Abstract: In the healthcare AI and Software as a Medical Device (SaMD) domain, MLOps faces unique challenges beyond the typical scale and performance considerations. Using Onc.AI’s oncology clinical management platform as a case study, this talk will address technical intricacies of processing diverse data types, including structured, unstructured, and volumetric imaging datasets. We'll explore methodologies for imaging biomarker discovery through AI/ML training paradigms, emphasizing the need for consistent and reproducible pipelines. Highlighting the importance of traceability across training, validation, and deployment phases, we'll discuss the mechanisms ensuring compliance in regulated industries like healthcare. Additionally, we'll outline the tailored validation workflows required to meet regulatory standards and share strategies for effective model deployment and serving. Throughout the presentation, the role of platforms such as Valohai in navigating these technical challenges will be underscored, providing a clear view of MLOps in healthcare applications.
Bio: Petr Jordan is the CTO and founding team member at Onc.AI, a digital health company dedicated to improving oncology decision-making. Using comprehensive multi-institutional real-world datasets, Onc.AI is pioneering the development of AI-driven therapy response prediction tools and imaging biomarkers in precision oncology. Petr is an experienced research and development leader with expertise in medical imaging, machine learning, oncology, and medical physics. Before joining Onc.AI, he led advanced machine learning and computer vision teams at Varian Medical Systems and Accuray Incorporated. He holds a PhD in Medical Engineering and Medical Physics from the Harvard-MIT Division of Health Sciences and Technology and completed core medical school curriculum at Harvard Medical School. He also holds an MS from Harvard University in Engineering Science and a BS from Lipscomb University in Computer Engineering. He is an inventor on 35 granted US patents.