xAILab Bamberg attends the 21st IEEE International Symposium on Biomedical Imaging in Athens

Research on Privacy and Adaptability

The paper introduces a method to enhance the transparency, adaptability, and privacy-awareness of image classification models. Traditional deep learning models typically encode knowledge within their model parameters, limiting adaptability and raising privacy concerns when handling sensitive, medical data.

To address this challenge, the paper integrates a k-Nearest Neighbor (k-NN) classifier with a vision-based foundation model pre-trained on natural images in a self-supervised manner. This integration allows for storing representative feature embeddings of the input data independently from the model’s weights, enabling dynamic data modifications without requiring retraining of the model.

Assessments across established benchmarks and tasks, including continual learning and data removal scenarios, demonstrated improvements in interpretability and adaptability, suggesting a potential solution for privacy-aware image classification in medical imaging.
For more details, you can access the full paper here and the associated code repository here.

Athens – A Historic Venue for Innovation

Athens, a city with a rich history spanning over 3,000 years, provided a vibrant and inspiring backdrop for ISBI 2024. The Megaron Athens International Conference Center, renowned for hosting world-class events, was the perfect venue for this year's symposium. Additionally, Participants had the opportunity to explore Athens' numerous cultural and historical landmarks, including the iconic Acropolis, the Panathenaic Stadium, ancient temples, and bustling neighborhoods like Plaka and Psyrri.

A Convergence of Experts and Ideas

The ISBI conference series, jointly organized by the IEEE Signal Processing Society and the IEEE Engineering in Medicine and Biology Society, aims to foster collaboration between researchers, healthcare professionals, and industry experts. This year's edition extended its focus to emerging AI frontiers in biomedical imaging, emphasizing interpretability, domain shifts, adaptation, and trustworthiness. The program featured keynote talks from world-renowned imaging scientists and clinicians, including Dr. Francis Bach, Dr. Katherine Ferrara, Dr. Anant Madabhushi, and Dr. Joseph Sifakis, a 2007 Turing Award laureate.

A notable addition this year was the Clinical Focus Sessions, central to the technical program, allowing participants to engage fully without conflicting schedules. The inaugural Pharma-Meets-Imaging event also brought together pharmaceutical companies and imaging scientists to discuss innovations in drug discovery and imaging technology.

An Enriching Experience

ISBI 2024 was not just about technical sessions and presentations. The conference also featured the Art-in-Biomedical-Imaging event, encouraging participants to explore the artistic aspects of medical imaging. This event showcased the creativity and innovation of researchers, blending science with art in an exceptional manner.

Additionally, in a unique event at the Panathenaic Stadium, the historic venue of the first modern Olympic Games, participants had the opportunity to run a symbolic marathon, connecting the legacy of ancient athletic traditions with modern scientific endeavors.

In summary, ISBI 2024 in Athens provided valuable insights and fostered collaborations in biomedical imaging through talks, presentations, plenums, and social events. The xAILab’s presentation on “Integrating kNN with Foundation Models for Adaptable and Privacy-Aware Image Classification” was well-received, and the team plans to continue their research in this area.