PhD Candidate · Johns Hopkins University
Tawsifur
Rahman
Clinical Data Scientist & Computational Pathology Researcher
Building deep learning systems that translate gigapixel tissue images and real-world clinical data into patient outcomes — from stroke prognosis at Johnson & Johnson to cancer detection at Johns Hopkins.
01 — About
Clinical AI That Ships
I am a PhD candidate in Biomedical Engineering at Johns Hopkins University, advised by Prof. Rama Chellappa and Dr. Alex Baras. My dissertation focuses on fast, scalable, and weakly supervised computational pathology for whole-slide image analysis — methods designed to close the gap between research accuracy and real-world clinical deployability.
Alongside my PhD, I work as a Clinical Data Science Intern at Johnson & Johnson MedTech (currently in my second stint), building multimodal deep learning pipelines for acute ischemic stroke prognosis and atrial fibrillation ablation outcome prediction that directly inform clinical decision-making.
My work spans MIL frameworks, vision transformers, contrastive self-supervised learning, and survival analysis, consistently applied to large-scale clinical datasets including TCGA, PANDA, and real-world EHR cohorts. I believe great clinical AI must be both statistically rigorous and operationally deployable — I build both.
02 — Experience
Where I’ve Worked
May 2025–May 2026
- Built BiLSTM and self-attention models integrating electrophysiology signals with 3D catheter coordinates to predict AFib ablation success, directly informing procedural planning.
- Designed multimodal deep learning pipelines combining pathology imaging and EHR variables (demographics, vitals, lab values) to predict 90-day functional outcomes in stroke cohorts.
- Partnered with clinical teams, biostatisticians, and medical device engineers to deliver production-ready decision-support tools under regulatory constraints.
- Developed MIL, GNN, and Vision Transformer pipelines for slide-level cancer classification from gigapixel WSIs (TCGA, PANDA, CAMELYON) achieving state-of-the-art accuracy with minimal labels.
- Designed DTC-WSI: similarity-guided token merging and importance-guided pruning for scalable, cost-efficient WSI inference on large patient cohorts.
- Fused pathology imaging with clinical reports via cross-attention and contrastive learning; applied SHAP and attention-clustering for biomarker interpretability.
- Led end-to-end pipelines from tissue segmentation and patch extraction through MIL/GNN aggregation and clinical evaluation (ROC/AUC, C-index, Kaplan–Meier).
Jan–Dec 2025
- Built EarSpot™: pediatric auricular deformity screening via YOLO-based ear localization and transformer classification from mobile phone images, replacing costly in-clinic measurements.
- Developed 3D cranial shape modelling from 2D images for plagiocephaly and brachycephaly severity scoring.
- Co-inventor on provisional US patent for AI-based auricular deformity screening.
- Built deep learning diagnostic systems for COVID-19, tuberculosis, and diabetic foot detection, contributing to publications with 5,000+ combined citations.
- Designed QCovSML (stacking ML for COVID-19 detection from CBC biomarkers) and BIO-CXRNet (multimodal imaging + clinical data mortality prediction).
03 — Publications
Selected Research 9,000+ citations
04 — Recognition
Awards & Patents
Patents & Innovation
05 — Contact
Let’s Correspond
Open to research collaborations, industry opportunities, and conversations about clinical AI, computational pathology, and deep learning for medicine.