Blog

Abstract

Early detection of the risk of lung damage is important in support of timely screening and intervention, while methods such as X-ray or CT require specialized equipment and high cost. This study proposes a deep-based supplementary approach that is to classify the risk of lung damage from iridology imaging. The research process includes pre-processing of images, extracting the area of myopia, training and comparing the four models MobileNetV2, MobileNetV3, Vision Transformer and TinyViT on the same dataset. Experimental results show TinyViT achieves the best performance with Accuracy 91.30%, Recall 95.80%, F1-score 94.48% and 103 ms response time, showing the ability to balance between accuracy and cost calculation. The study also directed the integration of the model into the mobile application to support the initial screening in the community. The proposed system does not replace standard medical diagnostic methods, but serves as a tool to support early warning at low cost and flexible deployment.

Authors:
Ngan Pham Thi Hong Institute of Digital Technology, Thu Dau Mot University Author
Cuong Nguyen Chi Institute of Digital Technology, Thu Dau Mot University Author
Hung Ho Dac Institute of Digital Technology, Thu Dau Mot University Author

DOI: https://doi.org/10.37550/tdmu.EJS/V2026.03.761