Multi-Scale CNN: An Explainable AI-Integrated Unique Deep Learning Framework for Lung-Affected Disease Classification DOI Creative Commons
Ovi Sarkar, Md. Robiul Islam, Md. Khalid Syfullah

и другие.

Technologies, Год журнала: 2023, Номер 11(5), С. 134 - 134

Опубликована: Сен. 30, 2023

Lung-related diseases continue to be a leading cause of global mortality. Timely and precise diagnosis is crucial save lives, but the availability testing equipment remains challenge, often coupled with issues reliability. Recent research has highlighted potential Chest X-ray (CXR) images in identifying various lung diseases, including COVID-19, fibrosis, pneumonia, more. In this comprehensive study, four publicly accessible datasets have been combined create robust dataset comprising 6650 CXR images, categorized into seven distinct disease groups. To effectively distinguish between normal six different lung-related (namely, bacterial opacity, tuberculosis, viral pneumonia), Deep Learning (DL) architecture called Multi-Scale Convolutional Neural Network (MS-CNN) introduced. The model adapted classify multiple numbers classes, which considered persistent challenge field. While prior studies demonstrated high accuracy binary limited-class scenarios, proposed framework maintains across diverse range conditions. innovative harnesses power combining predictions from feature maps at resolution scales, significantly enhancing classification accuracy. approach aims shorten duration compared state-of-the-art models, offering solution toward expediting medical interventions for patients integrating explainable AI (XAI) prediction capability. results an impressive 96.05%, average values precision, recall, F1-score, AUC 0.97, 0.95, 0.94, respectively, seven-class classification. exhibited exceptional performance multi-class classifications, achieving rates 100%, 99.65%, 99.21%, 98.67%, 97.47% two, three, four, five, six-class respectively. novel not only surpasses many pre-existing (SOTA) methodologies also sets new standard lung-affected using data. Furthermore, integration XAI techniques such as SHAP Grad-CAM enhanced transparency interpretability model’s predictions. findings hold immense promise accelerating improving confidence diagnostic decisions field identification.

Язык: Английский

Multi-Scale CNN: An Explainable AI-Integrated Unique Deep Learning Framework for Lung-Affected Disease Classification DOI Creative Commons
Ovi Sarkar, Md. Robiul Islam, Md. Khalid Syfullah

и другие.

Technologies, Год журнала: 2023, Номер 11(5), С. 134 - 134

Опубликована: Сен. 30, 2023

Lung-related diseases continue to be a leading cause of global mortality. Timely and precise diagnosis is crucial save lives, but the availability testing equipment remains challenge, often coupled with issues reliability. Recent research has highlighted potential Chest X-ray (CXR) images in identifying various lung diseases, including COVID-19, fibrosis, pneumonia, more. In this comprehensive study, four publicly accessible datasets have been combined create robust dataset comprising 6650 CXR images, categorized into seven distinct disease groups. To effectively distinguish between normal six different lung-related (namely, bacterial opacity, tuberculosis, viral pneumonia), Deep Learning (DL) architecture called Multi-Scale Convolutional Neural Network (MS-CNN) introduced. The model adapted classify multiple numbers classes, which considered persistent challenge field. While prior studies demonstrated high accuracy binary limited-class scenarios, proposed framework maintains across diverse range conditions. innovative harnesses power combining predictions from feature maps at resolution scales, significantly enhancing classification accuracy. approach aims shorten duration compared state-of-the-art models, offering solution toward expediting medical interventions for patients integrating explainable AI (XAI) prediction capability. results an impressive 96.05%, average values precision, recall, F1-score, AUC 0.97, 0.95, 0.94, respectively, seven-class classification. exhibited exceptional performance multi-class classifications, achieving rates 100%, 99.65%, 99.21%, 98.67%, 97.47% two, three, four, five, six-class respectively. novel not only surpasses many pre-existing (SOTA) methodologies also sets new standard lung-affected using data. Furthermore, integration XAI techniques such as SHAP Grad-CAM enhanced transparency interpretability model’s predictions. findings hold immense promise accelerating improving confidence diagnostic decisions field identification.

Язык: Английский

Процитировано

20