MUSYAFFA, MUHAMMAD RAIHAN ZUFFAR (2026) PERBANDINGAN ALGORITMA YOLOV8S DAN RETINANET RESNET50-FPN UNTUK DETEKSI KARDIOMEGALI PADA CITRA X-RAY THORAKS. S1 thesis, Universitas Mercu Buana Jakarta.
|
Text (HAL COVER)
Cover.pdf Download (840kB) | Preview |
|
![[img]](https://repository.mercubuana.ac.id/style/images/fileicons/text.png) |
Text (BAB I)
Bab 1.pdf Restricted to Registered users only Download (40kB) |
|
|
Text (BAB II)
Bab 2.pdf Restricted to Registered users only Download (321kB) |
|
|
Text (BAB III)
Bab 3.pdf Restricted to Registered users only Download (165kB) |
|
|
Text (BAB IV)
Bab 4.pdf Restricted to Registered users only Download (834kB) |
|
|
Text (BAB V)
Bab 5.pdf Restricted to Registered users only Download (32kB) |
|
|
Text (DAFTAR PUSTAKA)
Daftar Pustaka.pdf Restricted to Registered users only Download (92kB) |
|
|
Text (LAMPIRAN)
Lampiran.pdf Restricted to Registered users only Download (700kB) |
Abstract
Cardiomegaly is a medical condition that can indicate various cardiovascular diseases, making early detection essential for earlier medical intervention, however manual evaluation via Cardiothoracic Ratio (CTR) measurement still relies on radiologists' accuracy and is prone to inter-observer variability. This study aims to design, train, and evaluate two object detection algorithms, YOLOv8s and RetinaNet ResNet50-FPN, to analyze their performance differences in detecting cardiomegaly on chest X-ray images. This study employs a quantitative method with an experimental approach to objectively measure and compare both algorithms' performance based on numerical evaluation metrics. The sample consists of a primary dataset of 1,064 chest X-ray images privately managed on the Roboflow Enterprise platform and annotated by a pulmonary specialist, increased to 4,568 images through CLAHE augmentation, rotation, translation, and scaling, then analyzed using [email protected], [email protected]:0.95, precision, recall, and F1-score metrics. The results show that RetinaNet ResNet50-FPN achieved better overall detection performance compared to YOLOv8s. Statistically, RetinaNet ResNet50-FPN obtained an [email protected] of 0.9371, [email protected]:0.95 of 0.8355, precision of 0.9368, recall of 0.9362, and F1-score of 0.9361, while YOLOv8s obtained an [email protected] of 0.9125, [email protected]:0.95 of 0.8315, precision of 0.9258, recall of 0.9255, and F1- score of 0.9256. The superiority of RetinaNet ResNet50-FPN across all metrics indicates that the combination of feature pyramid network and focal loss more effectively handles object scale variation and foreground-background class imbalance compared to YOLOv8s. Therefore, the RetinaNet ResNet50-FPN-based system developed in this study is expected to serve as a diagnostic support and second opinion tool for medical professionals in analyzing chest X-ray images objectively and efficiently. Keywords: Chest X-ray, cardiomegaly, object detection, YOLOv8s, RetinaNet, deep learning Kardiomegali merupakan kondisi medis yang dapat mengindikasikan berbagai penyakit kardiovaskular, sehingga deteksi dini sangat diperlukan agar intervensi medis dapat diberikan lebih awal, namun evaluasi manual melalui pengukuran Cardiothoracic Ratio (CTR) masih bergantung pada ketelitian dokter radiologi dan rentan terhadap variabilitas hasil diagnosis antar-pengamat. Penelitian ini bertujuan merancang, melatih, dan mengevaluasi dua algoritma object detection, yaitu YOLOv8s dan RetinaNet ResNet50-FPN, guna menganalisis perbedaan kinerja keduanya dalam mendeteksi kardiomegali pada citra X-ray thoraks. Penelitian ini menggunakan metode kuantitatif dengan pendekatan eksperimental untuk mengukur dan membandingkan performa kedua algoritma secara objektif berdasarkan metrik evaluasi numerik. Sampel penelitian berupa dataset primer citra X-ray thoraks sebanyak 1.064 citra yang dikelola secara privat pada platform Roboflow versi Enterprise dan dianotasi oleh dokter spesialis paru-paru, ditingkatkan menjadi 4.568 citra melalui augmentasi CLAHE, rotasi, translasi, dan skala, kemudian dianalisis menggunakan metrik [email protected], [email protected]:0.95, precision, recall, dan F1-score. Hasil penelitian menunjukkan RetinaNet ResNet50-FPN menghasilkan kinerja deteksi kardiomegali yang lebih baik secara keseluruhan dibandingkan YOLOv8s. Secara statistik, RetinaNet ResNet50-FPN memperoleh [email protected] sebesar 0,9371, [email protected]:0.95 sebesar 0,8355, precision sebesar 0,9368, recall sebesar 0,9362, dan F1-score sebesar 0,9361, sedangkan YOLOv8s memperoleh [email protected] sebesar 0,9125, [email protected]:0.95 sebesar 0,8315, precision sebesar 0,9258, recall sebesar 0,9255, dan F1-score sebesar 0,9256. Keunggulan RetinaNet ResNet50-FPN pada seluruh metrik mengindikasikan bahwa kombinasi feature pyramid network dan focal loss lebih efektif menangani variasi ukuran objek dan ketidakseimbangan kelas foreground-background dibandingkan YOLOv8s. Dengan demikian, sistem berbasis RetinaNet ResNet50- FPN ini diharapkan dapat berperan sebagai sistem pendukung diagnosis dan second opinion bagi tenaga medis dalam menganalisis citra X-ray thoraks secara objektif dan efisien. Kata kunci: Citra X-ray thoraks, kardiomegali, object detection, YOLOv8s, RetinaNet, deep learning.
Actions (login required)
 |
View Item |
