AI Tuberculosis Detection Now Works With Just 75 X-Ray Examples
Could bring fast TB screening to clinics that lack specialists or big datasets.
A new paper shows that deep learning can detect tuberculosis from chest X-rays even when tested on images from a different dataset. The researchers used a DenseNet121 model trained on one TB X-ray source, then adapted it to another using only 75 labeled samples per class. Under full fine-tuning, the model reached 98.36% accuracy, demonstrating that adapting pretrained models is an effective way to handle domain shift. Since TB causes a large number of deaths worldwide each year, this low-resource approach could make accurate detection more practical in real-world clinical settings.
- AI detected tuberculosis in chest X-rays with 98.36% accuracy using only 75 labeled samples per class.
- The method works by fine-tuning a pre-trained model, which adapts to new hospital settings without huge datasets.
- This could make TB screening more affordable and accessible in low-resource clinics that lack radiologists.
Why It Matters
Faster, cheaper TB diagnosis in underserved areas could save over a million lives each year.