International Institute of Technology (IIT) – ICTU Students Publish Research Paper in International ICTA Journal
To boost student participation in scientific research and enhance international educational standards, a team of three students from the International Institute of Technology (IIT) at the University of Information and Communication Technology (ICTU) recently had their paper accepted for publication in an international journal supervised by Dr. Vu Duc Quang.
The participating students are:
- Luu The Ha (Class: CNTT QT K22B)
- Nguyen Thanh Trung (Class: CNTT QT K22A)
- Nguyen Tran Anh Hoang (Class: CNTT QT K22B)
Together with their co-authors, the team completed their research titled: “OWMM: Ordinal-Weighted Manifold Mixup with a Tempered Prior for Imbalanced Medical Image Grading.” The paper has been officially accepted and published in the proceedings of the International Conference on Information and Communication Technology and Applications (ICTA).
A Technological Breakthrough Solving Real-World Medical Challenges
Grading the severity of diseases from medical images (such as knee osteoarthritis or diabetic retinopathy) poses a major challenge in artificial intelligence (AI) due to severe data imbalance (where healthy samples vastly outnumber severe cases) and the inherent ordinal nature of disease grades.
The OWMM (Ordinal-Weighted Manifold Mixup) method proposed by the research team offers a groundbreaking solution:
- Smart Pair Selection: Pairs images for blending based on the ordinal distance between severity grades, combined with sample frequency weights.
- Feature Space Mixing (Manifold Mixup): Instead of mixing directly in pixel space (which destroys fine histological details), the method interpolates in the hidden feature space (ResNet layer3) to preserve crucial fine textures in medical images.
- Flexible Prior Regulation: Introduces a tempering factor $\gamma \in [0, 1]$ to seamlessly bridge Cross-Entropy loss and Balanced Softmax, effectively preventing majority-class prediction collapse under severe data skewness.
Impressive Experimental Results
Evaluated on two benchmark international medical imaging datasets—Knee OA (Knee Osteoarthritis) and EyePACS (Diabetic Retinopathy)—OWMM proved its superiority as the only method to outperform all baseline models across three rigorous evaluation metrics: Macro F1, Quadratic Weighted Kappa (QWK), and Mean Absolute Error (MAE).
A Source of Pride for the International Institute of Technology (IIT)
The success of Luu The Ha, Nguyen Thanh Trung, and Nguyen Tran Anh Hoang highlights the high academic standards and effective research orientation of the International Institute of Technology. It also serves as a strong inspiration for all IIT students striving for academic excellence on an international stage.
Congratulations to the team, and we wish them continued success in their academic and research journeys ahead!