual Stimuli

After reviewing the literature and studying several paradigms used for testing the reliability of the SSVEP and P300 in different applications, we decided to use two types of paradigms that enable the user to interact with visual stimuli and evoke the SSVEP and P300 wave. To design our paradigms, we used MATLAB GUI, Psych Toolbox MATLAB and BCI2000 platform. Before describing the platform design, we would like to introduce general implementation diagram as shown in Figure 3. According to the above block diagram, general process performed in this research consists of several steps that we describe them briefly; EEG recording device that we used is EMOTIV Epoc headset in our research and it records the brain signal after applying stimulation and sent it into a receiver module that is connected to a computer through USB port. BCI2000 software is used to save and transfer data to MATLAB. Fieldtrip Buffer can transfer data to MATLAB by onlinify code in real-time or process them in future and then classification results of system, give feedback to user for next action. For all experiments, we should perform all above steps to collect considered data.

The SSVEP Speller Paradigm

For designing SSVEP speller, we create a periodic stimuli mechanism with six low-range frequencies by a MATLAB Psych Toolbox that is used to evoke the SSVEP. The Psych toolbox is a software package that adds this capability to the MATLAB application on windows and Linux computers. MATLAB and Psych toolbox environment is flexible and relatively easy to learn. We used Psych toolbox version 3 (PTB-3) that is a free set of MATLAB functions for vision and neuroscience research. It makes it easy to synthesize and show accurately controlled visual and auditory stimuli and interact with the observer [10]. As described in chapter 2, the SSVEP is a periodic response evoked by a visual repetitive stimulus and it can be detected from the visual cortex. We use only six frequencies in the matrix. First, we designed SSVEP visual stimuli in MATLAB GUI with frequencies of 8.18, 8.97, 9.98, 11.23, 12.85, and 14.99 due to one of articles that we studied [11]. As shown in Figure 4, yellow lines are used to demonstrate the lack of repetitive frequencies appearing in the same column or row and all of squares flicker at different frequencies in each row or column. This platform had three problems:

a) The first problem in this platform can be referred to choosing the frequencies. Some of these frequencies were not match with refresh rate of MATLAB GUI and monitor.

b) The second problem was which frequencies were interrupted a short time after activity and this problem caused us to be faced with many errors. Because of this, we decided to switch to another platform design and software.

c) We did not pay attention to monitoring refresh rate. The visual stimulator plays an important role in a SSVEP based BCI that can be presented using flicker on a computer screen. In a 60Hz refresh rate monitor, there are 60 frames per second and the frame number per cycle is a constant. For example, we can use stimulus frequencies at 7.5 Hz (eight frames per period), 8.57 Hz (seven frame per period) and the number of frames is calculated by dividing monitor refresh rate (60 HZ) to flickering frequency (60/7.5=8 frames). Therefore, when we use a framebased design to ensure a flicker’s frequency stability, the number of stimuli is limited with the refresh rate of a monitor. Our second platform that is designed by Psych toolbox solve the previous problems. We choose six frequencies, which set at 7.5, 8.57, 9.23, 10, 12 and 13 Hz and are selected due to the several causes:

a) The SSVEP amplitude in the frequencies of 6-15 Hz is higher than the other frequency bands [68, 69] and desired frequencies are selected corresponding to our monitor refresh rate.

b) The stimuli with harmonic frequencies cannot be used for SSVEP detection [10].

c) The larger differences between frequencies will benefit the detection of the target frequency.

In addition, we design two types of SSVEP paradigms with Psych toolbox that are shown in Figures 5 & 6 and In Figure 5, we put together all squares without space and in Figure 6, we put together all squares with space and after recording data from both paradigms, we compare their performance. In Figure 7 frequencies layout due to our SSVEP design is shown with different colors and each color is related to a frequency. The P300-Speller Paradigm. The P300 speller paradigm use visual stimulation enabling to write by spelling. Our paradigm is designed by MATLAB Psych toolbox. To create visual stimuli for P300 speller, we use row-column (RC) paradigm as described in chapter 2. This panel is similar to RC paradigm that is introduced by Farewell et al. [6]. The method consists of displaying a 6×6 matrix composed by the numbers and letters as shown in Figure 7. Target letter is demonstrated to user with blue color in training run. Rows and columns of the matrix are successively highlighted. When the row or the column contains the chosen letter, a P300 ERP appears in EEG signal. In P300 paradigm, Synchronization is an important parameter to design because the P300 potentials are very hard to detect in EEG background. In this paradigm, we tried to show that time of onset intensification match with our EEG system results (signals recorded by our EMOTIV device) when we synchronize events accurately. To be more precise signals recorded by our EMOTIV device shows it’s a target or nontarget when the stimulus is on the screen. Since that our data are read from field trip buffer and there is a delay time, our platform was not reliable to BCI systems in order to type letters. Therefore, for solving this problem, we decide to use BCI2000 platform that do not have these problems. (Figures 8 & 9) shows BCI2000 row/ column P300 speller platform that is derived from Donchin’s matrix speller paradigm [12,13].

