blob loss: Instance Imbalance Aware Loss Functions for Semantic Segmentation

Kofler, Florian; Shit, Suprosanna; Ezhov, Ivan; Fidon, Lucas; Horvath, Izabela; Al-Maskari, Rami; Li, Hongwei Bran; Bhatia, Harsharan Singh; Loehr, Timo; Piraud, Marie; Ertürk, Ali ; Kirschke, Jan; Peeken, Jan C.; Vercauteren, Tom; Zimmer, Claus; Wiestler, Benedikt und Menze, Bjoern (2023): blob loss: Instance Imbalance Aware Loss Functions for Semantic Segmentation. 28th International Conference on Information Processing in Medical Imaging (IPMI), San Carlos De Bariloche, Argentina, 18. - 23. Juni 2023. Frangi, Alejandro; Bruijne, Marleen de; Wassermann, Demian und Navab, Nassir (Hrsg.): In: Information Processing in Medical Imaging, Lecture Notes in Computer Science Bd. 13939 Cham: Springer. S. 755-767

Abstract

Deep convolutional neural networks (CNN) have proven to be remarkably effective in semantic segmentation tasks. Most popular loss functions were introduced targeting improved volumetric scores, such as the Dice coefficient (DSC). By design, DSC can tackle class imbalance, however, it does not recognize instance imbalance within a class. As a result, a large foreground instance can dominate minor instances and still produce a satisfactory DSC. Nevertheless, detecting tiny instances is crucial for many applications, such as disease monitoring. For example, it is imperative to locate and surveil small-scale lesions in the follow-up of multiple sclerosis patients. We propose a novel family of loss functions, blob loss, primarily aimed at maximizing instance-level detection metrics, such as F1 score and sensitivity. Blob loss is designed for semantic segmentation problems where detecting multiple instances matters. We extensively evaluate a DSC-based blob loss in five complex 3D semantic segmentation tasks featuring pronounced instance heterogeneity in terms of texture and morphology. Compared to soft Dice loss, we achieve 5% improvement for MS lesions, 3% improvement for liver tumor, and an average 2% improvement for microscopy segmentation tasks considering F1 score.

• Dokumententyp: Konferenzbeitrag (Paper)
• Fakultät: Medizin > Munich Cluster for Systems Neurology (SyNergy) ; Medizin > Institut für Schlaganfall- und Demenzforschung (ISD)
• Themengebiete: 000 Informatik, Informationswissenschaft, allgemeine Werke > 004 Informatik ; 600 Technik, Medizin, angewandte Wissenschaften > 610 Medizin und Gesundheit
• ISBN: 978-3-031-34047-5 ; 978-3-031-34048-2 ; 978-3-031-34049-9
• ISSN: 0302-9743
• Ort: Cham
• Sprache: Englisch
• DFG: Gefördert durch die Deutsche Forschungsgemeinschaft (DFG) - 390857198