Projects

A selection of research and engineering projects spanning biomedical imaging, machine learning, computer vision, and computational methods.

Doctoral Research

My doctoral research focuses on developing optical and computational imaging methods for visualizing, quantifying, and interpreting dynamic biological processes in living tissue

Kinetics of Transient Tissue Transparency (T3RIE)

After establishing transient optical clearing in living tissue, this work focused on a key next question: how does transparency change over time in vivo? We developed complementary optical approaches to quantitatively monitor these dynamics in living animals.

Used an implanted self-illuminating light source to directly quantify changes in transmitted intensity and light-source profile width during T3RIE.
Combined visible and NIR imaging to develop a non-invasive reflectance-based method for monitoring transparency and anatomical image contrast over time.
Compared in-vivo and ex-vivo kinetics to distinguish physiological transparency dynamics from changes intrinsic to the clearing agent alone.

Biomedical ImagingT3RIENIR ImagingQuantitative ImagingImage ProcessingOptical Imaging

View publication →

Transient Optical Clearing for In Vivo Imaging

Biological tissue strongly scatters light, limiting how deeply internal structures can be observed using conventional optical imaging. In this work, we used absorbing molecules such as tartrazine to temporarily reduce optical scattering, enabling improved visualization of structures beneath living tissue.

Developed image-enhancement and segmentation workflows for cleared-tissue images, including contrast enhancement, denoising, and organ segmentation.
Analyzed ex-vivo and in-vivo imaging data to improve visualization of internal biological structures after optical clearing.

Computational ImagingImage ProcessingSegmentationIn Vivo ImagingPythonOpenCV

View publication →

Other Technical Projects

Additional projects exploring biomedical signal processing, medical imaging, computer vision, machine learning, algorithms, and engineering applications.

Real-Time ECG Classification & Mobile App Deployment

An end-to-end ECG monitoring system that streams two-channel ECG data over Bluetooth Low Energy to a Flutter mobile application, performs real-time signal processing and beat analysis, and runs a neural-network classifier directly on the device.

Streamed MLII and V1 ECG data to the phone over BLE at an effective sampling rate of 250 Hz using batched packet transmission.
Ported a db4 discrete wavelet transform filtering pipeline from MATLAB to Dart for real-time ECG denoising and baseline-wander correction.
Implemented R-peak detection, beat segmentation, RR-based features, and on-device Normal/Abnormal beat classification.
Achieved 85.81% classifier F1 score and 89.16% balanced accuracy on held-out test records, with an end-to-end F1 score of 84.98%.

ECGFlutterDartBluetooth LESignal ProcessingWaveletsNeural Networks

Breast Ultrasound Tumor Segmentation

A classical image-processing pipeline developed in MATLAB to identify and segment benign breast lesions in ultrasound images from the BUSI dataset.

Reduced ultrasound speckle noise using median filtering and separated darker lesion candidates using adaptive thresholding.
Combined morphological cleanup with feature-based candidate selection using shape, intensity, area, and local contrast.
Achieved a mean Dice score of 0.9051 and mean IoU of 0.8292 across the 10 benign development images.

MATLABMedical ImagingImage ProcessingSegmentationMorphologyUltrasound

Automated Inspection of IC Markings

End-to-end computer-vision pipeline for classifying integrated-circuit markings as good or defective, including missing, unclear, and distorted characters.

Combined grayscale preprocessing, Gaussian filtering, Canny edges, and hybrid Otsu/Sauvola binarization.
Used contours, projection profiles, Fourier analysis, and Hough methods for layout and orientation correction.
Explored MSER, SWT, watershed, and CRAFT for text extraction, with custom YOLOv8 recognition and rule-based final classification.

OpenCVscikit-imageYOLOv8CRAFTComputer VisionManufacturing QA

Shortest-Path Timing in Circuit Graphs (ISCAS’85)

C++ tool that models ISCAS’85 logic circuits as weighted directed graphs and computes minimum-delay paths between selected inputs and outputs.

Built a bench-file parser that preserves canonical signal names and validates user input.
Constructed graph connectivity using gate relationships and fan-out-based edge weights.
Implemented Dijkstra's algorithm with path reconstruction to identify timing-efficient routes through a circuit.

C++STLGraph AlgorithmsDijkstraVLSI TimingUnix

Deep Learning for Malaria Detection from Blood Smears

CNN-based classifier for distinguishing infected and uninfected red-blood-cell images using the NIH malaria dataset of more than 27,000 cell images.

Used Keras data augmentation with rotation, rescaling, zoom, and image flips.
Built and trained a custom CNN with dropout and early stopping.
Evaluated performance using confusion matrices and classification metrics, then exported the trained model for reuse.

PythonTensorFlow/Kerasscikit-learnDeep LearningMedical Imaging