Figure 5: Proposed SSVEP Visual Stimuli Paradigm without Spaces.

Figure 6: Proposed SSVEP visual stimuli paradigm with space between stimuli.

Figure 7: Frequencies Layout.

Figure 8: Proposed P300 Speller Paradigm.

Figure 9: BCI2000 P300 Speller Platform.

The P300-SSVEP Speller Hybrid paradigm

In this research, we design two-stage hybrid BCI with using two evoked potential detection patterns (SSVEP and P300) sequentially. The first protocol of acquisition is defined by SSVEP method and second by P300 as presented in Figure 10. Our main goal was that both paradigms are designed using same software (Psych toolbox), while in the last design our first platform is designed using Psych toolbox MATLAB and our second platform is designed using BCI2000 software. To design SSVEP, we used Psych toolbox that is described in section 4.3.1 and for P300 we used BCI2000 platform that is introduced in section 4.3.2. The general concept of our hybrid BCI system is demonstrated in Figures 11 & 12. SSVEP task is designed to frequency detection through six selections because we have only six frequencies that each frequency belongs to six characters (one row or one column). One of the six selected characters is target character that is determined by P300 task. P300 task is implemented by BCI2000 platform.

Figure 10: Hybrid BCI System Sequence.

Figure 11: Experiment with SSVEP platform with space.

Figure 12: Eexperiment with Proposed P300 platform.

Implementation of Offline SSVEP Speller Using EMOTIV EPOC

Our data are recorded from 14 channels by the means of EMOTIV EPOC Neuroheadset and are transferred to MATLAB via BCI2000 so that we used a MATLAB toolbox to make online analysis of EEG data transferred by BCI2000. It used Fieldtrip Buffer to transfer data to MATLAB in real-time. Obviously, fieldtrip buffer and BCI2000 must be available and they must be added to the MATLAB path. This BCI system is non-invasive and synchronous. Five healthy subjects (one female and four males) with age between 23 and 38 years participated in initial experiment as volunteer. The purpose of the required task was explained to subjects in detail before preparation for the EEG recording. For SSVEP detection, we design two platforms as we descried in section 4.3.1 and then we compare them.

The initial experiments is performed in a normal and quiet laboratory room, after preparation for the EEG recording, the subjects seated in a comfortable chair in front of the LED FLATRON monitor. First, we performed experiments in SSVEP Condition with simple paradigm, which is available at this site http://omidsani. com/. Then we started our tests with our both paradigms (with space and without space) show one of the SSVEP experiments in the laboratory. Only six frequencies used in two paradigms that set at 7.5, 8.57, 9.23, 10, 12 and 13 Hz. All five subjects participated in all experiments in six sessions so that each session includes three runs. Totally, all subjects should take part in eighteen runs and subjects should focus on one special character for along 15 and 40 seconds to detect frequency in each run. All of conditions are same for two experiments. Between each run, we considered 2-second rest time to avoid subject fatigue. Finally, we recorded EEG signals for all five subjects in offline runs and after processing them, we calculated classification accuracy, averaged accuracy and standard derivation.

Implementation of Offline P300 Speller-based BCI System Using EMOTIV EPOC

First, we performed experiments in P300 condition with Psych toolbox paradigm and then we started our tests with BCI2000 paradigm. One healthy male subject (38 years) are participated in these experiments. Each session included a number of runs. In each run, the subject focused attention on a series of characters. The sets of 12 flashing repeated 15 times for each character. There were 12×15 =180 intensifications totally for each character. Each row/column was randomly intensified for 100ms and after intensification of a row/column, the matrix was blank for 75ms (Figure 4) shows experimental environment with platform. After finishing initial experiment, we started tests with second paradigm. The second implemented system is an offline P300 speller that used platform of BCI2000 software. In BCI2000 platform, the P300 Speller implements a BCI that uses evoked responses to select items from a rectangular matrix, a paradigm originally described by Farwell et al. [6]. In this experiment we asked user to focus on the number of the characters one by one in calibration run and then main runs. We asked the subject to consider below words one by one in P300 speller platform.

