RT info:eu-repo/semantics/article
T1 In-depth evaluation of saliency maps for interpreting convolutional neural network decisions in the diagnosis of glaucoma based on fundus imaging
A1 Sigut Saavedra, José Francisco
A1 Fumero Batista, Francisco José
A1 Estévez Damas, José Ignacio
A1 Alayón Miranda, Silvia
A1 Díaz Alemán, Tinguaro
K1 Saliency methods
K1 Glaucoma diagnosis
K1 Convolutional neural networks
K1 Deep learning
K1 Retinal fundus images
AB Glaucoma, a leading cause of blindness, damages the optic nerve, making early diagnosischallenging due to no initial symptoms. Fundus eye images taken with a non-mydriatic retinographhelp diagnose glaucoma by revealing structural changes, including the optic disc and cup. Thisresearch aims to thoroughly analyze saliency maps in interpreting convolutional neural networkdecisions for diagnosing glaucoma from fundus images. These maps highlight the most influentialimage regions guiding the network’s decisions. Various network architectures were trained andtested on 739 optic nerve head images, with nine saliency methods used. Some other populardatasets were also used for further validation. The results reveal disparities among saliency maps,with some consensus between the folds corresponding to the same architecture. Concerning thesignificance of optic disc sectors, there is generally a lack of agreement with standard medical criteria.The background, nasal, and temporal sectors emerge as particularly influential for neural networkdecisions, showing a likelihood of being the most relevant ranging from 14.55% to 28.16% on averageacross all evaluated datasets. We can conclude that saliency maps are usually difficult to interpretand even the areas indicated as the most relevant can be very unintuitive. Therefore, its usefulnessas an explanatory tool may be compromised, at least in problems such as the one addressed inthis study, where the features defining the model prediction are generally not consistently reflectedin relevant regions of the saliency maps, and they even cannot always be related to those used asmedical standards.
SN 1424-8220
YR 2023
FD 2023
LK http://riull.ull.es/xmlui/handle/915/35148
UL http://riull.ull.es/xmlui/handle/915/35148
LA en
NO https://doi.org/10.3390/s24010239
DS Repositorio institucional de la Universidad de La Laguna
RD 30-nov-2024