AI Matches Radiologists in Detecting TB From Photographed Chest X-Rays: Study Shows
Ethiopia: Researchers from Ethiopia have demonstrated that computer-aided analysis of photographed chest X-ray films can perform on par with trained radiologists in detecting tuberculosis (TB), offering a promising diagnostic solution for low-resource, high-burden settings. The pilot study, published in Mayo Clinic Proceedings: Digital Health, was led by Zerubabel Desita from the University of Gondar, Ethiopia, and colleagues.
Tuberculosis remains a major global health challenge, with more than 10 million people affected worldwide and over one million deaths reported in 2023. Early detection is critical for controlling transmission, yet many low-income regions face a shortage of trained radiologists and lack access to digital radiography. In such settings, chest X-rays are often available only in analog film format, limiting the applicability of existing artificial intelligence (AI) tools that rely on digital images.
To address this gap, the researchers evaluated whether film chest X-rays photographed using mobile phones or digital cameras could be reliably analyzed using an AI-based computer-aided detection (CAD) system. The study assessed the diagnostic performance of a commercially available AI model (qXR, Qure.ai) when applied to photographed chest X-ray films, and compared its performance with that of experienced radiologists.
The retrospective pilot study included 498 chest X-ray films from patients presenting with symptoms suggestive of pulmonary TB between January 2017 and March 2018 in Ethiopia and Guinea-Bissau. Each film X-ray was photographed using a digital camera or mobile phone to create an image suitable for AI analysis. Final TB diagnosis was established using clinical assessment and laboratory confirmation, including Xpert MTB/RIF testing, which served as the reference standard. Two experienced Ethiopian radiologists independently reviewed the X-ray images.
The following findings were reported:
- One radiologist identified 50 chest X-ray films as indicative of tuberculosis, while the second radiologist identified 99 cases.
- The AI-based computer-aided detection model flagged 81 cases as suggestive of tuberculosis.
- When compared with laboratory-confirmed diagnoses, the AI system achieved an area under the receiver operating characteristic curve of 0.84, reflecting good overall diagnostic performance.
- At the predefined cutoff, the AI model demonstrated a sensitivity of 76.5% and a specificity of 85.9%.
- Radiologist A showed lower sensitivity (64.7%) but higher specificity (91.9%) compared with the AI model.
- Radiologist B demonstrated the same sensitivity as the AI model (76.5%) with slightly lower specificity (82.3%).
- Agreement between the two radiologists was moderate.
- The agreement between each radiologist and the AI software was also moderate.
The authors reported that the AI system performed comparably to experienced radiologists, even when applied to photographs of analog chest X-ray films rather than native digital images. They emphasized that this strategy could significantly lower diagnostic barriers in settings lacking digital radiography infrastructure.
Although the study was a pilot and requires further validation, the researchers underscored its implications for tuberculosis control. With global TB reduction targets unlikely to be achieved under current conditions, AI-assisted analysis of photographed chest X-rays may help close diagnostic gaps. Such tools could also support healthcare workers with limited radiology training in TB screening, improving early detection and access to care in underserved regions.
Reference:
Desita, Z., Tadesse, T., Bohlbro, A. S., Sifna, A., Fekadu, H., Bizuneh, S., Sridhar, S., Robert, D., Tadepalli, M., Wejse, C., Schön, T., & Rudolf, F. (2026). Computer-Aided Analysis of Photographed Chest X-Ray Films Performs Well Compared to Trained Radiologists. Mayo Clinic Proceedings: Digital Health, 100338. https://doi.org/10.1016/j.mcpdig.2026.100338
Next Story