Medical Image Analysis
Deep learning for clinically meaningful classification and segmentation in CT and radiographic imaging.
MSc Computer Science · University of Victoria
Nidita Roy
Graduate researcher in medical image analysis and continual learning
I investigate how medical-imaging models can learn new diagnostic tasks without losing previously acquired knowledge. My MSc research uses abdominal CT trauma detection to separate catastrophic forgetting from deeper problems in preprocessing, representation quality, labels, and task formulation.
FIG. 01Nidita Roy · MSc researcher
−1000 HU · airlearnable CT window+3000 HU · dense bone
8.50/9.0Cumulative MSc GPA
50+Students supported as a CSC 111 TA
3 × A+Data mining, massive datasets, medical imaging
Research focus
Continual Learning
Evaluating whether models retain old knowledge while acquiring new injury-detection tasks.
Learnable Preprocessing
Replacing fixed CT window presets with differentiable, task-aware Hounsfield Unit adaptation.
Multimodal Learning
Combining imaging with demographic, physiological, and structured clinical information.
Selected publications
Patient survival prediction using actigraphy and clinical data
A multimodal deep-learning study that achieved 92.4% accuracy using wearable and clinical information.
Open article →Offensive Bengali social-media meme detection
A weighted multimodal ensemble combining visual and textual representations for hierarchical classification.
Open on IEEE Xplore →Deep learning for wound tissue segmentation
A comparative evaluation of 82 segmentation and classification models across three labeling formats.
Open preprint →Spinal disease detection via deep multimodal fusion
Lead-author BSc thesis work combining X-ray imagery and demographic features across eight spine conditions. The camera-ready manuscript is available while the formal proceedings record is pending.
Education and preparation
2025–present
MSc in Computer Science, University of Victoria
Preliminary thesis: Continual Learning for Abdominal Trauma Detection in CT: A Task-Incremental Study of Forgetting, Windowing, and Learnability.
Graduate courses
CSC 502 Systems for Massive Datasets (97% A+); CSC 503 Data Mining (93% A+); CSC 581B Deep Learning for Medical Image Analysis (92% A+); CSC 581C Online Learning (81% A−); CSC 595 completed.
Supervision
Supervisor: Dr. Ashery Mbilinyi. Committee member: Dr. Jaya Prakash Champati.
Interested in reliable and adaptive medical AI?
I welcome conversations about PhD opportunities, research collaborations, and projects in continual learning, medical computer vision, and multimodal healthcare AI.
Get in touch