**LST-AI: A Deep Learning Ensemble for Accurate MS Lesion Segmentation** **Keywords:** Multiple Sclerosis, Artificial Intelligence, Lesion Segmentation, Magnetic Resonance Imaging, White Matter Lesions, Deep Learning **Abstract:** Automated segmentation of brain white matter lesions is crucial for clinical assessment and scientific research in multiple sclerosis (MS). Over a decade ago, we introduced LST, an engineered lesion segmentation tool. This paper presents LST-AI, an advanced deep learning-based extension of LST, consisting of an ensemble of three 3D U-Nets. LST-AI addresses the imbalance between white matter (WM) lesions and non-lesioned WM using a composite loss function that incorporates binary cross-entropy and Tversky loss. The network ensemble is trained on 491 MS pairs of T1-weighted and FLAIR images, 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, juxtacortical, and subcortical according to the 2017 McDonald criteria. Evaluations on 103 test cases using the Anima segmentation validation tools show that LST-AI achieves superior performance compared to existing methods, with Dice and F1 scores exceeding 0.62, outperforming LST, SAMESG, and the popular nnUNET framework. Notably, LST-AI demonstrated exceptional performance on the MSSEG-1 challenge dataset, achieving a Dice score of 0.65 and an F1 score of 0.63, surpassing all other competing models at the time of the challenge. **Introduction:** Multiple sclerosis (MS) is a complex chronic inflammatory disease of the central nervous system. Automated lesion segmentation is crucial for clinical assessment and research. While manual segmentation by trained neuroradiologists is the gold standard, it is time-consuming and prone to inter- and intra-rater variability. Recent advancements in artificial intelligence (AI) have led to the development of automated lesion segmentation tools based on convolutional neural networks (CNNs), which often outperform earlier machine learning-based methods. LST-AI is an open-source tool that provides an ensemble of three 3D U-Nets, trained on an in-house dataset of 491 paired T1-weighted and FLAIR images. The tool is validated on multiple external datasets, demonstrating superior performance compared to existing methods, including LST, SAMESG, and nnUNET. LST-AI also includes a lesion location annotation feature, which is crucial for MS research and clinical applications. **Methods:** LST-AI is evaluated using multiple datasets, including the ISBI 2015, msljub, mssegtest, and mssegment datasets. Preprocessing steps include registration to the MNI ICBM**LST-AI: A Deep Learning Ensemble for Accurate MS Lesion Segmentation** **Keywords:** Multiple Sclerosis, Artificial Intelligence, Lesion Segmentation, Magnetic Resonance Imaging, White Matter Lesions, Deep Learning **Abstract:** Automated segmentation of brain white matter lesions is crucial for clinical assessment and scientific research in multiple sclerosis (MS). Over a decade ago, we introduced LST, an engineered lesion segmentation tool. This paper presents LST-AI, an advanced deep learning-based extension of LST, consisting of an ensemble of three 3D U-Nets. LST-AI addresses the imbalance between white matter (WM) lesions and non-lesioned WM using a composite loss function that incorporates binary cross-entropy and Tversky loss. The network ensemble is trained on 491 MS pairs of T1-weighted and FLAIR images, 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, juxtacortical, and subcortical according to the 2017 McDonald criteria. Evaluations on 103 test cases using the Anima segmentation validation tools show that LST-AI achieves superior performance compared to existing methods, with Dice and F1 scores exceeding 0.62, outperforming LST, SAMESG, and the popular nnUNET framework. Notably, LST-AI demonstrated exceptional performance on the MSSEG-1 challenge dataset, achieving a Dice score of 0.65 and an F1 score of 0.63, surpassing all other competing models at the time of the challenge. **Introduction:** Multiple sclerosis (MS) is a complex chronic inflammatory disease of the central nervous system. Automated lesion segmentation is crucial for clinical assessment and research. While manual segmentation by trained neuroradiologists is the gold standard, it is time-consuming and prone to inter- and intra-rater variability. Recent advancements in artificial intelligence (AI) have led to the development of automated lesion segmentation tools based on convolutional neural networks (CNNs), which often outperform earlier machine learning-based methods. LST-AI is an open-source tool that provides an ensemble of three 3D U-Nets, trained on an in-house dataset of 491 paired T1-weighted and FLAIR images. The tool is validated on multiple external datasets, demonstrating superior performance compared to existing methods, including LST, SAMESG, and nnUNET. LST-AI also includes a lesion location annotation feature, which is crucial for MS research and clinical applications. **Methods:** LST-AI is evaluated using multiple datasets, including the ISBI 2015, msljub, mssegtest, and mssegment datasets. Preprocessing steps include registration to the MNI ICBM