Research
As a part of this club, we try to connect students with labs and professors offering research in areas of interest as well as conduct our own interdisciplinary research in fields of embedded systems, AI, and medical diagnostics. Explore this page to learn about opportunities for research within our club.
Our Research
Sleep Apnea Kit, A Research Concept Brief
Develop a comprehensive understanding of how multimodal biosignals, such as blood oxygen saturation (SpO₂), respiratory rate, heart rhythm, and neural activity, correlate with sleep apnea events and their severity. The goal is to enable continuous, non-invasive detection and classification of sleep apnea episodes through wearable technology.
Rationale
Sleep apnea is a common but underdiagnosed sleep disorder characterized by repeated interruptions in breathing during sleep. These episodes cause drops in oxygen levels, cardiovascular strain, and disrupted sleep cycles, contributing to conditions such as hypertension, arrhythmia, and cognitive impairment. Conventional diagnosis relies on an invasive overnight polysomnography (PSG) in clinical settings, an accurate but expensive, uncomfortable, and resource-intensive process.
Advances in wearable biosensors now make it possible to capture high-resolution physiological data continuously in natural sleep environments. However, for this data to be clinically meaningful, it must be properly interpreted, linking sensor-derived signals to physiological markers of apnea severity, frequency, and recovery response. This project focuses on building that interpretive framework.
Research Questions
- Signal Characterization:
- How do changes in SpO₂, heart rate variability (HRV), and respiratory effort correspond to the onset and resolution of sleep apnea events?
- What EEG or neural signal features reflect sleep stage transitions and arousal responses during apnea episodes?
- Biomarker Correlation and Severity Analysis:
- Which biosignals or combinations thereof most accurately predict apnea severity (mild, moderate, severe) as classified by the apnea-hypopnea index (AHI)?
- How can we quantify physiological stress or compensatory responses, such as elevated heart rate or delayed oxygen recovery following apnea events?
- Data Interpretation and Modeling:
- Which computational or machine learning approaches (e.g., time-series modeling, anomaly detection, supervised classification) best identify apnea patterns in multi-sensor data streams?
- How can signal processing techniques such as Fourier or wavelet analysis enhance detection of micro-arousals and irregular breathing patterns?
- Device Validation and Real-World Application:
- How does wearable-derived data compare to clinical-grade PSG results under varying conditions and sleep environments?
- What sensor configurations and data fusion strategies yield reliable apnea detection without compromising comfort or signal quality?
Desired Research Outcomes
Establish a scientifically validated link between wearable biosignals and clinical indicators of sleep apnea severity. This research phase aims to generate analytical models, signal mappings, and feature sets that can drive real-time apnea detection and scoring. Ultimately, these insights will inform the design of a portable, low-cost sleep monitoring system capable of accurately identifying apnea events and quantifying their physiological impact in home settings.
Learn More About the Project
To learn more about the sleep apnea detection project, visit our project page.
Tri-modal Home Kit for Alzheimers and Parkinsons, A Research Concept Brief
Investigate, characterize, and validate behavioral and physiological markers, derived from tremor analysis, gait monitoring, and eye-movement tracking, that correlate with the onset and progression of Alzheimer’s and Parkinson’s diseases. The goal is to establish a scientific foundation for a non-invasive, in-home monitoring system capable of distinguishing between normal age-related changes and true neurodegenerative progression.
Rationale
Alzheimer’s and Parkinson’s diseases are progressive neurodegenerative disorders that develop long before clinical diagnosis, often after irreversible damage has occurred. Current diagnostic methods rely on in-clinic evaluations and specialized imaging, making continuous monitoring difficult and costly.
A tri-modal home kit, integrating sensors to monitor motor function, ocular behavior, and activity patterns, offers the opportunity for daily, unobtrusive tracking of early disease indicators. However, for such data to be meaningful, it must be interpreted within a robust behavioral and neurophysiological framework. Differentiating between normal movement variability and disease-specific changes is critical to transforming raw sensor outputs into actionable clinical insights.
Research Questions
- Motor and Tremor Analysis:
- What quantitative features of hand tremor (e.g., frequency, amplitude, rhythmicity) best distinguish Parkinsonian tremors from normal micro-movements or essential tremor?
- How do motor variability and bradykinesia metrics evolve with disease progression, and how can they be captured through wearable or environmental sensors?