The Quick Brown Fox

To calibrate the P300 speller for an individual subject, we first operated in a ‘copy spelling’ mode that prompts the user to pay attention to desired letters in sequence. As the user counts the number of times the desired letter intensifies in the word, a P300 response is created. The purpose of the calibration session is to de detect those features that discriminate between the desired and undesired rows and columns. After the initial runs, we used the BCI2000 offline analysis tool to determine which features correspond to row/column of the desired letter and we saw a feature map that is shown in Figure 13. In this plot, the vertical axis corresponds to the different locations, while the horizontal axis corresponds to the time after stimulus. The signals are shown at a particular location and time after stimulus in 14 channels and the color shows the value of r2 calculated between correct and incorrect row/columns. Red colors demonstrate a high correlation of the brain signal amplitude at the time/location with flashing of the desired row/column, and blue colors demonstrate low correlation. We are interested in finding large r2 values between 200 and 550ms because they had most distinction between target and non-target.

Figure 13: Topographic Feature map of P300 produced by the Offline Analysis.

As we expected, the large r2 values between 250 and 500ms are shown and electrodes of O1 and O2 located in occipital lobe have better features than the others for our subject. We choose four best data points have r2 values 0.03268, 0.02596, 0.02506 and 0.02808 that occur at times of 386.7ms, 394.5ms, 402.3ms and 378.9ms respectively and all four data points are detected by channel eight. We could product time-course plot for target/non-target row/ columns. The target reaction was stronger than for the non-target. Then we used BCI2000 P300 classifier tool that determines optimal features such as channels and signal times corresponding weights. After calibration run, we started main runs. We asked a 38-years male to focus on each character one by one in P300 Speller. We considered two string characters as shown below:

First String of Characters: BROWNTHE_MANCATDOGFAKHRSHAM

Second String of Characters: GBADMANGFTMKHODATF

The first string of characters is used as train data and the second string of characters are used as test data. Experiment conditions was similar to initial test. There were 180 intensifications totally for each character. Each row/column was randomly intensified for 100ms and after intensification of a row/column, the matrix was blank for 75ms and stimulus duration last 175ms for each intensification. To focus and avoid distraction of user, we asked him to count numbers of time that desired character intensified. The data recorded with 128Hz sampling rate. We considered 2-second rest time after each four or five characters to prevent subject fatigue.

Implementation of Offline P300-SSVEP Speller Using EMOTIV EPOC

In hybrid condition, we also used the EMOTIV Epoc to capture EEG signals. One healthy subject (38 years male) is participated in offline experiment that includes two steps. In first step, the subject focused on desired character (for example A) to spell and if the frequency is detected, the characters corresponding to detected frequency are shown in another color in platform (Figure 14). In next step, the subject focused on same character (for example A) in P300 paradigm that finally target character is detected. SSVEP task last 15 second for each character and P300 task take long 31 second in each run.

Figure 14: An Example of Frequency Detection.

EEG Signal Processing and Machine Learning Algorithms

a) Data Acquisition: In the data acquisition part of the BCI application, the selected input is acquired by the recording electrodes, amplified and digitized. Our data are recorded from 14 channels by the means of EMOTIV EPOC Neuroheadset that are transferred to MATLAB via BCI2000 or without using BCI2000 are read from fieldtrip buffer directly.

b) Pre-processing Methods: The pre-processing stage provides the signals in a suitable form for further processing, so careful selection of preprocessing steps is crucial to the success of any classification scheme. Appropriate application of preprocessing steps can decrease data dimensionality and emphasize portions of the data with discriminative power, thereby reducing computation time and improving classification rates. For SSVEP, we used Fast Fourier Transform (FFT) and Power Spectrum Density (PSD) in pre-processing stage. For P300, we used averaging in preprocessing stage. PSD finds the power density for filtered EEG recordings. Mathematically, PSD is defined as a Fourier Transform of the autocorrelation sequence of the time series. One of the non-parametric power spectral density estimation methods is Welch method that it divides the EEG signal into overlapped windowed segments and then average the periodograms for all segments. Welch method uses windowed segments to reduce the effect of the discontinuity of the segments.

