Development and application of a deep learning-based tuberculosis diagnostic assistance system in remote areas of Northwest China.
Authors
Affiliations (5)
Affiliations (5)
- Xinjiang Key Laboratory of Artificial Intelligence Assisted Imaging Diagnose, The First People's Hospital of Kashi Prefecture, Kashi, China.
- Department of Scientific Research, Xinjiang Academy of Medical Sciences, Urumqi, China.
- Xinjiang Key Laboratory of Artificial Intelligence Assisted Imaging Diagnose, The First People's Hospital of Kashi Prefecture, Kashi, China. [email protected].
- Department of Scientific Research, Xinjiang Academy of Medical Sciences, Urumqi, China. [email protected].
- Department of Radiology, The First People's Hospital of Kashi Prefecture, Kashi, China. [email protected].
Abstract
The Kashgar region, located in Northwest China, has a significantly higher incidence of tuberculosis (TB) compared to the national average. Local governments conduct annual TB screening using medical imaging. However, due to a shortage of radiologists, insufficient diagnostic capabilities, low levels of informatization, and reliance on manual processes in primary healthcare institutions, traditional medical imaging methods for TB screening suffer from low efficiency, high rates of misdiagnosis, and missed diagnoses. To develop a deep learning-based TB diagnostic assistance system tailored to local conditions, addressing the shortcomings of primary healthcare institutions, improving TB screening efficiency, and reducing misdiagnosis and missed diagnosis rates. We collected chest X-ray images from 10,897 patients across multiple centers, with 10,002 cases used for training and 895 cases for testing. We trained a TB-UNET model and developed a TB diagnostic assistance system based on this model, deploying it in 12 counties and 178 township hospitals in the Kashgar region. The system enhanced the informatization of primary healthcare institutions. In clinical testing, Radiologists' sensitivity in diagnosing TB increased by 11.8% (from 62.7% [95% CI 58.2-67.1%] to 74.5% [95% CI 70.3-78.4%], p < 0.001), accuracy improved by 2.8% (from 87.2% [95% CI 85.1-89.1%] to 90.0% [95% CI 88.2-91.6%], p = 0.023), and the average time spent on reading images decreased from 38.83 s (95% CI 36.21-41.45 s) to 15.93 s (95% CI 14.32-17.54 s, p < 0.001). The system significantly enhanced TB screening efficiency in the Kashgar region, reducing misdiagnosis and missed diagnosis rates. It has high practical value and offers a replicable model for screening other infectious diseases in remote areas.