DeepMedic is software for 3D image segmention, based on a multi-scale 3D Deep Convolutional Neural Network, from the BioMedIA Group of Imperial College London. The system has been shown to achieve excellent performance on brain lesion segmentation for various tasks, including brain injuries, brain tumors, and ischemic stroke lesions. This method is part of the systems that won the ISLES 2015 and BRATS 2017 competitions.
A complete set of versatile and easy to use tools to train, evaluate, and make use of 3D image segmentation models in a simple and intuitive manner for non-technical users.
The framework includes automated data harmonisation, and informative dataset metadata checks, made easily and intuitively customisable via its Graphical User Interface.
Graphical User Interface enables easy training and testing of Deep Learning models.
Easy to customize and tune for your data and needs to achieve higher performance levels.
Distributing pre-trained segmentation models for common medical tasks.
DeepMedic was successfully adopted in the clinical workflow [for segmenting organs-at-risk in prostate radiotherapy] maintaining in the clinical setting the accuracy obtained in the feasibility study.
no statistically significant differences were found between the performance metrics [between DeepMedic and human reader for stroke lesion segmentation in DW MRI]
the system [DeepMedic] learned to outperform its “human teachers” and produced output that was better, on average, than its training data.
the Dice score of the algorithm results were comparable with the Dice scores between human readers in our subcohort [for stroke lesion segmentation in DW MRI]
a whole training process within a single day utilizing a NVIDIA GeForce GTX 580 graphics board which makes this library also very interesting for research purposes on low-end GPU hardware.
Various independent studies found that DeepMedic out-performed other state-of-the-art deep-learning models in different tasks, such as for liver segmentation in CT or brain vessel segmentation from cerebrovascular images.
DeepMedic is hosted on GitHub, where the codebase is kept, maintained, and updated.