Evidence record · UDINUS · SDGs
- Unit
- Rektorat UDINUS
- Period
- 2026
- Goals
- Record
- 68515808-0c78-41df-952e-84b24b3aee87
- Published
- 2026-10-30
Universitas Dian Nuswantoro conducted a study to develop and evaluate MK–TripNet, a deep learning framework for real-time multi-class lung sound classification. The activity involved Erini, Widya Surya; Thomas, Gracia Putri; Badia, Giulia Salzano; Rahadian, Arief; Raharjo, Sofyan Budi; and Wulandari, Sari Ayu as participants in the research activity. The study addressed respiratory diseases including asthma, pneumonia, and Chronic Obstructive Pulmonary Disease, which remain major global health challenges, particularly in resource-limited settings where access to pulmonary specialists and early diagnostic tools is limited.
The activity responded to limitations identified in existing automatic lung sound classification methods, including reliance on single-scale feature extraction, conventional loss functions, and offline analysis. UDINUS proposed MK–TripNet as an integrated architecture combining multi-scale feature extraction, discriminative embedding learning, and real-time inference. The framework unified a Multi-Kernel convolutional architecture, Triplet Loss-based embedding learning, and Sliding Window segmentation within an end-to-end approach. This design was intended to capture fine-grained temporal patterns and broader spectral characteristics while maximizing inter-class separability in the embedding space and enabling segment-level classification in real-time scenarios.
The model was evaluated with a newly constructed dataset comprising 1,409 lung sound segments obtained from primary digital stethoscope recordings and publicly available respiratory sound databases. Experimental results showed that MK–TripNet outperformed the CNN-BiGRU, CNN-BiGRU-UMAP, and VGGish-Triplet baseline models, with an accuracy of 89.1%, an F1-score of 0.89, and a recall of 0.88. Ablation studies confirmed that the combined use of Multi-Kernel convolution, Triplet Loss, and Sliding Window segmentation produced the most robust and generalizable performances. The activity relates to health research focused on respiratory screening and digital auscultation. UDINUS frames this work as institutional research addressing real-time respiratory sound classification and its potential application in resource-limited and telemedicine settings.