Signal processing and Classification:

a) Canonical Correlation Analysis (CCA): We used Canonical Correlation Analysis (CCA) method for SSVEP frequency detection in this thesis. The CCA is used to process the SSVEP data and the Fourier Transform (FT) is used to evaluate the performance of the CCA. CCA is generally used for finding the correlations between two sets of multi-dimensional variables. Before using CCA method for SSVEP detection, we asked users to perform the training stage to each frequency for 60 sec. Then, the resulting signal is analyzed to two-second blocks that for each of these two-second window blocks, the CCA coefficients corresponding to each frequency are calculated. Therefore, after acquiring training data for each user and each two-second window size of its signal, create a six-labeled data (label 1 denoting observed frequency by user). Then, we used Logistic Regression (LR) method to detect SSVEP event and in result, generated vector of coefficients that were inner multiplied in the feature vector. Finally, for frequency detection we used CCA method and obtained results for each subject.

b) Support Vector Machine (SVM) & Stepwise Linear Discriminant Analysis (SWLDA): SVM analysis also has three main steps: Data processing to ease later analysis, SVM classifier training on the labeled training set, test set classification using the trained SVM classifier. SVM classifier and SWLDA classifier are used for P300 speller classification separately and then we compared these two methods.

Evaluation

a) Classification Accuracy (CA): We calculated classification accuracy to evaluate our experiments and compared results of our tests that was calculated with following formula:

Classification accuracy = (the number of sample cases correctly classified/ total number of sample cases) ×100.

Discussions and Results

Results of BCI Competition Data Sets

In 2004, a BCI competition was held and the sets of this competition are available for free on the website of the competition. In this competition, one of the datasets is obtained from a complete record of P300 evoked potentials recorded with 64 channel EEG according to 10-20 system by BCI2000 software. Some of the typed characters were said to the participants and they should have achieved another typed characters by signal processing. In this thesis, we also processed the same datasets and obtained the relatively good accuracy. We used 8-order band-pass filter (Chebyshev Type I) for pre-processing and a SVM classifier with Gaussian kernel was used for classification (Figure 15) presents the mean signals averaged over all training signals for the channel Cz for both target and non-target. As expected, the blue curve demonstrates the target (with P300) and the pink curve shows non-target (non-P300) in time course of average signal waveforms at Cz channel. This Event Related Potential occurs around 250- 500ms after the stimuli that is acceptable. Illustrates results of classification character with SVM classifier. SVM is trained with 85 known characters in training phase then we can detect 100 unknown characters with relatively good accuracy (87%). These classification results are achieved for subject A.

Figure 15: Averaged signals over Cz channel among all training data.

Results of Proposed Offline SSVEP Speller Classification

As we described in chapter four, the experiments were performed for five healthy subjects in eighteen runs. For SSVEP detection, we estimated power spectral density for each frequency. Obtained results from welch power spectrum estimation for 10Hz frequency are shown in Figures 2 & 3 at only over O1 and O2 electrodes. The first harmonic frequency of 10Hz is shown in 20Hz. We used CCA method and then evaluated our results with classification accuracy as shown in Tables 1 & 2 shows the results of CCA method of subject A for one run (15 sec). The numbers that are shown in red color are maximum value of each group as frequency detection through six frequencies. These six columns are corresponding to PSD peak in six frequencies. The histogram of frequency detection using CCA for 10 Hz is shown in Figure 5, Table 3: Offline Performance Comparisons Between Two SSVEP Paradigms (40 sec). In Table 4, we compare performance of two SSVEP paradigms that we proposed in this thesis for 15 seconds. SSVEP paradigm 1 performed better than SSVEP paradigm 2 with averaged accuracy 86.64 and 75.5 respectively. In Table 4, we compare performance of two SSVEP paradigms for along 40 second. SSVEP paradigm 1 performed better than SSVEP paradigm 2 with averaged accuracy 81.06 and 74.4 respectively.

Table 1: Results of Classification with SVM.

Table 2: Frequency Detection Using CCA algorithm for Subject A.

Table 3: Offline Performance Comparisons Between Two SSVEP Paradigms (15 sec).

