LST-AI: A deep learning ensemble for accurate MS lesion segmentation

• Published in: NeuroImage clinical Vol. 42; p. 103611
• Main Authors: Wiltgen, Tun, McGinnis, Julian, Schlaeger, Sarah, Kofler, Florian, Voon, CuiCi, Berthele, Achim, Bischl, Daria, Grundl, Lioba, Will, Nikolaus, Metz, Marie, Schinz, David, Sepp, Dominik, Prucker, Philipp, Schmitz-Koep, Benita, Zimmer, Claus, Menze, Bjoern, Rueckert, Daniel, Hemmer, Bernhard, Kirschke, Jan, Mühlau, Mark, Wiestler, Benedikt
• Format: Journal Article
• Language: English
• Published: Netherlands Elsevier Inc 2024; Elsevier
• Subjects: Adult; Artificial Intelligence; Brain - diagnostic imaging; Brain - pathology; Deep Learning; Female; Humans; Image Processing, Computer-Assisted - methods; Lesion Segmentation; Magnetic Resonance Imaging; Magnetic Resonance Imaging - methods; Male; Multiple Sclerosis; Multiple Sclerosis - diagnostic imaging; Multiple Sclerosis - pathology; Neuroimaging - methods; Neuroimaging - standards; Radiology; White Matter - diagnostic imaging; White Matter - pathology; White Matter Lesions; RRMS; N/A; CNN; PPMS; MRI; PV; Lesion Segmentation; CPU; SAMSEG; GPU; SC; Multiple Sclerosis; White Matter Lesions; ASD; PPVL; DSC; WM; PPV; FLAIR; ON; SPMS; MS; Artificial Intelligence; AI; IQR; IT; CIS; LST; LST-LPA; Deep Learning; TE; Magnetic Resonance Imaging; SensL; TI; LST-LGA; JC; ReLU; FA; T1w; TR; positive predictive value; not applicable/available; central processing unit; inversion time; artificial intelligence; average surface distance; echo time; convolutional neural networks; lesion-wise positive predictive value; secondary progressive multiple sclerosis; rectified linear unit; lesion segmentation tool lesion growth algorithm; repetition time; T1-weighted; optic neuritis; subcortical; primary progressive multiple sclerosis; periventricular; flip angle; magnetic resonance imaging; lesion-wise sensitivity; interquartile range; white matter; lesion segmentation tool lesion prediction algorithm; fluid-attenuated inversion recovery; dice similarity coefficient; multiple sclerosis; infratentorial; sequence adaptive multimodal segmentation; relapsing-remitting multiple sclerosis; lesion segmentation tool; juxtacortical; graphics processing unit; clinically isolated syndrome
• ISSN: 2213-1582; 2213-1582
• DOI: 10.1016/j.nicl.2024.103611

•Free and open-source lesion-segmentation tool (LST-AI), allowing widespread use and community contributions.•Tested on multiple external datasets and consistently outperforms existing models.•Ensemble of 3D U-Nets and composite loss functions to optimize performance.•Enhanced detection rate, especially for small lesions 10–100 mm3.•Includes lesion location annotation per 2017 McDonald criteria. 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 U-Nets. 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 T1-weighted and FLAIR images, collected in-house from a 3T MRI scanner, and expert neuroradiologists manually segmented the utilized lesion maps for training. LST-AI also includes a lesion location annotation tool, labeling lesions as periventricular, infratentorial, and juxtacortical according to the 2017 McDonald criteria, and, additionally, as 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 10 mm3 and 100 mm3. 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.