DETECTX: a novel deep learning approach for detecting pneumonia, tuberculosis, and aspergillosis from chest X-rays.
Authors
Affiliations (8)
Affiliations (8)
- Department of Internal Medicine and Pediatrics, Medical School, Henan Provincial Research Center of Engineering Technology for Nuclear Protein Medical Detection, Zhengzhou Health College, Zhengzhou, Henan, China.
- School of Software, Nanjing University of Information Science and Technology, Nanjing, China.
- College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, Nanjing, China.
- Department of Pulmonary and Critical Care Medicine, People's Hospital of Fangcheng County, Nanyang, Henan, China.
- Department of Microbiology and Immunology, Medical School; Henan Provincial Research Center of Engineering and Technology for Medical Detection of Nuclear Protein, Zhengzhou Health College, Zhengzhou, Henan, China.
- Kaifeng Key Laboratory for Infection and Biosafety, School of Basic Medical Sciences, Henan University, Kaifeng, Henan, China.
- Department of Health Management, ZhengZhou Health College, Zhengzhou, Henan, China.
- Institute of Translational Medicine, Medical College, Yangzhou University, Yangzhou, Jiangsu, China.
Abstract
Aspergillosis, pneumonia, and tuberculosis are distinct respiratory illnesses, but they share similar symptoms, including cough, fever, and shortness of breath. This overlap can make accurate diagnosis challenging. Delayed or incorrect identification, particularly of aspergillosis, can lead to worsening patient outcomes and significant lung damage. To address this diagnostic challenge, we developed a deep learning based system designed to differentiate between aspergillosis, pneumonia, tuberculosis, and healthy lungs using chest X-ray images. We implemented three models: Convolutional Neural Networks (CNN), ResNet-50, and a novel Siamese Convolutional Neural Network. The models were trained on a dataset of 7,200 images, categorized into four groups: aspergillosis, pneumonia, tuberculosis, and healthy. We utilized various training-testing ratios to optimize performance, which was evaluated using accuracy, precision, recall, F1-score, and confusion matrices. Among the models tested, the Siamese neural network outperformed the others, achieving an accuracy of 98.72%, surpassing the performance of state-of-the-art models in comparative studies. The novel Siamese CNN model demonstrated excellent accuracy in differentiating respiratory diseases from chest X-ray images, offering a promising tool for more accurate and timely diagnoses. We also developed an intuitive web interface that enables healthcare providers to upload and analyze medical images, facilitating quicker and more reliable diagnoses to support timely interventions.