We are delighted to celebrate the achievement of Tanmoy Debnath and Miroslaw Narbutt, whose paper, “Toward Digital Twin-Ready Brain Tumor Segmentation: A Multi-Pillar Comparison of Transformer and Convolutional Models,” received the Best Paper Award at BIOKDD 2026.
BIOKDD (Bioinformatics Knowledge Discovery and Data Mining) International Workshop on Data Mining in Bioinformatics is among the longest-running workshops at the ACM SIGKDD Conference and the longest-running event dedicated to biomedical data mining. For more than two decades, it has showcased cutting-edge research at the intersection of data science, artificial intelligence, and healthcare. Receiving the Best Paper Award at this prestigious forum is a significant recognition of the impact and quality of the team’s research.
The award-winning study addresses a critical challenge in neuro-oncology: accurately segmenting brain tumours from medical imaging to support patient-specific modelling and emerging digital twin technologies. Digital twins are virtual representations of individual patients that can help clinicians better understand disease progression, evaluate treatment options, and support personalised care.
While many segmentation studies focus primarily on overlap accuracy, the researchers demonstrate that this alone is insufficient for applications such as digital twin development. Their work introduces a rigorous, multi-dimensional evaluation of three leading brain tumour segmentation frameworks: nnU-Net v2, Universal Model, and VSmTrans.
Using the widely recognised MSD Task01 Brain Tumour benchmark, five-fold cross-validation, and comprehensive statistical analysis, the team compared model performance across multiple criteria, including overlap accuracy, boundary precision, and volumetric reliability. The results showed that VSmTrans achieved the strongest overall performance, providing an effective balance between geometric accuracy and volumetric agreement across tumour subregions. At the same time, the study highlighted an important clinical trade-off: nnU-Net v2 demonstrated lower volumetric bias for enhancing tumour estimation, despite VSmTrans delivering superior overall geometric fidelity.
These findings provide valuable evidence for researchers and clinicians seeking segmentation approaches that are robust enough for patient-specific modelling and digital twin-ready workflows. By demonstrating the importance of evaluating both geometric and volumetric performance, the study advances the field toward more reliable and clinically meaningful AI-driven tools in brain cancer care.
Congratulations to Tanmoy Debnath and Miroslaw Narbutt on this outstanding achievement and on receiving the BIOKDD 2026 Best Paper Award. Their work exemplifies the transformative potential of AI and data science in advancing precision medicine and improving patient outcomes.