Evidence record · UDINUS · SDGs
- Unit
- Rektorat UDINUS
- Period
- 2025
- Goals
- Record
- 7c9b4771-f963-44fa-98c2-516a22abec7f
- Published
- 2026-10-30
Universitas Dian Nuswantoro conducted a health research activity on breast cancer detection using medical imaging. The activity involved Kurniawan, Defri; Salam, Abu; Astuti, Yani Parti; Supriyanto, Catur; Shidik, Guruh Fajar; Andono, Pulung Nurtantio; and Safar, Noor Zuraidin Mohd. The work addressed the challenge of detecting breast cancer from a large volume of mammograms and from class imbalance in datasets.
The activity used the RSNA screening mammography breast cancer detection dataset. UDINUS applied a regions of interest, or ROIs, based approach by focusing on specific areas within mammograms. According to the source, this approach reduced the computational load and enabled the model to concentrate on the most critical areas. The activity also applied SMOTE Tomek Link to address class imbalance. This preprocessing generated synthetic samples for the minority cancerous class and removed noisy or overlapping samples.
Three dataset splits were created for evaluation. Split 1 used a 5:1 ratio of normal to cancer cases, while Split 2 used a 3:1 ratio. A fully balanced Random Under-Sampling dataset was also prepared. Various convolutional neural network models were evaluated across the different dataset splits, including InceptionV3, ResNet152V2, DenseNet201, and EfficientNetB7.
The findings reported by UDINUS show that EfficientNetB7, used together with ROI extraction and SMOTE Tomek Link, achieved the highest accuracy reported in the source on the Split 2 dataset. The reported accuracy was 97.41%. The findings identify the combined use of ROI extraction and SMOTE Tomek Link as effective preprocessing techniques within the described deep learning-based breast cancer detection activity. In relation to RAW_GOALS 3, this activity demonstrates research addressing breast cancer detection through medical imaging, dataset preparation, imbalance mitigation, and evaluation of deep learning models. The activity is institutionally framed as UDINUS health research focused on the described breast cancer detection method.