JMIR Publications

ABSTRACT

Background:

Pulmonary tuberculosis (PTB) poses a global health challenge as obtaining results from traditional smear and culture tests require a few hours to weeks.

Objective:

We aimed to assess infectivity using AI-based chest radiography (CXR) alone with comparable performance to traditional tests for PTB.

Methods:

By employing DenseNet121 and visualisation techniques such as gradient-weighted class activation mapping (Grad-CAM++) and local interpretable model-agnostic explanations (LIME) to illustrate model's decision-making process, we analysed 36,142 CXR images of 4,492 patients with PTB obtained from Severance Hospital, focusing specifically on the lung region through segmentation and cropping with TransUNet. We used data from 2004–2021 to build the model, and data from 2022–2023 for internal validation. Additionally, we used 1,978 CXR images of 299 patients with PTB obtained from Yongin Severance Hospital for external validation.

Results:

: In the internal validation, the model achieved an accuracy of 73.27%, AUROC of 0.7917, and AUPRC of 0.7716. In the external validation, it demonstrated an accuracy of 70.29%, AUROC of 0.7686, and AUPRC of 0.7970. Additionally, Grad-CAM++ and LIME provide insights into the decision-making process of the AI model.

Conclusions:

This proposed AI tool can evaluate PTB infectivity from CXR images and potentially offer more accurate and faster screening and results than traditional smear and culture tests.