Muscular dystrophies and in general neuromuscular diseases are serious illnesses that mostly affect children at a young age, often with fatal outcomes. The development of effective therapies has also been challenging because of the lack of objective biomarkers for the evaluation of the natural course of the disease and the efficacy of the therapy. While MR imaging is a good candidate to provide such biomarkers, an accurate segmentation of the single muscle groups is an important step for the correct image analysis and evaluation.The segmentation of medical images is a task that has been optimally tackled by deep learning methods in the recent years. However, the accuracy of a deep learning model heavily depends on the type, the quantity and the quality of the data used for the training, and the relative rarity of these diseases often prevents a single site or clinic from having sufficient data for the task, and often from even having the appropriate software tools for an accurate and effective segmentation.In this project, I propose to develop a collaborative segmentation platform that would be accessible to users around the globe, who can use it as a tool to segment their own images, and by doing so, they would contribute to the improvement of the accuracy of the platform itself. Thanks to the principles of federated and incremental learning, the users would not need to share their data, thus preserving privacy and anonymity. The platform will be developed in a modular way, in order to be extensible to multiple clinical questions and input image modalities and contrasts.