Testing the robustness of attribution methods for convolutional neural\n networks in MRI-based Alzheimer's disease classification

2019/09/19 by Fabian Eitel, Kerstin Ritter, Eitel, Fabian +1 · 1 citation
Neuroscience · Engineering · Computer Science · #Brain Tumor Detection and Classification #Medical Imaging and Analysis #Machine Learning in Healthcare

paper · pdf · doi:10.48550/arxiv.1909.08856

Abstract

Attribution methods are an easy to use tool for investigating and validating\nmachine learning models. Multiple methods have been suggested in the literature\nand it is not yet clear which method is most suitable for a given task. In this\nstudy, we tested the robustness of four attribution methods, namely\ngradient*input, guided backpropagation, layer-wise relevance propagation and\nocclusion, for the task of Alzheimer's disease classification. We have\nrepeatedly trained a convolutional neural network (CNN) with identical training\nsettings in order to separate structural MRI data of patients with Alzheimer's\ndisease and healthy controls. Afterwards, we produced attribution maps for each\nsubject in the test data and quantitatively compared them across models and\nattribution methods. We show that visual comparison is not sufficient and that\nsome widely used attribution methods produce highly inconsistent outcomes.\n

Cited by

• Latent space projections and atlases: A cautionary tale in deep neuroimaging using autoencoders

Related

• Convolutional neural networks for classification of Alzheimer's disease: Overview and reproducible evaluation
• Comparison of Convolutional neural network training parameters for detecting Alzheimers disease and effect on visualization
• Detection of Alzheimers Disease from MRI using Convolutional Neural\n Networks, Exploring Transfer Learning And BellCNN
• A Comprehensive Study of Alzheimer's Disease Classification Using Convolutional Neural Networks
• Alzheimer's Disease Brain MRI Classification: Challenges and Insights