Research output

Publications

A curated record of work in trustworthy medical AI, remote sensing, and scientific computing.

Recent work presented at leading international venues

Medical AI

Trustworthy learning for medical images and multimodal systems

Chest X-ray retrieval explanations from Graph-of-Differences: two query radiographs beside their top retrievals, overlaid with coloured anatomy-graph nodes on the lung and heart regions.

Graph-of-Differences: Anatomy-Structured Difference Alignment for Medical Image Re-Identification

Nichula Wasalathilaka, Abhijit Das, Imran Razzak, Dwarikanath Mahapatra

MICCAI 2026 Medical Image Computing and Computer Assisted Intervention, Strasbourg, France Published

Paper arXiv

PROTON architecture: the vision and text pipelines of a frozen medical vision-language model feeding MCM scoring, an online prototype bank and a confidence gate.

PROTON: Prototype-based Test-time Online OOD Detection for Medical Vision-Language Models

Abhijit Das, Nichula Wasalathilaka, Yifan Lu, Adinath Dukre, Dwarikanath Mahapatra, Shadab Khan, Imran Razzak

MICCAI 2026 Medical Image Computing and Computer Assisted Intervention, Strasbourg, France Published

Paper arXiv

Remote Sensing

Robust visual learning for Earth observation

Controlled benchmark pipeline: a LoveDA RGB tile entering three visual state-space encoder families through identical stems and downsampling stages.

A Controlled Benchmark of Visual State-Space Backbones with Domain-Shift and Boundary Analysis for Remote-Sensing Segmentation

Nichula Wasalathilaka, Dineth Perea, Oshadha Samarakoon, Buddhi Wijenayake, Roshan Godaliyadda, Vijitha Herath, Parakrama Ekanayake

IGARSS 2026 IEEE International Geoscience and Remote Sensing Symposium, Washington, DC, USA Accepted

arXiv Code

Stage B of the prompt-controlled diffusion augmentation: a LoveDA semantic map and its class-share prompt feeding a ControlNet U-Net through CLIP and one-hot encoding, while the paired remote-sensing image is encoded by a frozen VAE and noised, the two meeting at a FiLM gate over the fine-tuned Stable Diffusion U-Net that is trained against the added noise.

Mitigating Long-Tail Bias in LoveDA via Prompt-Controlled Diffusion Augmentation

Buddhi Wijenayake, Nichula Wasalathilaka, Roshan Godaliyadda, Vijitha Herath, Parakrama Ekanayake, Vishal M. Patel

IGARSS 2026 IEEE International Geoscience and Remote Sensing Symposium, Washington, DC, USA Accepted

arXiv Code Website Dataset

Mamba-FCS overview: pre-change and post-change satellite images entering the shared encoder and the three-branch semantic and binary change decoder.

Journal

Mamba-FCS: Joint Spatio-Frequency Feature Fusion, Change-Guided Attention, and SeK Inspired Loss for Enhanced Semantic Change Detection in Remote Sensing

Buddhi Wijenayake, Athulya Ratnayake, Praveen Sumanasekara, Roshan Godaliyadda, Parakrama Ekanayake, Vijitha Herath, Nichula Wasalathilaka

IEEE J-STARS · 2026 IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing Published

IEEE Xplore Code Website

Qualitative change-detection results on three scenes, one per row: the pre-event and post-event satellite images, the ground-truth change map, and two predicted change maps in which agreement with the ground truth is white and the pixels that disagree with it are picked out in red and green.

Precision Spatio-Temporal Feature Fusion for Robust Remote Sensing Change Detection

Buddhi Wijenayake, Athulya Ratnayake, Praveen Sumanasekara, Nichula Wasalathilaka, Mathivathanan Piratheepan, Roshan Godaliyadda, Mervyn Ekanayake, Vijitha Herath

ICIIS 2025 19th IEEE International Conference on Industrial and Information Systems, Kandy, Sri Lanka Published

IEEE Xplore Code

Scientific Computing

Machine learning and computational models for physical and biomedical systems

A CT-derived airway reconstruction that breaks into distal fragments, beside a RespGeomLib procedural airway with smooth branching at every junction.

RespGeomLib: A Reproducible Parametric Engine for Generating Analysis-Ready Human Airway Lumen Geometry

Nichula Wasalathilaka, Parakrama Ekanayaka, Roshan Godaliyadda

IEEE Mercon 2026 Moratuwa Engineering Research Conference Published

IEEE Xplore arXiv Code Website

Three-panel workflow: local CFD on tube, bifurcation and trifurcation primitives with velocity and pressure fields; a closure-aware DeepONet with branch and trunk networks and hard closure on outlet fluxes; and zero-shot whole-tree composition compared against CFD reference with absolute speed error.

Local Consistency Does Not Guarantee Global Conservation: Auditing Zero-Shot Composition of Airway Flow Operators

Nichula Wasalathilaka, Navodya Heshan Samarasinghe, Dhanujaya Suraweera, Kevin Dawson, Chinthaka Jacob, Mervyn Parakrama Bandara Ekanayake, Roshan Godaliyadda

arXiv · 2026

arXiv