Penerapan Semi-Supervised Deep Learning dengan Remixmatch untuk Klasifikasi Penyakit Paru-Paru Menggunakan Citra Chest X-Ray

Fajri Fajri (Universitas Islam Negeri Sultan Syarif Kasim, Riau)
Benny Sukma Negara (Universitas Islam Negeri Sultan Syarif Kasim, Riau)
Muhammad Irsyad (Universitas Islam Negeri Sultan Syarif Kasim, Riau)
Febi Yanto (Universitas Islam Negeri Sultan Syarif Kasim, Riau)
Iis Afrianty (Universitas Islam Negeri Sultan Syarif Kasim, Riau)

TIN: TERAPAN INFORMATIKA NUSANTARA
Vol 7 No 2 (2026): July 2026

Lung diseases such as pneumonia and COVID-19 viral infection remain significant health problems that require fast and accurate diagnostic processes. The utilization of deep learning-based Computer-Aided Diagnosis (CAD) on Chest X-Ray (CXR) images has demonstrated promising capabilities in assisting disease classification. However, the implementation of deep learning models in the medical field still faces a major challenge, namely the limited availability of labeled data due to the time-consuming annotation process and the involvement of medical experts. This study applies a semi-supervised learning approach using the ReMixMatch algorithm with DenseNet169 architecture as a feature extraction backbone to reduce the dependency on large amounts of labeled data. Experiments were conducted using the public dataset Covid19-Pneumonia-Normal Chest X-Ray Images available on Mendeley Data. The ReMixMatch method utilizes both labeled and unlabeled data through pseudo-labeling, distribution alignment, MixUp augmentation, and consistency regularization mechanisms during the model training process. The evaluation was performed using several labeled data scenarios, namely 10, 20, 30, and 40 labels per class. The experimental results show that the combination of ReMixMatch and DenseNet169 achieved high classification performance with an accuracy of 96.43% on the validation data. The model evaluation obtained a precision value of 96.47%, a recall value of 96.43%, and an F1-score of 96.42%. These results indicate that the semi-supervised learning approach is able to effectively utilize information from unlabeled data, thereby maintaining high Chest X-Ray image classification performance under limited annotation conditions. This study offers an alternative approach to developing a chest X-ray image classification system through the application of the ReMixMatch algorithm combined with the DenseNet169 architecture, enabling the model to achieve good classification performance even with a limited amount of labeled data.