Automatic Eye-Tracking-Assisted Chest Radiography Pathology Screening

• Tipo: Artigo em Livro de Atas de Conferência Internacional
• Data: 2023

Autores

Santos, R (Autor) Outra; Pedrosa, J (Autor) Outra; Ana Maria Mendonça (Autor) FEUP; Aurélio Campilho (Autor) FEUP

Ata de Conferência Internacional

Título: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

Páginas: 520-532

11th Iberian Conference on Pattern Recognition and Image Analysis, IbPRIA 2023

Alicante, 27 June 2023 through 30 June 2023

Indexação

Scopus - 0 Citações

Outras Informações

ID Authenticus: P-00Y-K8D

DOI: 10.1007/978-3-031-36616-1_41

Abstract (EN): Chest radiography is increasingly used worldwide to diagnose a series of illnesses targeting the lungs and heart. The high amount of examinations leads to a severe burden on radiologists, which benefit from the introduction of artificial intelligence tools in clinical practice, such as deep learning classification models. Nevertheless, these models are undergoing limited implementation due to the lack of trustworthy explanations that provide insights about their reasoning. In an attempt to increase the level of explainability, the deep learning approaches developed in this work incorporate in their decision process eye-tracking data collected from experts. More specifically, eye-tracking data is used in the form of heatmaps to change the input to the selected classifier, an EfficientNet-b0, and to guide its focus towards relevant parts of the images. Prior to the classification task, UNet-based models are used to perform heatmap reconstruction, making this framework independent of eye-tracking data during inference. The two proposed approaches are applied to all existing public eye-tracking datasets, to our knowledge, regarding chest X-ray screening, namely EGD, REFLACX and CXR-P. For these datasets, the reconstructed heatmaps highlight important anatomical/pathological regions and the area under the curve results are comparable to the state-of-the-art and to the considered baseline. Furthermore, the quality of the explanations derived from the classifier is superior for one of the approaches, which can be attributed to the use of eye-tracking data.

Idioma: Inglês

Tipo (Avaliação Docente): Científica

Nº de páginas: 12