- Gait and Posture Characterization:
- Which gait parameters, such as stride length, step variability, or postural sway, correlate most strongly with cognitive decline and motor deterioration?
- Can temporal changes in gait dynamics serve as reliable longitudinal biomarkers for disease progression or medication response?
- Oculomotor and Cognitive Markers:
- What eye-movement behaviors (e.g., fixation duration, saccade latency, blink rate) are indicative of neurological impairment in Alzheimer’s or Parkinson’s patients?
- How can these ocular indicators be passively tracked to infer cognitive slowing or attention deficits over time?
- Multimodal Integration and Classification:
- How can data from motion, gait, and eye-tracking sensors be combined through machine learning or statistical modeling to enhance sensitivity and specificity?
- What normalization or baseline-adjustment methods can effectively separate disease-linked signals from those caused by aging, fatigue, or environmental factors?
Desired Research Outcomes
Define a set of quantifiable neurophysiological and behavioral biomarkers that reliably indicate progression of Alzheimer’s and Parkinson’s diseases in naturalistic environments. The research phase aims to produce validated algorithms and interpretive frameworks that transform multimodal sensor data into clinically interpretable trends. These findings will guide the development of a next-generation, tri-modal home monitoring kit capable of early detection, continuous assessment, and personalized tracking of neurodegenerative disease progression.
Learn More About the Project
To learn more about the Tri-Modal Home Kit project, visit our project page.
Retinal Scan / Hearable for Alzheimers, A Research Concept Brief
Investigate and characterize retinal and in-ear biosignatures that correlate with early-stage Alzheimer’s disease (AD). The goal is to establish reliable physiological and electrophysiological markers measurable through a non-invasive, wearable device, enabling early screening and continuous monitoring outside clinical settings.
Rationale
Alzheimer’s disease is often diagnosed only after cognitive decline becomes apparent, by which point significant neural degeneration has already occurred. Conventional screening methods, such as PET imaging and cerebrospinal fluid (CSF) assays, are costly, invasive, and inaccessible for large-scale early detection.
Evidence suggests that Alzheimer’s pathology extends beyond the brain, manifesting as measurable vascular and neural changes in peripheral regions such as the retina and inner ear. Retinal imaging can reveal amyloid accumulation, microvascular thinning, and altered blood flow patterns, while in-ear biosensing (EEG or hemodynamic) can detect neural activity disruptions associated with cognitive impairment. Together, these modalities may provide a powerful, non-invasive biomarker set for early Alzheimer’s detection.
Research Questions
- Biomarker Identification and Validation:
- Which retinal features, such as vessel tortuosity, layer thickness, or amyloid deposition, correlate most strongly with known Alzheimer’s stages and cognitive decline?
- What EEG or bioelectrical patterns detectable via the ear (e.g., auditory evoked potentials, alpha/theta power ratios) reflect Alzheimer’s-related neurodegeneration?
- Multimodal Integration:
- How can retinal and in-ear biomarkers be combined to improve sensitivity and specificity in Alzheimer’s screening?
- Can multimodal data fusion models (e.g., deep learning, Bayesian integration) effectively distinguish Alzheimer’s from other neurodegenerative or vascular conditions?
- Data Modeling and Interpretation:
- Which analytical approaches (e.g., signal processing, graph-based modeling, or explainable AI) best map physiological signals to pathological processes?
- What thresholds or composite indices can be derived from the biosensor data to classify disease likelihood or progression?
- Clinical and Ethical Considerations:
- How do we validate these biomarkers across diverse populations and ensure reproducibility in non-clinical environments?
- What ethical and interpretive frameworks should guide the communication of risk from non-invasive Alzheimer’s screening?
Desired Research Outcomes
Build a comprehensive mapping between retinal/EEG-derived biomarkers and Alzheimer’s pathology. This research will form the foundation for the next phase of device development, transforming raw biosensor data into clinically meaningful indicators. The ultimate outcome is a validated, cost-effective, non-invasive screening method that identifies Alzheimer’s risk and progression, enabling earlier intervention, improved patient outcomes, and easier monitoring of disease progression.
Learn More About the Project
To learn more about the Alzheimer’s retinal and in-ear biosensing project, visit our project page.
VOC Analyzer for Lung Cancer, A Research Concept Brief
Explore and define methodologies for detecting volatile organic compounds (VOCs) in human breath that may serve as biomarkers for early-stage lung cancer. The overarching aim is to establish the biochemical and analytical foundation for a future handheld detection device.
