Walter Hugo Lopez Pinaya - CMIC/WEISS Joint Seminar Series

Invited Speaker: Walter Hugo Lopez Pinaya, Department of Biomedical Engineering - King's College London.

Title:  Unsupervised Anomaly Detection and Segmentation in Medical Images using Deep Generative Models 

Abstract: 

The detection and segmentation of lesions in medical imaging is an important problem whose solution is of potential value across many clinical tasks, including diagnosis, prognosis, and treatment selection. Ordinarily, segmentation is performed by hand, making this process time-consuming and dependent on human expertise. Therefore, the development of accurate automatic segmentation methods is crucial to allow the widespread use of precise measurements in clinical. Over the last few years, in our research, we have proposed to perform unsupervised anomaly detection and segmentation using deep generative models, such as autoregressive transformers and, more recently, diffusion models. These generative models learn the probability density function of normal data and then highlight pathological features as deviations from normality. In this presentation, we will describe the extensive experiments we performed over different data types, going from whole-body PET/CT to brain MRI/CTs samples, and different pathologies, from various cancer types to intracerebral haemorrhage and early stages of schizophrenia. Finally, we discuss the promising use of these models not just to identify lesions, but also to highlight anomalous data that should not be processed by the downstream machine learning models in an AI-powered pipeline, offering a more clinically safe use of AI. 

Bio: 

Walter Hugo Lopez Pinaya is a Research Fellow from the Department of Biomedical Engineering at King's College London (KCL), working in the Artificial Medical Intelligence Group (AMIGO). He is currently working with the "Programme for High-Dimensional Neurology", a collaboration between UCL and KCL, aiming to create individually predictive models of clinical outcomes from high-dimensional analysis of large-scale clinical datasets. Previously, he held a Lecturer position at the Federal University of ABC (UFABC, Brazil) on Theoretical and Computational Neuroscience (2017-2020). He also worked at the Institute of Psychiatry, Psychology & Neuroscience (KCL) with Prof. Andrea Mechelli on the project "Using deep learning technology to make individualised inferences in brain-based disorders" (https://www.neurofind.ai/). He completed his PhD in Neuroscience and Cognition in 2017 at UFABC and has a background in Computer Science and Information Engineering, working with deep neural networks applied to medical imaging for the last 10 years. 

Chair: James Cole and Mariam Zabihi