Deep Learning in Multi-Class Lung Diseases’ Classification on Chest X-ray Images
Problem
Classifying lung diseases such as pneumonia, pneumothorax and tuberculosis from chest X-rays is slow and error-prone, limiting computer-aided diagnosis.
Why it matters
Misdiagnosed lung diseases delay treatment and can cost lives.
What the AI did
A transfer-learned EfficientNet v2-M deep learning model classifies raw chest X-ray images end-to-end into disease categories.
Results
NIH 3-class dataset: loss 0.6933, accuracy 82.15%, sensitivity 81.40%, specificity 91.65%; SCH 4-class dataset results not stated in source.
Impact
Demonstrated moderate accuracy on public and hospital datasets; clinical adoption not stated in source.
Record
| Venue | Diagnostics |
|---|---|
| Published | 2022 |
| Work type | article |
| Authors | Sungyeup Kim, Beanbonyka Rim, Seongjun Choi, Ahyoung Lee, Se Dong Min, Min Hong |
| Citations | 143 (OpenAlex, at import time) |
| References | 33 |
| Model or method | EfficientNet v2-M (transfer learning) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.3390/diagnostics12040915 |
| Open access | gold |
| Keywords | deep learning, chest X-ray, EfficientNet, pneumonia, pneumothorax, transfer learning |
Importance rating
significant — thousands to millions affected
CXR-based diagnosis of pneumonia, pneumothorax and tuberculosis affects large patient populations worldwide; misdiagnosis delays treatment, though abstract gives no scale and results are moderate.