Development and Validation of Deep Learning–Based Infectivity Prediction in Pulmonary Tuberculosis Through Chest Radiography: Retrospective Study
2024

AI Tool for Predicting Infectivity in Pulmonary Tuberculosis Using Chest X-rays

Sample size: 4492 publication 10 minutes Evidence: high

Author Information

Author(s): Mavragani Amaryllis, Yoo Sunyong, Wani AasimAyaz, Mao Siqi, Mistry Jinal, Chung Wou young, Yoon Jinsik, Yoon Dukyong, Kim Songsoo, Kim Yujeong, Park Ji Eun, Kang Young Ae

Primary Institution: Yonsei University College of Medicine

Hypothesis

Can an AI-based tool using chest radiography evaluate the infectivity of patients with pulmonary tuberculosis more quickly and accurately than traditional methods?

Conclusion

The AI tool provides a rapid and accurate alternative for assessing PTB infectivity, enhancing screening efficiency in clinical settings.

Supporting Evidence

  • The AI model achieved an accuracy of 73.27% in internal validation.
  • It demonstrated an area under the receiver operating characteristic curve (AUROC) of 0.79.
  • The model's sensitivity was 67.55% and specificity was 78.15% in internal validation.
  • In external validation, the model showed an accuracy of 70.29% and AUROC of 0.77.
  • The AI tool can provide results in seconds, significantly faster than traditional methods.
  • Visualization techniques improved the transparency of the AI model's decision-making process.
  • The study included a large dataset collected over nearly two decades.
  • The model can assist in making quicker clinical decisions regarding patient management.

Takeaway

This study shows that an AI tool can help doctors quickly figure out if someone with tuberculosis is contagious just by looking at their chest X-ray.

Methodology

The study analyzed 36,142 chest X-ray images from 4,492 patients, using deep learning models for infectivity prediction.

Potential Biases

Potential bias due to the study being conducted only on Korean patients.

Limitations

The study only included Korean patients and did not assess long-term patient outcomes after using the AI tool.

Participant Demographics

The cohort included 4,492 patients, with a slight majority being male (60.51%) and an average age of 65 years.

Statistical Information

P-Value

<0.001

Confidence Interval

95% CI for accuracy: 0.72-0.75

Statistical Significance

p<0.05

Digital Object Identifier (DOI)

10.2196/58413

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication