Science Fair Projects
GaitSense: Gait Analysis Using a Novel Smart Embedded Insole with Machine Learning for Enhanced Diagnosis and Monitoring of Parkinson’s Disease
Parkinson’s Disease (PD) is a leading neurodegenerative disorder characterized by motor decline due to dopamine depletion in the brain’s nigrostriatal pathway. Current diagnosis methods rely on subjective gait analysis, which lacks quantitative biomarkers and fails to capture motor fluctuations in daily life. This research developed GaitSense, a clinical decision-support platform using a custom built non-invasive pressure sensing insole to extract objective biomarkers from gait (walk pattern) for automated PD diagnosis and monitoring. An operational prototype was engineered using eight anatomically-aligned pressure sensors to record plantar pressure dynamics. The low-profile and flexible insole along with no-contact electronics ensure natural walk. Raw pressure data is wirelessly transmitted to a software pipeline that extracts high-fidelity biomarkers, specifically Center of Pressure (CoP) trajectories and Double Support Phase duration. These biomarkers are used to diagnose the presence and severity of PD through a multi-stage Machine Learning inference engine trained on an available clinical dataset. Clinicians can track a patient’s biomarkers over time and identify drug treatment efficacy using a streamlined website. The pipeline achieved 90% accuracy and 98% sensitivity in distinguishing PD patients from healthy controls. The system demonstrated a 2% False Negative Rate, ensuring a PD diagnosis is not missed. The insole and biomarker-driven diagnosis pipeline was validated through a clinical trial involving 11 healthy control subjects. The results demonstrated exceptional efficacy, with key biomarkers derived from the participants' walk data showing a close correlation (within 5%) with the clinically proven dataset.
Skills: Computer Science · Deep Learning · Artificial Intelligence (AI) · Python (Programming Language) · Machine Learning · Computer Vision · Arduino C++ · Electrical Engineering · Circuits · PCB Design
CardioSense: Electrocardiogram Analysis Using Machine Learning and Digitization Techniques for Enhanced Clinical Diagnosis
The leading cause of death worldwide is Cardiovascular Disease (CVDs), accounting for 17 million deaths worldwide. CVDs are vital to diagnose early as they can lead to major heart problems. The predominant method to diagnose CVDs is an Electrocardiogram (ECG). ECG electrodes are placed on 12 parts of the human body, represented as 12 waveforms indicating the heart’s activity. Currently, CVDs are primarily diagnosed by human interpretation of an ECG. This has shown insufficient accuracy of diagnosis. The emergence of machine interpretation of an ECG has shown similar accuracy as human diagnoses. Additionally, ECGs are commonly represented as a printed paper strip with 12 leads. This makes it very difficult to access machine interpretation ECG methods, as they would require a digital ECG. To address this gap, this project presents an approach that digitizes paper ECGs and applies artificial intelligence techniques to diagnose CVDs from scanned paper ECG strips. The CVDs aiming to be detected are Atrial Fibrillation, Tachycardia, Bradycardia, and Sinus Rhythm, STEMI (heart attack), Arrhythmia, and Atrial Pacing. Using machine-learning based image processing, a 12-lead ECG is split into digitized individual leads. QRS complexes in an ECG are detected by the Pan-Tompkins algorithm. Labeled QRS complex statistics were used to train various machine learning models. Preliminary results show the highest accuracy from a Random Forest Classifier model and a CNN-LSTM model. In the final system, users can upload scanned images of an ECG strip, which the system processes using machine-learning algorithms to classify the ECG into various CVDs.The leading cause of death worldwide is Cardiovascular Disease (CVDs), accounting for 17 million deaths worldwide. CVDs are vital to diagnose early as they can lead to major heart problems. The predominant method to diagnose CVDs is an Electrocardiogram (ECG). ECG electrodes are placed on 12 parts of the human body, represented as 12 waveforms indicating the heart’s activity. Currently, CVDs are primarily diagnosed by human interpretation of an ECG. This has shown insufficient accuracy of diagnosis. The emergence of machine interpretation of an ECG has shown similar accuracy as human diagnoses. Additionally, ECGs are commonly represented as a printed paper strip with 12 leads. This makes it very difficult to access machine interpretation ECG methods, as they would require a digital ECG. To address this gap, this project presents an approach that digitizes paper ECGs and applies artificial intelligence techniques to diagnose CVDs from scanned paper ECG strips. The CVDs aiming to be detected are Atrial Fibrillation, Tachycardia, Bradycardia, and Sinus Rhythm, STEMI (heart attack), Arrhythmia, and Atrial Pacing. Using machine-learning based image processing, a 12-lead ECG is split into digitized individual leads. QRS complexes in an ECG are detected by the Pan-Tompkins algorithm. Labeled QRS complex statistics were used to train various machine learning models. Preliminary results show the highest accuracy from a Random Forest Classifier model and a CNN-LSTM model. In the final system, users can upload scanned images of an ECG strip, which the system processes using machine-learning algorithms to classify the ECG into various CVDs.
Skills: Computer Science · Deep Learning · TensorFlow · Computer Aided Diagnosis · Artificial Intelligence (AI) · Python (Programming Language) · Machine Learning
A Novel Approach to Improve Autonomous Vehicles Navigation in Inclement Weather Conditions
An autonomous vehicle or self-driving vehicle is a vehicle that can drive without human interaction. Autonomous vehicles rely on camera sensors to navigate. Autonomous vehicles struggle to navigate in inclement weather conditions like heavy snow. Falling snow obstructs autonomous vehicles cameras making it difficult to detect road signs and other objects, hampering navigation. Failing to navigate in these condition can lead to severe collisions. This project creates a system which will remove snow from autonomous vehicles camera live feed to enable detection of road signs, lane markers, and many other objects, in turn improving autonomous vehicles navigation in snow. This utilizes machine learning computer vision techniques to do this task. This system worked and was able to remove 89% of snow.
An autonomous vehicle or self-driving vehicle is a vehicle that can drive without human interaction. Autonomous vehicles rely on camera sensors to navigate. Autonomous vehicles struggle to navigate in inclement weather conditions like heavy snow. Falling snow obstructs autonomous vehicles cameras making it difficult to detect road signs and other objects, hampering navigation. Failing to navigate in these condition can lead to severe collisions. This project creates a system which will remove snow from autonomous vehicles camera live feed to enable detection of road signs, lane markers, and many other objects, in turn improving autonomous vehicles navigation in snow. This utilizes machine learning computer vision techniques to do this task. This system worked and was able to remove 89% of snow.
Skills: Computer Science · Deep Learning · Artificial Intelligence (AI) · Python (Programming Language) · Machine Learning · Computer Vision
