Rupam Mukherjee

Research Overview

I work at the intersection of Medical Image Computing, AI for Health, and Computational Pathology. My current focus is on lung nodule analysis & malignancy risk estimation from LDCT and oral cytology WSI analysis for early OPMD detection.

Primary themes: Multimodal fusion, weak supervision, calibrated risk modeling.

Data modalities: LDCT, WSI, clinical biomarkers.

Publications & Manuscripts

Multimodal RGB-HSI Feature Fusion with Patient-Aware Incremental Heuristic Meta-Learning for Oral Lesion Classification

Mukherjee, R., Daniel, R., Hazra, S., Dasgupta, S., Mandal, S.

In review; arXiv version available.

arXiv:2511.12268

Multimodal learning Computational pathology Meta-learning
RAA-MIL: A Novel Framework for Classification of Oral Cytology

Mukherjee, R., Daniel, R., Hazra, S., Dasgupta, S., Mandal, S.

In review; arXiv version available.

arXiv:2511.12269

MIL Cytology Weak supervision

Ongoing Projects

M.Tech Thesis — Deep Learning for Lung Nodule Analysis & Malignancy Risk Estimation (IIT Kharagpur)
- Preprocessing of LDCT volumes (.mha, .nii), patient-wise splits, and volumetric augmentation.
- Training 3D CNNs (ResNet, I3D) and experimenting with nnU-Net for segmentation-driven features.
- Integrating imaging and clinical biomarkers for calibrated malignancy prediction.
- Keywords: 3D CNN, volumetric learning, radiomics, risk modeling.
Oral Cytology — Classification & Segmentation of Cervical/Oral Cells
- Hybrid pipeline combining classical morphology + deep learning for robust cell analysis.
- Weakly/fully supervised setups on cytology datasets; emphasis on generalization and Dice/AUC.
- Exploring stain normalization, patch-level transformers, and multi-task learning.

Completed & Prior Work

Brain Tumor Segmentation — Image Processing Lab Capstone
- Classical pipeline: anisotropic diffusion, top-hat filtering, histogram-based segmentation.
- Deep learning extension with U-Net, achieving improved Dice and AUC on public MRI slices.
Pancreatic Cancer Prediction from Imaging & Biomarkers — B.Tech Thesis (KGEC)
- Developed CNNs for CT-based detection; modeled urinary biomarkers with ML (RF/XGB).
- Reported improvement over prior SOTA by ~0.7% with careful preprocessing and feature selection.
Telemedicine Platform — Neurology Hospital Management
- Web system (Apache/PHP) integrating AI-assisted image analysis streams for clinical workflow.
- User roles, data management, and basic visualization for clinician-facing modules.

Methods & Tooling

ML/DL: PyTorch, MONAI, scikit-learn, Optuna (HPO)
Medical Imaging: SimpleITK, NiBabel, PyRadiomics, OpenCV
Data & Visualization: NumPy, Pandas, Matplotlib, Seaborn, SciPy
Ops: Git, Docker, Linux, HPC