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.
Original language | English |
---|---|
Title of host publication | Information Processing in Medical Imaging - 28th International Conference, IPMI 2023, Proceedings |
Editors | Alejandro Frangi, Marleen de Bruijne, Demian Wassermann, Nassir Navab |
Publisher | Springer Science and Business Media Deutschland GmbH |
Pages | 755-767 |
Number of pages | 13 |
ISBN (Print) | 9783031340475 |
DOIs | |
Publication status | Published - 2023 |
Event | 28th International Conference on Information Processing in Medical Imaging, IPMI 2023 - San Carlos de Bariloche, Argentina Duration: 18 Jun 2023 → 23 Jun 2023 |
Publication series
Name | Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) |
---|---|
Volume | 13939 LNCS |
ISSN (Print) | 0302-9743 |
ISSN (Electronic) | 1611-3349 |
Conference
Conference | 28th International Conference on Information Processing in Medical Imaging, IPMI 2023 |
---|---|
Country/Territory | Argentina |
City | San Carlos de Bariloche |
Period | 18/06/2023 → 23/06/2023 |
Keywords
- instance imbalance awareness
- lightsheet microscopy
- multiple sclerosis
- semantic segmentation loss function
Fingerprint
Dive into the research topics of 'blob loss: Instance Imbalance Aware Loss Functions for Semantic Segmentation'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver
}
Information Processing in Medical Imaging - 28th International Conference, IPMI 2023, Proceedings. ed. / Alejandro Frangi; Marleen de Bruijne; Demian Wassermann; Nassir Navab. Springer Science and Business Media Deutschland GmbH, 2023. p. 755-767 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13939 LNCS).
Research output: Chapter in Book/Report/Conference proceeding › Conference paper › peer-review
TY - CHAP
T1 - blob loss
T2 - 28th International Conference on Information Processing in Medical Imaging, IPMI 2023
AU - Kofler, Florian
AU - Shit, Suprosanna
AU - Ezhov, Ivan
AU - Fidon, Lucas
AU - Horvath, Izabela
AU - Al-Maskari, Rami
AU - Li, Hongwei Bran
AU - Bhatia, Harsharan
AU - Loehr, Timo
AU - Piraud, Marie
AU - Erturk, Ali
AU - Kirschke, Jan
AU - Peeken, Jan C.
AU - Vercauteren, Tom
AU - Zimmer, Claus
AU - Wiestler, Benedikt
AU - Menze, Bjoern
N1 - Funding Information: Acknowledgements. This work was supported by the National Natural Science Foundation of China (31971288, U1801265, 61936007, U20B2065, 61976045, 62276050, and 61976045), National Key R&D Program of China (2020AAA0105701), Sichuan Science and Technology Program (2021YJ0247), Innovation and Technology Commission-Innovation and Technology Fund ITS/100/20, Doctor Dissertation of Northwestern Polytechnical University CX2022053. Funding Information: Acknowledgements. Computing resources and support were provided by AINOS-TICS Ltd., enabled through funding by Innovate UK. MG is funded by UKRI under grant MR/W004097/1. Funding Information: Acknowledgements. This research was in part supported by the NSF grant IIS-1724174, the NIH NINDS and NIA via RF1NS121099 to Vemuri and the MOST grant 110-2118-M-002-005-MY3 to Yang. Funding Information: Foundation, DMS 1912194, Simons Foundation Collab. on the Global Brain. Yemini: Klingenstein-Simons Fellowship in Neuroscience, Hypothesis Fund. Dey: NIH NIBIB NAC P41EB015902, NIBIB 5R01EB032708. Varol: 1K99MH128772-01A1. Venkatacha-lam: Burroughs Wellcome Fund and NIH R01 NS126334. Funding Information: Acknowledgments. This work was supported by the Hong Kong Innovation and Technology Commission under Project No. ITS/238/21. Funding Information: Acknowledgements. Support for this work was provided by National Science Foundation grant 1944247 and National Institutes of Health grants U19AG056169 and 5R01AG070937 to C.M. Funding Information: Partially supported by the EANS 2021 Leica Research Grant. Funding Information: This work is partly supported by JSPS KAKENHI JP23KJ0118. Funding Information: Acknowledgements. This research/project is supported by the Singapore Ministry of Education (Academic research fund Tier 1) and A*STAR (H22P0M0007). Additional funding is provided by the National Science Foundation MDS-2010778, National Institute of Health R01 EB022856, EB02875. This research was also supported by the A*STAR Computational Resource Centre through the use of its high-performance computing facilities. Funding Information: financed by grants from Region Stockholm ALF Medicine, Styrgruppen KI/Region Stockholm for Research in Odontology and research funds from Karolinska Institutet and KTH. Funding Information: Acknowledgements. This research was partially supported by the Bavarian State Ministry of Science and the Arts and coordinated by the bidt, the BMBF (DeepMentia, 031L0200A), the DFG and the LRZ. Funding Information: Acknowledgments. This work was supported in part by the United States National Institutes of Health (NIH) through grants MH125479 and EB008374. Funding Information: Acknowledgement. This work was supported by Shenzhen Science and Technology Innovation Committee (Project No. SGDX20210823103201011) and Hong Kong Innovation and Technology Fund (Project No. ITS/028/21FP). Funding Information: Acknowledgments. This work is supported in part by the Natural Sciences and Engineering Research Council of Canada (NSERC), and NVIDIA Hardware Award, and Institute of Information & Communications Technology Planning & Evaluation (IITP) grant funded by the Korea government (MSIT) No. 2022-0-00959 ((Part 