What I work on
My work sits at the intersection of medical image processing and machine learning, focused on making perfusion MRI a more objective tool for neuro-oncology.
Quantitative DCE-MRI
Perfusion imaging of brain tumors to derive objective, reproducible biomarkers for grading and treatment assessment.
Vessel Segmentation
Swin UNETR, Attention U-Net and U-Net pipelines for automatic segmentation of large blood vessels in tumor MRI.
Glioma Classification
Differentiating glioblastoma, diffuse midline glioma and grade-3 glioma from quantitative perfusion parameters.
Radiomics & Texture
PyRadiomics workflows — 107 features per sequence, 851 with wavelet decomposition — paired with ML classifiers.
MRI Contrast Agent Optimization
Collaborative work with IIT Mandi — MRI scanning of contrast-agent samples (n = 27) and T1/T2 relaxivity computation for selecting optimal agents for specific MRI applications.
Brain–Computer Interfaces
EEG-based study of advertisement impact on consumer preference and mental stress detection, using SVM classification (M.Tech research).
Removing vascular bias from tumor perfusion maps
Pipeline overview — Pipeline overview — multiparametric MRI → perfusion maps (CBV, Slope-2) → k-means clustering of the CBV×Slope-2 product → generated LBV mask, refined to ground truth.
Large intra- and peri-tumoral blood vessels distort quantitative DCE-MRI perfusion maps, making glioma grading operator-dependent and hard to reproduce.
A Swin UNETR transformer segments large vasculature from brain-tumor DCE-MRI — benchmarked against Attention U-Net and U-Net — then masks it out of the perfusion computation.
More reproducible perfusion estimates and cleaner grade separation — published in Magnetic Resonance Imaging (2025) and presented as a Power Pitch at ISMRM.