Events

The seminar examined the ethical challenges of explainability in medical AI, highlighting concerns around opaque decision-making in high-stakes clinical settings. It discussed the implications of AI as a “black box” for accountability, responsibility, and patient autonomy, emphasizing the importance of addressing explainability as a core ethical requirement in healthcare AI.

The session presented advanced AI applications in cardiovascular and critical care, including a robust deep learning model for atrial fibrillation detection across diverse populations, challenges in ICU monitor data collection, and a causal machine learning framework for treatment recommendations in acute heart failure. It also showcased projects developed at the TERA Hackathon, demonstrating AI-driven solutions for early infection risk detection and improved birth weight prediction.

The seminar explored innovation in medicine through close collaboration between clinicians, scientists, and engineers, highlighting the full innovation cycle from unmet clinical needs to technological solutions and regulatory approval. Case studies in cardiology and medical robotics illustrated how interdisciplinary research and algorithmic motion planning were advancing minimally invasive procedures and shaping the future of medical innovation at Rambam and the Technion.

The sessions highlighted the application of AI in healthcare, from building a clinician-first, machine learning–based platform for primary care to introducing a novel text-mining tool that enhanced medical diagnosis through large-scale, up-to-date analysis of biomedical literature.

Advances in deep learning and the availability of large ECG datasets led to rapid growth in machine learning research for ECG analysis. The seminar critically examined common limitations in the literature, including label disagreement, noise, bias, and evaluation issues, and discussed strategies to address these challenges and identify promising directions for future research.

The seminar presented advances in AI-driven oncology research, including the use of deep learning to predict PD-L1 expression in breast cancer from standard H&E-stained images, reducing reliance on costly immunohistochemistry. It also introduced the application of machine learning models in oncology research, focusing on classification approaches and the early prediction of immune-related treatment toxicity.