Rationale
Lung cancer remains one of the deadliest cancers due to late-stage diagnosis, when treatment options are limited and survival rates drop below 20%. Prior studies suggest that specific VOCs in exhaled breath reflect metabolic alterations linked to early tumorigenesis, changes that can appear months to years before imaging detects a lesion.
While gas chromatography-mass spectrometry (GC-MS) provides the gold-standard precision for VOC analysis, its cost, complexity, and laboratory dependence limit scalability. The long-term vision is to translate GC-MS-level insights into a portable diagnostic platform, but at this stage, the focus is on identifying, validating, and modeling the relevant VOC patterns.
Research Questions
- Biochemical Signature Mapping:
- What specific VOCs (or combinations thereof) show consistent correlation with early lung cancer across diverse patient cohorts?
- Detection Feasibility:
- Which classes of sensors (e.g., MOS, PID, NDIR, electrochemical) are most responsive to these candidate compounds under controlled conditions?
- Sampling and Environmental Control:
- How do factors such as humidity, diet, or medication influence breath VOC profiles? What protocols ensure reproducible sampling?
- Data Characterization:
- What analytical or AI-driven methods (PCA, clustering, supervised learning) best differentiate VOC signatures between healthy and cancer-affected individuals?
Desired Research Outcomes
Establish the scientific foundation: VOC targets, data patterns, and environmental controls needed to guide the next phase of engineering and prototype development. The research stage in this project should generate evidence to answer: Can VOC-based detection be reliably tied to early lung cancer signatures in a real-world, non-invasive format?
Learn More About the Project
To learn more about the VOC analyzer for lung cancer project, visit our project page
Multilingual Healthcare Communication App, A Research Concept Brief
Develop an AI-powered multilingual communication platform capable of seamless, contextually accurate translation between text and speech across multiple languages. The goal is to enable effective and reliable communication in both everyday and high-stakes contexts, particularly within healthcare settings where linguistic precision is critical to patient safety and understanding.
Rationale
Miscommunication between healthcare providers and patients due to language barriers remains a major global challenge, often leading to misunderstandings, reduced quality of care, and even medical errors. While existing translation tools can convert literal meaning between languages, they often fail to capture the cultural nuance, idiomatic expression, and emotional tone essential to accurate, human-like understanding.
This project aims to bridge that gap by combining natural language processing (NLP), speech recognition, and text-to-speech synthesis in a unified AI model trained for cultural and contextual awareness. By embedding sociolinguistic and medical-domain knowledge into the translation process, the system can ensure clarity, empathy, and precision across diverse linguistic contexts.
Research Questions
- Linguistic and Cultural Context Modeling:
- How can translation systems incorporate cultural idioms, regional dialects, and slang to preserve intended meaning rather than literal phrasing?
- What frameworks or datasets can be developed to represent language variation across demographics, regions, and cultural backgrounds?
- Speech-to-Text and Text-to-Speech Integration:
- How can we optimize cross-modal conversion (speech ↔ text) to maintain accuracy and tone across languages with differing phonetic structures?
- What approaches ensure real-time performance and low-latency translation without sacrificing comprehension or emotional fidelity?
- Medical Communication and Terminology:
- How can medical-specific terminology be standardized and correctly interpreted across languages, particularly in emergency or clinical settings?
- What ethical safeguards and data validation strategies are necessary to minimize misinformation and translation errors in medical contexts?
- AI Ethics and Sociolinguistic Responsibility:
- How can translation models be trained to avoid cultural bias, stereotyping, or exclusion of minority dialects and linguistic identities?
- What role can human-in-the-loop systems play in maintaining trust, transparency, and accountability in cross-cultural AI communication?
Desired Research Outcomes
Establish a comprehensive linguistic and cultural understanding framework that enhances AI-driven translation accuracy and empathy across languages. The research will produce language datasets, contextual modeling techniques, and validation protocols specifically tuned for medical and interpersonal communication. Ultimately, the outcome will guide the development of a multilingual app capable of bridging linguistic divides in healthcare and beyond, ensuring that every user, patient, provider, or community member can communicate effectively and safely.
Learn More About the Project
To learn more about the Multilingual Healthcare Communication App project, visit our project page.