Background: We explore the use of a soft ground-truth mask (“soft mask”) to train a Fully Convolutional Neural Network (FCNN) for segmentation of Multiple Sclerosis (MS) lesions. Detection and segmentation of MS lesions is a complex task largely due to the extreme unbalanced data, with very small number of lesion pixels that can be used for training.
Methods: Utilizing the anatomical knowledge that the lesion surrounding pixels may also include some lesion level information, we suggest to increase the data set of the lesion class with neighboring pixel data - with a reduced confidence weight. A soft mask is constructed by morphological dilation of the binary segmentation mask provided by a given expert, where expert-marked voxels receive label 1 and voxels of the dilated region are assigned a soft label. In the methodology proposed, the FCNN is trained using the soft mask.
Results: On the ISBI 2015 challenge dataset, this is shown to gain a clear improvement in both Dice, precision and recall measures. We also show that by using this soft mask scheme we can improve the network segmentation performance when compared to a second independent expert.
Conclusion: In this study we showed that training the FCNN with the proposed soft mask for the task where the training data is highly unbalanced leads to better model generalization. We demonstrate the propose method on MS lesion segmentation, however this concept is general and can be harnessed to improve other medical image segmentation tasks.