2) Few-Shot Learning of Causal Inference in Vision and Language for Decision Making), and the MOTIE (Ministry of Trade, Industry, and Energy) in Korea, under Human Resource Development Program for Industrial Innovation (Global) (P0017311) supervised by the Korea Institute for Advancement of Technology (KIAT). Funding Information: Acknowledgements. This work was partially supported by the EPSRC Centre for Doctoral Training in Fluid Dynamics (EP/L01615X/1) and the Royal Academy of Engineering Chair in Emerging Technologies (CiET1919/19). The computational work was undertaken on the UK National Tier-2 high performance computing service JADE-2 (EP/T022205/1). Funding Information: Acknowledgement. This work was partially supported by grants from NIH (R21AG065942, R01EY032125, and R01DE030286). Funding Information: Acknowledgement. The reported research was partly supported by NIH award # 1R21CA258493-01A1, NSF awards IIS-2212046 and IIS-2123920, and Stony Brook OVPR seed grants. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. Funding Information: Acknowledgement. This work was supported by the Wellcome/EPSRC Centre for Interventional and Surgical Sciences [203145Z/16/Z] and the International Alliance for Cancer Early Detection, an alliance between Cancer Research UK [C28070/A30912; C73666/A31378], Canary Center at Stanford University, the University of Cambridge, OHSU Knight Cancer Institute, University College London and the University of Manchester. Funding Information: Supported by Natural Sciences and Engineering Research Council of Canada (NSERC) and Reseau de BioImagrie du Quebec (RBIQ). Funding Information: Work supported by MIT JClinic, Philips, and Wistron. Funding Information: Acknowledgement. This research was partly supported by the National Key R&D Program of China (Grant No. 2020AAA0108303), Shenzhen Science and Technology Project (Grant No. JCYJ20200109143041798), Shenzhen Stable Supporting Program (Grant No. WDZC20200820200655001), and Shenzhen Key Lab of next generation interactive media innovative technology (Grant No. ZDSY S20210623092001004). Funding Information: Acknowledgement. This work is supported by the Fundamental Research Funds for the Central Universities. Funding Information: Acknowledgements. We gratefully acknowledge the tissue donors and their families. This work was supported by the NIH (Grants RF1 AG069474, P30 AG072979 and R01 AG056014), a UCLM travel and research grant (to R.I), and an Alzheimer’s Association grant (AARF-19-615258) (to L.E.M.W). Funding Information: Acknowledgements. ALY is supported by an MRC Skills Development Fellowship (MR/T027800/1). Data collection and sharing for this project was funded by the Alzheimer’s Disease Neuroimaging Initiative (ADNI) (National Institutes of Health Grant U01 AG024904) and DOD ADNI (Department of Defense award number W81XWH-12-2-0012) (For a full list of ADNI funders see: https://adni.loni.usc.edu/ wp-content/uploads/how to apply/ADNI Data Use Agreement.pdf). Funding Information: H. Dai and S. Joshi were supported by NSF grant DMS-1912030. P. T. Fletcher was supported by NSF grant IIS-2205417. M. Bauer was supported by NSF grants DMS-1912037, DMS-1953244 and by FWF grant FWF-P 35813-N. Funding Information: Acknowledgments. This work was supported by NIH R01-AG067103 grant. Computations were performed on Washington University Center for High Performance Computing. Funding Information: Acknowledgement. This work was supported in part by National Natural Science Foundation of China (grant number 62131015), Science and Technology Commission of Shanghai Municipality (STCSM) (grant number 21010502600), and The Key R&D Program of Guangdong Province, China (grant number 2021B0101420006). Funding Information: Acknowledgments. This work is supported in part by the Natural Sciences and Engineering Research Council of Canada (NSERC), Public Safety Canada (NS-5001-22170), in part by NVIDIA Hardware Award, and in part by the Hong Kong Innovation and Technology Commission under Project No. ITS/238/21. Funding Information: Acknowledgment. This work was supported in part by National Natural Science Foundation of China (No. 62131015), Science and Technology Commission of Shanghai Municipality (STCSM) (No. 21010502600), The Key R&D Program of Guangdong Province, China (No. 2021B0101420006), and the China Postdoctoral Science Foundation (Nos. BX2021333, 2021M703340). This work was completed under the close collaboration between C. Jiang and Y. Pan, and they contributed equally to this work. Publisher Copyright: © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.
PY - 2023
Y1 - 2023
N2 - 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.
AB - 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.
KW - instance imbalance awareness
KW - lightsheet microscopy
KW - multiple sclerosis
KW - semantic segmentation loss function
UR - http://www.scopus.com/inward/record.url?scp=85163977420&partnerID=8YFLogxK
U2 - 10.1007/978-3-031-34048-2_58
DO - 10.1007/978-3-031-34048-2_58
M3 - Conference paper
AN - SCOPUS:85163977420
SN - 9783031340475
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 755
EP - 767
BT - Information Processing in Medical Imaging - 28th International Conference, IPMI 2023, Proceedings
A2 - Frangi, Alejandro
A2 - de Bruijne, Marleen
A2 - Wassermann, Demian
A2 - Navab, Nassir
PB - Springer Science and Business Media Deutschland GmbH
Y2 - 18 June 2023 through 23 June 2023
ER -