Chest X-Ray Analysis of Tuberculosis by Deep Learning with Segmentation and Augmentation
Problem
Detecting tuberculosis from chest X-ray images via computer-aided diagnosis, particularly when the available dataset is small and class-imbalanced.
Why it matters
Reliable computer-aided TB detection from chest X-rays could support diagnosis where training data is scarce and unbalanced.
What the AI did
A deep convolutional neural network was trained on lung-segmented and augmented chest X-ray images to predict tuberculosis; segmentation enabled training where the unsegmented dataset did not.
Results
Lossless augmentation of the segmented dataset gave the lowest validation loss without overfitting, with accuracy within one standard deviation of alternatives; lossy augmentation lowered validation loss but reduced validation accuracy.
Impact
Shows preprocessing and augmentation enable CNN training on small, imbalanced chest X-ray datasets.
Record
| Venue | IEEE International Conference on Electronics and Nanotechnology (ELNANO) |
|---|---|
| Published | 2018 |
| Work type | conference-paper |
| Authors | Sergii Stirenko, Yuriy Kochura, Oleg Alienin, Oleksandr Rokovyi, Yuri G. Gordienko, Peng Gang, Wei Zeng |
| Citations | 109 (OpenAlex, at import time) |
| References | 15 |
| Model or method | deep convolutional neural network (CNN) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1109/elnano.2018.8477564 |
| Open access | green |
| Keywords | tuberculosis, chest X-ray, CNN, lung segmentation, data augmentation, CADx |
Importance rating
significant — thousands to millions affected
Tuberculosis diagnosis via chest X-ray CADx could support diagnosis where data is scarce; a significant diagnostic problem, though the abstract gives no disease burden numbers and focuses on preprocessing methods.