SA-ResNet50: A Four-Class Classification Model for Lung Diseases in Chest X-rays
Year: 2025
Type: article
Abstract: Deep learning in medical image analysis is becoming more important for diagnosing disease, especially when it comes to spotting stuff like COVID-19 and pneumonia in CXR pictures of someone's chest. But it is inevitable that there will be some limitations for CNN in terms of CXR images because it always has blurred and unclear edges, strange shape and low contrast with other tissues around the lesion area. And these usually cause wrong localizations and some important informations will be lost when extracting features. So it is hard for classification models to do well. In order to solve above problems, I propose the SA-ResNet50 architecture in which a spatial attention block is introduced inside residual bottleneck blocks of ResNet-50. This work contributes in some ways. First is to improve classification accuracy by improving the localization and representation of important lesion area. Second, in a complete pneumonia data set, the Accuracy is 0.9847, the Precision is 0.9826, the Recall is 0.9827, the F1-Score is 0.9826, and the Specificity is 0.9939. By doing full-comparison experiment about every baseline model and detailed ablation studies, I will strictly demonstrate it is effective and better than others. The experimental results not only validate the efficacy of this model, but also propose new ways and references to deal with similar issues in other fields related to recognizing pictures of medicine.
Author: Qi Deng
Institution: Southeast University
Language: English
Topic: COVID-19 diagnosis using AI
Subfield: Radiology, Nuclear Medicine and Imaging
Field: Medicine
Domain: Health Sciences
Open Access status: hybrid
OpenAlex ID: W4414478702
FWCI:
Cites:
Cited by:
Related to: