• LST-AI: a Deep Learning Ensemb...

Result: LST-AI: a Deep Learning Ensemble for Accurate MS Lesion Segmentation.

Title:
LST-AI: a Deep Learning Ensemble for Accurate MS Lesion Segmentation.
Authors:
Wiltgen T; Department of Neurology, School of Medicine, Klinikum rechts der Isar, Technical University of Munich, Munich, Germany.; TUM-Neuroimaging Center, School of Medicine, Technical University of Munich, Munich, Germany., McGinnis J; Department of Neurology, School of Medicine, Klinikum rechts der Isar, Technical University of Munich, Munich, Germany.; TUM-Neuroimaging Center, School of Medicine, Technical University of Munich, Munich, Germany.; Department of Computer Science, Institute for AI in Medicine, Technical University of Munich, Munich, Germany., Schlaeger S; Department of Diagnostic and Interventional Neuroradiology, School of Medicine, Klinikum rechts der Isar, Technical University of Munich, Munich, Germany., Kofler F; Department of Computer Science, Institute for AI in Medicine, Technical University of Munich, Munich, Germany.; Department of Diagnostic and Interventional Neuroradiology, School of Medicine, Klinikum rechts der Isar, Technical University of Munich, Munich, Germany.; TranslaTUM, Center for Translational Cancer Research, Munich, Germany.; Helmholtz AI, Helmholtz Munich, Neuherberg, Germany., Voon C; Department of Neurology, School of Medicine, Klinikum rechts der Isar, Technical University of Munich, Munich, Germany.; TUM-Neuroimaging Center, School of Medicine, Technical University of Munich, Munich, Germany., Berthele A; Department of Neurology, School of Medicine, Klinikum rechts der Isar, Technical University of Munich, Munich, Germany., Bischl D; Department of Diagnostic and Interventional Neuroradiology, School of Medicine, Klinikum rechts der Isar, Technical University of Munich, Munich, Germany., Grundl L; Department of Diagnostic and Interventional Neuroradiology, School of Medicine, Klinikum rechts der Isar, Technical University of Munich, Munich, Germany., Will N; Department of Diagnostic and Interventional Neuroradiology, School of Medicine, Klinikum rechts der Isar, Technical University of Munich, Munich, Germany., Metz M; Department of Diagnostic and Interventional Neuroradiology, School of Medicine, Klinikum rechts der Isar, Technical University of Munich, Munich, Germany., Schinz D; Department of Diagnostic and Interventional Neuroradiology, School of Medicine, Klinikum rechts der Isar, Technical University of Munich, Munich, Germany.; Institute of Radiology, University Hospital Erlangen, Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany., Sepp D; Department of Diagnostic and Interventional Neuroradiology, School of Medicine, Klinikum rechts der Isar, Technical University of Munich, Munich, Germany., Prucker P; Department of Diagnostic and Interventional Neuroradiology, School of Medicine, Klinikum rechts der Isar, Technical University of Munich, Munich, Germany., Schmitz-Koep B; Department of Diagnostic and Interventional Neuroradiology, School of Medicine, Klinikum rechts der Isar, Technical University of Munich, Munich, Germany., Zimmer C; Department of Diagnostic and Interventional Neuroradiology, School of Medicine, Klinikum rechts der Isar, Technical University of Munich, Munich, Germany., Menze B; Department of Quantitative Biomedicine, University of Zurich, Zurich, Switzerland., Rueckert D; Department of Computer Science, Institute for AI in Medicine, Technical University of Munich, Munich, Germany.; Department of Computing, Imperial College London, London, United Kingdom., Hemmer B; Department of Neurology, School of Medicine, Klinikum rechts der Isar, Technical University of Munich, Munich, Germany.; Munich Cluster for Systems Neurology (SyNergy), Munich, Germany., Kirschke J; Department of Diagnostic and Interventional Neuroradiology, School of Medicine, Klinikum rechts der Isar, Technical University of Munich, Munich, Germany., Mühlau M; Department of Neurology, School of Medicine, Klinikum rechts der Isar, Technical University of Munich, Munich, Germany.; TUM-Neuroimaging Center, School of Medicine, Technical University of Munich, Munich, Germany., Wiestler B; Department of Diagnostic and Interventional Neuroradiology, School of Medicine, Klinikum rechts der Isar, Technical University of Munich, Munich, Germany.; TranslaTUM, Center for Translational Cancer Research, Munich, Germany.
Source:
MedRxiv : the preprint server for health sciences [medRxiv] 2024 Mar 11. Date of Electronic Publication: 2024 Mar 11.
Publication Type:
Preprint; Journal Article
Language:
English
Journal Info:
Country of Publication: United States NLM ID: 101767986 Publication Model: Electronic Cited Medium: Internet NLM ISO Abbreviation: medRxiv Subsets: PubMed not MEDLINE
Comments:
Update in: Neuroimage Clin. 2024;42:103611. doi: 10.1016/j.nicl.2024.103611.. (PMID: 38703470)
References:
Neuroimage. 2012 Feb 15;59(4):3774-83. (PMID: 22119648)
Neuroimage Clin. 2019;21:101638. (PMID: 30555005)
Comput Biol Med. 2022 Jun;145:105402. (PMID: 35344864)
Lancet. 2018 Apr 21;391(10130):1622-1636. (PMID: 29576504)
IEEE J Biomed Health Inform. 2022 Jun;26(6):2680-2692. (PMID: 35171783)
Nat Methods. 2021 Feb;18(2):203-211. (PMID: 33288961)
Neuroimage. 2017 Mar 1;148:77-102. (PMID: 28087490)
Gigascience. 2024 Jan 2;13:. (PMID: 39102518)
Sci Rep. 2023 Mar 13;13(1):4102. (PMID: 36914715)
Front Neuroinform. 2020 Nov 20;14:610967. (PMID: 33328949)
Sci Rep. 2018 Sep 12;8(1):13650. (PMID: 30209345)
Nat Rev Dis Primers. 2018 Nov 8;4(1):43. (PMID: 30410033)
Mult Scler Relat Disord. 2019 Apr;29:124-129. (PMID: 30711877)
Front Neurosci. 2020 Apr 29;14:125. (PMID: 32410929)
Neuroimage Clin. 2020;27:102357. (PMID: 32739882)
Neuroimage. 2021 Dec 1;244:118589. (PMID: 34563682)
Lancet Neurol. 2018 Feb;17(2):162-173. (PMID: 29275977)
Nat Commun. 2022 Jul 15;13(1):4128. (PMID: 35840566)
Neuroimage. 2018 Dec;183:650-665. (PMID: 30125711)
Neuroinformatics. 2018 Jan;16(1):51-63. (PMID: 29103086)
Neuroimage. 2021 Jan 15;225:117471. (PMID: 33099007)
Hum Brain Mapp. 2022 Feb 15;43(3):929-939. (PMID: 34704337)
Neuroimage Clin. 2023;37:103311. (PMID: 36623350)
Hum Brain Mapp. 2019 Dec 1;40(17):4952-4964. (PMID: 31403237)
Hum Brain Mapp. 2023 Oct 1;44(14):4893-4913. (PMID: 37530598)
Front Neurosci. 2022 Oct 26;16:975862. (PMID: 36389254)
Neuroradiology. 2022 Nov;64(11):2103-2117. (PMID: 35864180)
Proc IEEE Int Symp Biomed Imaging. 2021 Apr;2021:1971-1974. (PMID: 34367472)
Ther Adv Neurol Disord. 2023 Mar 24;16:17562864231161892. (PMID: 36993939)
Sci Rep. 2021 Jan 13;11(1):1087. (PMID: 33441684)
Nat Commun. 2022 Dec 5;13(1):7346. (PMID: 36470898)
Med Image Anal. 2022 Feb;76:102312. (PMID: 34894571)
Brain. 2012 Oct;135(Pt 10):2952-61. (PMID: 23065788)
Neuroimage Clin. 2020;27:102335. (PMID: 32663798)
Neuroimage Clin. 2021;31:102707. (PMID: 34111718)
Neuroimage. 2006 Jul 1;31(3):1116-28. (PMID: 16545965)
Grant Information:
R01 NS112161 United States NS NINDS NIH HHS
Contributed Indexing:
Keywords: Artificial Intelligence; Deep Learning; Lesion Segmentation; Magnetic Resonance Imaging; Multiple Sclerosis; White Matter Lesions
Entry Date(s):
Date Created: 20231204 Latest Revision: 20250903
Update Code:
20250904
PubMed Central ID:
PMC10690346
DOI:
10.1101/2023.11.23.23298966
PMID:
38045345
Database:
MEDLINE

