FAUDIN, Arif Nur and Farikhin, Farikhin and Syafei, Wahyul Amien (2026) OPTIMALISASI PREDIKSI KARDIOMEGALI MENGGUNAKAN ARSITEKTUR HYBRID YOLOV8-RESNET50 PADA CITRA RONTGEN DADA. Masters thesis, UNIVERSITAS DIPONEGORO.

	Text cover aja.pdf Download (164kB)
	Text Cover.pdf Restricted to Repository staff only Download (4MB)
	Text Bab 1.pdf Download (4MB)
	Text Bab 2.pdf Download (4MB)
	Text Bab 3.pdf Restricted to Repository staff only Download (4MB)
	Text Bab 4.pdf Restricted to Repository staff only Download (4MB)
	Text Bab 5.pdf Restricted to Repository staff only Download (4MB)
	Text Daftar Pustaka.pdf Download (4MB)
	Text Lampiran.pdf Restricted to Repository staff only Download (4MB)

Abstract

Kardiomegali merupakan pembesaran jantung yang menjadi indikator awal gangguan kardiovaskular serius. Penilaian kardiomegali melalui perhitungan Cardiothoracic Ratio (CTR) pada citra rontgen dada memerlukan keahlian khusus dan berpotensi menimbulkan variabilitas antar pemeriksa. Penelitian ini bertujuan mengoptimalkan prediksi kardiomegali melalui arsitektur hybrid YOLOv8-ResNet50 yang dikombinasikan dengan optimalisasi augmentasi data dan hyperparameter tuning. Proses optimalisasi dilakukan dalam tiga tahap: modifikasi backbone YOLOv8 dengan integrasi ResNet50, eksplorasi konfigurasi augmentasi data, dan penyesuaian hyperparameter pelatihan. Dataset terdiri dari 563 citra latih dan 131 citra validasi dari Shenzhen dan Montgomery. Hasil optimalisasi menunjukkan model terbaik mencapai mAP50-95 sebesar 0,7573 (meningkat 19,16% dari baseline tanpa augmentasi), mAP50 0,9922, precision 0,9917, recall 0,9924, dan latensi 1,41 ms/citra. Optimalisasi augmentasi data meningkatkan mAP50-95 hingga 19,7%, sementara hyperparameter tuning memberikan peningkatan tambahan 1,18%. Sistem diintegrasikan ke dalam aplikasi web untuk perhitungan CTR otomatis dan klasifikasi kardiomegali. Temuan menunjukkan optimalisasi arsitektur hybrid, augmentasi data, dan hyperparameter tuning secara kolektif meningkatkan akurasi prediksi kardiomegali dan mengurangi beban kerja radiolog.
Kata Kunci : Optimalisasi, Kardiomegali, YOLOv8-ResNet50, Deep Learning, Augmentasi Data, Hyperparameter Tuning, Citra Rontgen Dada

Cardiomegaly is heart enlargement serving as an early indicator of serious cardiovascular disorders. Cardiomegaly assessment through Cardiothoracic Ratio (CTR) calculation on chest X-rays requires specialized expertise and is susceptible to inter-observer variability. This study aims to optimize cardiomegaly prediction through a hybrid YOLOv8-ResNet50 architecture combined with data augmentation and hyperparameter tuning optimization. The optimization process involves three stages: YOLOv8 backbone modification with ResNet50 integration, data augmentation configuration exploration, and training hyperparameter adjustment. The dataset comprises 563 training and 131 validation images from Shenzhen and Montgomery repositories. Optimization results show the best model achieves mAP50-95 of 0.7573 (19.16% increase from baseline without augmentation), mAP50 of 0.9922, precision of 0.9917, recall of 0.9924, and latency of 1.41 ms/image. Data augmentation optimization increases mAP50-95 by 19.7%, while hyperparameter tuning provides additional 1.18% improvement. The system is integrated into a web application for automatic CTR calculation and cardiomegaly classification. Findings indicate that optimization of hybrid architecture, data augmentation, and hyperparameter tuning collectively enhances cardiomegaly prediction accuracy and reduces radiologist workload.
Keywords : Optimization, Cardiomegaly, YOLOv8-ResNet50, Deep Learning, Data Augmentation, Hyperparameter Tuning, Chest X-ray Images

Actions (login required)

View Item