2014/07/21 by Mathieu Dubois, Dubois, Mathieu, Fouad Hadj‐Selem +11
Computer Science · Engineering · Medicine · #Advanced MRI Techniques and Applications #FOS: Computer and information sciences #Machine Learning (stat.ML) #Medical Image Segmentation Techniques #Radiomics and Machine Learning in Medical Imaging #Sparse and Compressive Sensing Techniques
paper · pdf · doi:10.48550/arxiv.1407.5602
openalex publication_date 2014/07/21 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
The use of machine-learning in neuroimaging offers new perspectives in early\ndiagnosis and prognosis of brain diseases. Although such multivariate methods\ncan capture complex relationships in the data, traditional approaches provide\nirregular (l2 penalty) or scattered (l1 penalty) predictive pattern with a very\nlimited relevance. A penalty like Total Variation (TV) that exploits the\nnatural 3D structure of the images can increase the spatial coherence of the\nweight map. However, TV penalization leads to non-smooth optimization problems\nthat are hard to minimize. We propose an optimization framework that minimizes\nany combination of l1, l2, and TV penalties while preserving the exact l1\npenalty. This algorithm uses Nesterov's smoothing technique to approximate the\nTV penalty with a smooth function such that the loss and the penalties are\nminimized with an exact accelerated proximal gradient algorithm. We propose an\noriginal continuation algorithm that uses successively smaller values of the\nsmoothing parameter to reach a prescribed precision while achieving the best\npossible convergence rate. This algorithm can be used with other losses or\npenalties. The algorithm is applied on a classification problem on the ADNI\ndataset. We observe that the TV penalty does not necessarily improve the\nprediction but provides a major breakthrough in terms of support recovery of\nthe predictive brain regions.\n