I build AI that can tell, before treatment, how a head and neck cancer will behave.

Biography

Hasan Shaikh is a data scientist and medical imaging researcher in the Quantitative Imaging Research and Artificial Intelligence Lab (QIRAIL), Department of Radiation Oncology, Christian Medical College (CMC) Vellore, one of India's leading cancer centres. He builds AI pipelines that help oncologists make smarter, faster and more personalised treatment decisions for patients with head and neck cancer.

Prediction. A large part of his work asks a simple question: before we treat, can we tell who is likely to relapse? He engineered a 10-feature radiomics and clinical signature (2 clinical, 8 radiomic) for locoregional recurrence, choosing the features with five metaheuristic algorithms searched over more than 1,000 candidates. It reached an AUC of 0.82 in 327 patients treated with primary radiation, and SHAP explains each prediction so a clinician can see why. The work was a poster highlight at ESTRO 2026 in Stockholm, and a companion paper was accepted at AIHC 2025 (Springer LNEE).

Segmentation. With collaborators at NIT Karnataka, Surathkal, he curated 167 head and neck CT cases from MAASTRO and CMC Vellore and trained a 3D nnU-Net to outline gross tumour volume from CT alone. He is a co-author of AGSR-Net, a segmentation network that adds anatomy-guided skip connections, and of a medRxiv preprint on the nnU-Net results. An oral presentation of this work won third prize at the 2nd National Symposium on Health Data and AI.

Data engineering. Good models need clean data. He built the DICOM retrieval pipeline on Orthanc PACS for a prospective cohort of more than 2,000 head and neck cancer patients in a DBT/Wellcome Trust India Alliance study, automated segmentation and PyRadiomics extraction across it, and set up S3 backups for a five-year multisite radiotherapy study. He also contributed 304+ de-identified cases to CHAVI, India's first national oncology imaging biobank, working with Tata Memorial Centre and IIT Kharagpur.

Reproducibility. He also checks whether published models hold up. He reproduced a CNN-based prognosis study from the authors' own code, data and Docker container. The paper reported 95% CLAIM compliance, yet AUCs differed by up to 0.22, and external validation on a separate cohort gave an AUC close to chance. The lesson, now under submission to Scientific Reports, is that a high checklist score does not guarantee a trustworthy model.

Background. Hasan holds an M.Tech in Computer Science and Engineering from Aligarh Muslim University, where he worked with Prof. Rashid Ali on multimodal deep learning for cancer survival. His model, MGAttCNNMD, fuses clinical, copy-number and gene-expression data and reached 91.2% accuracy on METABRIC; it became a chapter in Data Science in the Medical Field (Elsevier). Before joining CMC Vellore he spent six months as a research analyst at STARlab Capital in Lucknow, backtesting options strategies and contributing to an open-source backtesting library.

Community. He organises the lab's hands-on Workshops on Radiomics and Auto Segmentation for clinicians, medical physicists and researchers (three so far), and helped run the Head and Neck Cancer Survivorship Program and the 2nd Annual Winter Symposium on Health Data and AI at CMC Vellore.

Outlook. He believes AI will eventually automate radiation oncology end to end: treatment delivery, planning, auto-segmentation, adaptive radiotherapy and a paperless, AI-driven medical record. He also believes the biggest impact of medical AI will come in under-resourced settings, where expert clinicians are scarce.

He enjoys talking with researchers, clinicians, engineers and anyone curious about AI in healthcare, so please get in touch.