Improving Osteosarcoma Detection through SMOTE-Driven Machine Learning Approaches

Muhammad Ainul, Fikri and Ajie Kusuma, Wardhana and Yudha, Riwanto and Inggrid Yanuar Risca, Partiwi and Fauzia Anis Sekar, Ningrum and Iqbal Kurniawan Asmar, Putra (2025) Improving Osteosarcoma Detection through SMOTE-Driven Machine Learning Approaches. IJID (International Journal on Informatics for Development), 13 (2). pp. 517-529. ISSN 2549-7448

	Text rev-ijid-paper-format-ver.23-kanker-tulang-revisedfeb25_2.pdf Download (696kB)
	Text Sertifikat Turnitin 2024 - Muhammad Ainul Fikri - 1 September 2025.pdf Download (283kB)
	Text Improving Osteosarcoma Detection through SMOTE-Driven Machine Learning Approaches.pdf Download (826kB)

Abstract

Osteosarcoma is an aggressive and highly malignant bone cancer that predominantly affects adolescents and young adults, with males being more commonly affected. Early detection is crucial for improving treatment outcomes and survival rates. While deep learning models such as YOLO, which achieves an accuracy of 95.73%, and VGG19, with an accuracy of 95.25%, have demonstrated effectiveness in osteosarcoma detection, their large model sizes, reliance on extensive computational resources, and susceptibility to dataset imbalance limit their feasibility in resource-constrained environments. To address these challenges, this study proposes a lightweight AI approach that optimizes osteosarcoma detection while maintaining high diagnostic accuracy. By leveraging machine learning models under 5MB, manual or semi-automatically extracted features, and SMOTE for data balancing, the proposed method reduces computational demands and improves efficiency. Experimental results show that Random Forest, SVM, and XGBoost achieve accuracies of 94.70%, 94.23%, and 94.39%, respectively, closely matching the performance of YOLO and VGG19 while maintaining computational efficiency. Furthermore, the inference time for SVM is under one second (0.97s), demonstrating the speed advantage of lightweight models. These findings highlight the potential of small-size (lightweight) machine learning models to deliver high diagnostic accuracy with minimal computational requirements, providing a scalable and practical solution for early osteosarcoma detection in resource-limited settings. By balancing simplicity, efficiency, and high performance, this study establishes a new benchmark for achieving state-of-the-art results with lightweight models, paving the way for improved healthcare accessibility in underserved regions.

Actions (login required)

View Item