Keynote Speaker

Topic of the keynote:

Enhancing 3D spatial perception in endoscopy

Endoscopy is a fundamental tool for the early detection and treatment of diseases such as gastrointestinal cancer or inflammatory pathologies. However, standard procedures rely on monocular imaging, providing only two-dimensional views of a complex environment. This makes it difficult to accurately estimate lesion size, perceive depth, and maintain spatial awareness, often leading to operator-dependent variability. This talk presents a set of computational approaches to enhance spatial understanding directly from conventional endoscopic video. We first address the problem of unreliable lesion size estimation through a geometry-driven method that combines multi-view reconstruction with depth-aware correction, enabling more accurate measurements in realistic conditions. We then tackle the lack of annotated data by introducing a high-fidelity synthetic dataset that closely mimics real endoscopic scenes, supporting the development of learning-based models. Building on this, we explore methods for endoscopic localisation and mapping that allow the reconstruction of consistent 3D representations from monocular sequences. Finally, we present a reconstruction framework based on Gaussian Splatting that produces detailed, stable scene representations in challenging settings. Overall, this work shows how integrating geometry, learning, and realistic simulation can move endoscopy beyond 2D visualisation toward more accurate and reliable 3D understanding to support clinical practice.

Scientific Biography

Chiara Lena is a researcher in biomedical engineering and computer vision, with a particular emphasis on artificial intelligence in medical imaging. Her research focuses on the development of deep and machine learning methodologies for endoscopic applications, with a particular emphasis on gastrointestinal imaging and fetoscopy. She has contributed to several peer-reviewed publications on depth estimation, 3D reconstruction, and image mosaicking from endoscopic video. Her research encompasses the development of learning-based frameworks for occlusion recovery and navigation in fetoscopy, as well as novel approaches for reconstructing three-dimensional structures from colonoscopy data. Her work also includes methodologies for the segmentation and registration of anatomical structures to enhance visual understanding in minimally invasive procedures. More recently, she has explored the use of foundation models and advanced AI techniques, including large language models and temporally aware architectures, to improve the analysis and interpretation of surgical video and endoscopic imagery. Her research interests include computer vision, deep learning, and machine learning for medical applications, with a particular focus on enhancing perception, navigation, and decision support in minimally invasive surgical procedures.

Program

The SASYR program can be downloaded here:

Publication

available soon