Table 4: Offline Performance Comparisons Between Two SSVEP Paradigms (40 sec).

Results of Proposed Offline P300 Speller Classification

As we described in chapter 4, we designed a P300 Speller paradigm with MATLAB Psych toolbox and collected data from this platform. Before classification, we got average from all trials for each subject that specifies target and no-target plot (Figure 5) shows target and non-target plots for character O at O1 electrode with our platform that is not acceptable to detect P300 because onset of our stimulation was not synchronized with onset of recording through the fieldtrip buffer (Figures 16-19). In order to obtain results from our platform, we had to use BCI2000 platform and acquired new data from it. All data collected from suitable conditions as we presented before. (Figure 20) shows example of averaged signal over O1 channel for character H. sampling rate is 128 Hz. Classification accuracy calculated from BCI2000 platform by SVM classifier for one subject is shown in Table 5. 5-fold cross validation is done. The data set is divided into 5 subsets, and the k-fold method is repeated 5 times. Each time, one of the 5 subsets is used as the test set and the other 4 subsets are put together to form a training set. Averaged accuracy is performed on different trials. The results of SWLDA classifier of BCI2000 are shown in Tables 5 & 6. Waveform plot is produced by the offline analysis for channel 8 is shown in Figure 21. In left figure, Waveform of target or non-target for channel 8 is shown and in right figure, waveform of r2 due to the topographic feature map in chapter 4 is shown. As shown in Figure 22, we showed graph of accuracy in terms of number of averaged trials according to (Tables 5 & 6). We achieved 100% accuracy from the 12th trial with SWLDA classifier.

Table 5: Classification accuracy with SVM.

Table 6: some results of test data with SWLDA classifier.

Figure 16: Averaged signals over Cz channel among all training data.

Figure 17: 10 Hz Frequency Detection at O2 electrode.

Figure 18: Frequency Detection with CCA in 10 Hz.

Figure 19: Averaged signal over O1 channel for character O.

Figure 20: Averaged signal over O1 channel for character H.

Figure 21: Waveform plot produced by the Offline Analysis.

Figure 22: Graph of Accuracy in terms of number of averaged trials.

Results of Proposed Offline Hybrid Speller Classification

For our hybrid BCI system in offline experiments, we achieved classification accuracy for both of tasks consist of SSVEP condition and P300 condition that have been shown in Table 7. Three subjects are participated in these experiments. Each subject focused on one character on first platform (SSVEP) for 15 seconds to frequency detection then the subject focused on another platform (P300) for character detection for about 31 seconds. In first step, frequency of the desired character is detected and then all characters with the same frequency are changed to another color in SSVEP platform. In second step, the desired character is identified through six frequencies in P300 platform. This time includes 15 trials. The results obtained from five healthy subjects show that an offline classification accuracies of 89.98% and 88.6% were achieved using the proposed hybrid BCI speller.

Table 7: Accuracies of Our Hybrid BCI Offline Experiments.

Conclusion

In this project, we have investigated a hybrid BCI system sequentially to purpose to type letters for disabled people. The whole system includes three main parts: design visual stimuli paradigm for SSVEP detection, P300 detection and hybrid detection. The experiments are performed using EMOTIV Epoc device. We succeed to create an acceptable platform for each experiment and get suitable data that are used in signal processing step. Since CCA had been used successfully in many BCI applications, we used CCA to classify the SSVEP data and frequency detection. SWLDA and SVM are used for P300 detection and we conclude that both methods are suitable for P300 classification. We designed a hybrid SSVEP-P300 BCI platform with the random flashings and periodic flickers to evoke the P300 and SSVEP simultaneously that we did not result.

Then, we decided to design a sequential hybrid BCI platform by same software but we could not detect P300 response from its collected signals. Finally, we designed a hybrid SSVEP-P300 BCI platform by two different software. Our hybrid BCI system consists of two sections including SSVEP condition and P300 condition. In first section, as respect we had only six frequencies in a 6×6 speller matrix, we should detect one-group characters with the same frequencies through six groups. In second section, we must detect desired character that there is in selected group in previous section. This hybrid system can help to SSVEP condition to choose more commands using P300 condition so that it can help to increase the number of commands from 6 to 36 commands. In our system, we can achieve only six choices by SSVEP condition.

More BJSTR Articles: https://biomedres01.blogspot.com/