Further Information

Automated segmentation of brain white matter lesions is crucial for both clinical assessment and scientific research in multiple sclerosis (MS). Over a decade ago, we introduced an engineered lesion segmentation tool, LST. While recent lesion segmentation approaches have leveraged artificial intelligence (AI), they often remain proprietary and difficult to adopt. As an open-source tool, we present LST-AI, an advanced deep learning-based extension of LST that consists of an ensemble of three 3D-UNets. LST-AI explicitly addresses the imbalance between white matter (WM) lesions and non-lesioned WM. It employs a composite loss function incorporating binary cross-entropy and Tversky loss to improve segmentation of the highly heterogeneous MS lesions. We train the network ensemble on 491 MS pairs of T1w and FLAIR images, collected in-house from a 3T MRI scanner, and expert neuroradiologists manually segmented the utilized lesion maps for training. LST-AI additionally includes a lesion location annotation tool, labeling lesion location according to the 2017 McDonald criteria (periventricular, infratentorial, juxtacortical, subcortical). We conduct evaluations on 103 test cases consisting of publicly available data using the Anima segmentation validation tools and compare LST-AI with several publicly available lesion segmentation models. Our empirical analysis shows that LST-AI achieves superior performance compared to existing methods. Its Dice and F1 scores exceeded 0.62, outperforming LST, SAMSEG (Sequence Adaptive Multimodal SEGmentation), and the popular nnUNet framework, which all scored below 0.56. Notably, LST-AI demonstrated exceptional performance on the MSSEG-1 challenge dataset, an international WM lesion segmentation challenge, with a Dice score of 0.65 and an F1 score of 0.63-surpassing all other competing models at the time of the challenge. With increasing lesion volume, the lesion detection rate rapidly increased with a detection rate of >75% for lesions with a volume between 10mm <sup>3</sup> and 100mm <sup>3</sup> . Given its higher segmentation performance, we recommend that research groups currently using LST transition to LST-AI. To facilitate broad adoption, we are releasing LST-AI as an open-source model, available as a command-line tool, dockerized container, or Python script, enabling diverse applications across multiple platforms.

Declaration of Competing Interests: The authors declare no competing interests.