Seminar “U-Mamba2 and U-Mamba2-SSL: Optimising Speed for Accurate CBCT Segmentation”

Mercoledì 27 Maggio 2026 09:00

TL;DR: On May 27 (9:00–13:00), Dr. Zhi Qin Tan from King’s College London will present at the UNIMORE AI Center to introduce U-Mamba2 and U-Mamba2-SSL, the award-winning AI models for 3D dental anatomy segmentation from CBCT scans. The seminar will highlight how optimizing for faster computational speed was the real key to making AI practical in clinical settings, giving you an inside look at how to win three international challenges at MICCAI!

Seminar announcement

On May 27, at UNIMORE AI Center, from 9:00 to 13:00, Dr. Zhi Qin Tan, from King’s College, London will deliver a seminar titled: “U-Mamba2 and U-Mamba2-SSL: Optimising Speed for Accurate CBCT Segmentation”

Abstract. Accurate 3D segmentation of orofacial anatomies such as jaws, teeth, and pulp via Cone-Beam Computed Tomography (CBCT) images is critical for clinical applications such as dental surgical planning, yet it remains computationally challenging and time-consuming. In this talk, we will present an in-depth look at U-Mamba2 and U-Mamba2-SSL, the winning solutions for the three CBCT multi-anatomy segmentation challenges at the MICCAI 2025 ODIN Workshop. We will walk through our entire development pipeline: from the initial idea to rigorous optimisation and evaluation results. The competition results demonstrate that our models succeeded not by chasing marginal accuracy improvement, but by delivering faster inference speed. Our findings also show that prioritising inference speed over accuracy can yield practical outcomes in resource-constrained clinical settings.  

Biography. Zhi Qin Tan is a final-year PhD student at King’s College London under the academic supervision of Dr. Yunpeng Li and Prof. Owen Addison. Prior to joining King’s College London, he obtained his Bachelor of Computer Science degree with first-class honours from the University of Malaya and worked as a Senior Data Scientist at Datium Insights. His research interests lie in the intersection between trustworthy and multimodal AI, as well as their application in healthcare, specifically the dentistry domain.

This event is sponsored by the Emilia-Romagna Region under the Operation “Seminars on Advanced Topics in Artificial Intelligence” (Ref. P.A. 2024-23126/RER, Project 1), approved by Regional Government Decree (DGR) No. 1964/2024: ACTION 2. CUP E93C24001700002.

Autore: Costantino GRANA - Ultima modifica: Venerdì 22 Maggio 2026 18:44