The Ohio State University
Undergraduate AI/ML Researcher
Digital Pathology, under Dr. Anil Parwani
Conducting digital pathology computer vision and clinical explainability research at The Ohio State University Wexner Medical Center under internationally recognized pathology informatics pioneer Dr. Anil Parwani. Developing transparent, low-latency deep learning pipelines for whole-slide imaging (WSI) and formulating automated area calibration models for clinical allergy diagnostics.
Technical Workstreams & Architecture
3 Core AreasExplainable AI (XAI) & Grad-CAM Saliency Engine
Formulated an 8.09ms vectorized Grad-CAM saliency generation pipeline using NumPy linear Jet colormaps, producing real-time diagnostic heatmaps without GUI rendering locks or memory leaks.
- ―Replaced Matplotlib rendering loops with vectorized NumPy linear interpolation, eliminating 120MB+ RAM overhead.
- ―Generated piecewise colormap overlays illustrating model feature attribution directly over histological tissue beds.
- ―Benchmarked latency on 224x224 RGB tensors to guarantee clinical UI responsiveness.
Automated Wheal & Flare Area Calibration
Engineering computational vision algorithms for standardized allergic skin prick test wheal measurements, replacing subjective manual caliper estimations with mathematically calibrated area contours.
- ―Constructed contour detection and perspective correction pipelines for non-planar patient skin photography.
- ―Implemented HSV color-space thresholding to separate acute erythema from elevated central wheals.
- ―Benchmarked segmentation accuracy against verified clinician ground-truth annotations.
Histological Morphology & Tissue Bed Segmentation
Built image processing models to identify granulation, slough, and necrotic margins on histological biopsy tiles, accelerating pathological tissue classification.
- ―Engineered sub-2ms morphological operators to parse cellular boundary density.
- ―Designed Poisson-disc sampling routines for cellular spatial distribution analysis.
Verified Deliverables & Outcomes
Empirical EvidenceVectorized NumPy linear Jet colormap generated without GUI locks.
HSV morphology separating necrotic and granulated margins.
Eliminated Matplotlib backend locks in clinical browser/API loops.
- ―Working on technical tasks in digital pathology using AI/ML methods under Dr. Anil Parwani at OSU.
Related Systems & Case Studies
2 Architectural Deep-DivesDigital Pathology & Allergy Wheal CV Research
Automated computer vision image-processing pipeline quantifying allergy skin prick test (SPT) wheal and flare areas, replacing subjective millimeter calipers with standardized clinical imaging metrics.
Healios V2 — Multimodal Post-Op Recovery Platform
Clinical recovery monitoring ecosystem integrating Explainable AI (Grad-CAM in 8.09ms), tissue morphology HSV segmentation (2.00ms), edge HTML5 Canvas Laplacian blur gatekeeping (<35ms), and an EHR-integrated clinical recovery agent grounded in ERAS protocols.
Technologies & Tooling
Specific tools, libraries, and frameworks leveraged during this tenure:
Engineering Methodologies
Operational practices and systemic disciplines enforced:
“In clinical pathology, an accurate model that cannot explain its reasoning is a liability. Speed and interpretability must go hand in hand: if a saliency map takes 5 seconds to render, a clinician will skip it; at 8ms, it becomes part of their natural workflow.”