Bridging Statistical Learning and Deep Learning for Probabilistic Segmentation in Molecular Imaging
Abstract
The integration of artificial intelligence (AI) in medical imaging is transforming cancer treatment, particularly in radiotherapy (RT), where accurate tumour detection and segmentation are essential for effective treatment planning. Functional imaging techniques like Positron Emission Tomography (PET) provide valuable insights, but current AI-driven segmentation methods often struggle with precisely identifying tumour boundaries. This is especially critical for advanced treatment strategies such as intensity-modulated radiation therapy (IMRT) dose painting, which requires highly accurate contouring to ensure optimal radiation delivery.
In this talk, I will introduce KsPC-Net, a novel AI-powered approach that combines deep learning with statistical techniques to improve segmentation accuracy in molecular imaging. By integrating a convolutional neural network (CNN) with a probability-based contouring method, KsPC-Net not only enhances tumour boundary detection but also provides confidence estimates for treatment planning. Unlike conventional methods, our approach learns key parameters automatically, reducing the need for manual adjustments and improving consistency. Tested on a leading medical imaging dataset, KsPC-Net outperforms existing techniques, bringing AI-driven segmentation closer to clinical application. This talk will explore how blending deep learning with statistical modeling can lead to more reliable and interpretable AI solutions in healthcare.
Short Biography
Prof. Surajit Ray is a Professor of Statistics at the University of Glasgow. His research focuses on uncertainty quantification of AI algorithms for medical image analysis, as well as the theory and geometry of mixture models and functional data analysis. He is particularly interested in challenges arising from high-dimensional and large-scale data, both in terms of vector dimensions and sample sizes. His methodological expertise includes multivariate mixture models, structural equation modeling, high-dimensional clustering, and functional clustering. He collaborates extensively across disciplines, with key projects in medical image segmentation, immunology, and climate-ecosystem dynamics modeling. Beyond academia, he contributes to advisory boards, including the Newton Gateway to Mathematics, fostering interdisciplinary engagement. His work bridges theoretical advancements with real-world applications, making significant contributions to medical imaging and statistical science.
