Seminars on AI for cardiovascular and cancer research
par BUTEL Nathalie
A series of three talks on AI for cardiovascular and cancer research will be delivered by Prof. Anubha Gupta, from IIIT-Delhi, India, who's visiting the i3S Lab from Sept. 22 to Sept. 30, in the framework of the Franco-Indian Campus in Life Sciences of Université Côte d'Azur.
Talks 2 and 3 will be jointly presented with Prof. Ritu Gupta, from All India Institute of Medical Sciences.
The talks will take place in conference room (007), Les Algorithmes, Euclide-B, i3S Laboratory.
Details on the talks' contents can be found below and the speakers' biographies are attached.
Program
- Tuesday, September 23, 2pm-3pm - Talk 1 : Explainable AI models for CVD detection and risk prediction
Abstract : In this talk, we will cover our work on deep learning architectures including CNNs, ViT, and knowledge fusion on ECG data analysis for the identification of cardiac disorders, stress level prediction, mortality prediction using GNN (GCN, GraphSAGE, graph attention), augmentation of synthetic data generated using diffusion models, and our exploration of CVD detection using retinal images. In addition, we have tried to understand the interpretability of these AI models by identifying changes on ECG waveforms as observed by doctors on raw ECG waveforms.
- Wednesday, September 24, 9am-10am - Talk 2 : AI in blood cancer imaging
Abstract : SBILab, IIIT-Delhi and AIIMS, New Delhi have been collaborating in the area of blood cancer imaging for the past 10 years. During this time, we have worked on two blood cancer types: B-ALL and Multiple Myeloma. We have also released four curated datasets publicly and also organized two medical imaging challenges in the International conferences. We have built deep learning based inhouse solutions for blood cancer diagnosis. In this talk, we will briefly discuss the problem statement, datasets, workflow, the data challenges, and the solutions proposed.
- Friday, September 26, 10am-11am - Talk 3 : Design of a targeted panel of 295 genes in multiple myeloma using deep learning on NGS data
Abstract : Multiple myeloma (MM), a hematological cancer, evolves from a benign premalignant stage of monoclonal gammopathy of undetermined significance (MGUS) at a low and variable rate. MM shares genomic features with MGUS. Identifying distinctive features of these two entities is vital to understanding myeloma pathogenesis. We present a clinically-oriented targeted panel of 295 genes that potentially cause MGUS-to-MM transition and influence survival outcome in MM. Additionally, we introduce the concepts of transformative genes that are significantly altered in the malignant stage (MM) but not in the premalignant stage (MGUS). This is achieved by designing an attention-based graph neural network, namely BIO-DGI, that extracts gene-gene interactions utilizing a-priori information from nine protein-protein interaction (PPI) databases. BIO-DGI analyzes the whole-exome sequencing data of 1154 MM and 61 MGUS samples collected from a diverse population of American, European, and Indian subcontinents. This 295-gene panel can provide critical insights into disease progression.