2019/01/17 by Mostafa Salem, Sergi Valverde, Salem, Mostafa +13 · 2 citations
Biochemistry, Genetics and Molecular Biology · Computer Science · Engineering · #AI in cancer detection #Cell Image Analysis Techniques #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Image Processing Techniques and Applications
paper · pdf · doi:10.48550/arxiv.1901.05733
openalex publication_date 2019/01/17 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
In this paper, we propose generating synthetic multiple sclerosis (MS)\nlesions on MRI images with the final aim to improve the performance of\nsupervised machine learning algorithms, therefore avoiding the problem of the\nlack of available ground truth. We propose a two-input two-output fully\nconvolutional neural network model for MS lesion synthesis in MRI images. The\nlesion information is encoded as discrete binary intensity level masks passed\nto the model and stacked with the input images. The model is trained end-to-end\nwithout the need for manually annotating the lesions in the training set. We\nthen perform the generation of synthetic lesions on healthy images via\nregistration of patient images, which are subsequently used for data\naugmentation to increase the performance for supervised MS lesion detection\nalgorithms. Our pipeline is evaluated on MS patient data from an in-house\nclinical dataset and the public ISBI2015 challenge dataset. The evaluation is\nbased on measuring the similarities between the real and the synthetic images\nas well as in terms of lesion detection performance by segmenting both the\noriginal and synthetic images individually using a state-of-the-art\nsegmentation framework. We also demonstrate the usage of synthetic MS lesions\ngenerated on healthy images as data augmentation. We analyze a scenario of\nlimited training data (one-image training) to demonstrate the effect of the\ndata augmentation on both datasets. Our results significantly show the\neffectiveness of the usage of synthetic MS lesion images. For the ISBI2015\nchallenge, our one-image model trained using only a single image plus the\nsynthetic data augmentation strategy showed a performance similar to that of\nother CNN methods that were fully trained using the entire training set,\nyielding a comparable human expert rater performance\n