Our work introduces MML-SurgAdapt, a unified CLIP-based multi-task framework capable of handling diverse surgical tasks through natural language supervision. To address the challenge of partial annotations when integrating multiple tasks, we employ Single Positive Multi-Label (SPML) learning, which enables effective learning even with noisy labels while reducing annotation cost by requiring only one positive label per image. We validate our approach on a combined laparoscopic cholecystectomy dataset spanning tasks of varying granularity and demonstrate that MML-SurgAdapt achieves performance comparable to traditional task-specific models trained with complete annotations.
I recently graduated with a bachelor's degree in Electronics and Communications Engineering and have been studying and working in the field of AI since college. This is my first work to be published in a conference. It is a great honor to have my work accepted at such a prestigious conference, especially at such an early stage of my research career. Presenting at MICCAI as an undergraduate is both exciting and humbling, and I am grateful that my work will be shared with and recognized by leading researchers in the field. This opportunity not only validates my efforts but also motivates me to keep learning and growing.
In addition to exploring the latest works in the surgical AI community, I am eager to explore research in radiology, which aligns with my current work. I am also interested in learning more about how popular frameworks such as generative models, language models, and multimodal approaches are being applied in medical AI.
At the conference, I am looking forward to meeting experienced researchers, making connections, and having good conversations about topics that I am interested in. Presenting my poster is another highlight, as it gives me the opportunity to share my work and receive valuable feedback from knowledgeable experts. I am also excited to attend the keynote sessions and gain insights from leaders in the field.