Medical Image Segmentation


Using Statistical Learning together with Machine Learning and AI to develop a framework for contouring medical images

Updated on August 14, 2025 by Surajit Ray

medical imaging Pet CT AI Contouring Uncertainty Quantification

16 min READ

AISTAT combines a pre-trained U-Net deep learning model with a statistical contour-based framework, enabling semi-automated segmentation that is both interpretable and clinically actionable. Unlike standard pixel-based methods, it provides contour-level outputs suitable for radiotherapy (RT) including intensity modulated (IMRT) and includes a traffic light-style uncertainty measure, supporting more informed clinical decisions.

The tool has already been trained on publicly available AUTOPET dataset and currently being tested on 1,400 non-small cell lung cancer and 60 lymphoma patients. In this GKEF project we will focus exclusively on H&N cancers, using both the publicly available HECKTOR dataset and a real-world set of 200 patients from NHSGGC, including radiotherapy plans to provide a demonstration of clinical utility on real-world tumours that will in turn provide a platform for further investment and development.

Goal

Combine Statistical Learing with AI to provide a new framework for probabilistic contouring and Uncertainty Quantification

Researchers

Partners

Deep Probability Contour Framework for Tumour Segmentation and Dose Painting in PET Images
Zhang. W and Ray S. International Conference on Medical Image Computing and Computer-Assisted Intervention.
Journal Page | PDF | Open Access | Cite |
Abstract

From coarse to fine: a deep 3D probability volume contour framework for tumor segmentation and dose painting in PET images
Zhang. W and Ray S. Frontiers In Radiology,. 3
Journal Page | PDF | Open Access | Cite |
Abstract

DenResCov-19: A deep transfer learning network for robust automatic classification of COVID-19, pneumonia, and tuberculosis from X-rays
Mamalakis M., Swift A.J., Vorselaars B., Ray S., Weeks S., Ding W., Clayton R.H., Mackenzie L.S., and Banerjee A. Computerized Medical Imaging and Graphics. 94
Journal Page | Scopus Link | Cite | Citing Papers |
Abstract

Signaling local non-credibility in an automatic segmentation pipeline
Levy J.H., Broadhurst R.E., Ray S., Chaney E.L., and Pizer S.M. Progress in Biomedical Optics and Imaging - Proceedings of SPIE. 6512 (PART 3)
Journal Page | Open Access | Scopus Link | Cite | Citing Papers |
Abstract

Preprints and conference proceedings

Kernel Smoothing-based Probability Contours for Tumour Segmentation
Zhang, Wenhui; Ray, and Surajit 26th UK Conference on Medical Image Understanding and Analysis (MIUA 2022), University of Cambridge, 27-29 July 2022..
Link |
Abstract

Kernel Smoothing-based Probability Contours for Tumour Segmentation
Zhang, Wenhui; Ray, and Surajit Classification and Data Science in Digital Age - 17th Conference of the International Federation of Classification Society (IFCS 2022), Porto, Portugal, 19-23 July 2022.
Link |
Abstract

Analysis of PET Imaging for Tumor Delineation
Ray and Surajit 11th SINAPSE Annual Scientific Meeting, Dundee, UK, 21 Jun 2019.
Link |
Abstract