Deteksi Daerah Cardiomegaly pada Citra Chest X-Ray Berbasis Object Detection dan Deep Learning

Indriani, Ratna (2025) Deteksi Daerah Cardiomegaly pada Citra Chest X-Ray Berbasis Object Detection dan Deep Learning. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5023211064-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (1MB) | Request a copy

Abstract

Cardiomegaly merupakan pembesaran struktural jantung sering kali menandakan adanya gangguan kardiovaskular seperti cardiomyopathy, yang mengurangi kemampuan jantung dalam memompa darah. Pemeriksaan Chest X-ray (CXR) rutin penting untuk mendeteksi kondisi ini, karena mampu menunjukkan perubahan ukuran dan bentuk jantung secara jelas. Namun, meskipun CXR merupakan alat diagnostik yang umum digunakan dan mudah diakses oleh ahli radiologi, proses evaluasinya sering kali memakan waktu dan berpotensi meningkatkan risiko kesalahan diagnosis. Selain itu, citra mentah yang dihasilkan oleh X-ray sering kali memiliki kualitas yang kurang optimal akibat adanya noise digital dan kontras yang rendah, yang dapat memperumit proses deteksi tepi objek dalam gambar. Pada penelitian terdahulu masih menggunakan pendekatan tradisional seperti selective search sering memakan waktu untuk menghasilkan region proposals. Oleh karena itu, penelitian ini mengembangkan sistem deteksi otomatis menggunakan metode Faster R-CNN untuk mempercepat dan meningkatkan akurasi deteksi cardiomegaly pada citra CXR. Tahap preprocessing yang dilakukan meliputi pengubahan ukuran citra menjadi 512×512 piksel, peningkatan kontras menggunakan CLAHE, serta intensity thresholding untuk meningkatkan kejelasan objek. Ekstraksi fitur dilakukan menggunakan arsitektur backbone VGG-16 yang telah disesuaikan dengan citra grayscale dan menggunakan bobot pralatih dari ImageNet. Proposal wilayah objek dihasilkan menggunakan Region Proposal Network (RPN) dengan anchor box adaptif dari hasil Mean Shift Clustering, yang kemudian diproses lebih lanjut oleh ROI Align dan jaringan fully connected untuk klasifikasi dan regresi. Model dilatih menggunakan kombinasi citra cardiomegaly dan normal selama 30 epoch. Hasil pelatihan menunjukkan penurunan total loss yang signifikan dari 0,40 menjadi 0,0966. Evaluasi terhadap sistem menunjukkan akurasi sebesar 88,5%, precision 86,67%, recall 91,0%, dan F1-score 88,78%. Selain itu, hasil perhitungan Cardiothoracic Ratio (CTR) pada salah satu sampel citra menghasilkan nilai sebesar 0,576, yang mengindikasikan adanya kondisi cardiomegaly.
==================================================================================================================================
Cardiomegaly, a structural enlargement of the heart, often indicates cardiovascular disorders such as cardiomyopathy, which reduces the heart's ability to pump blood. Routine chest X-ray (CXR) examinations are essential for detecting this condition, as they clearly demonstrate changes in the heart's size and shape. However, although CXR is a commonly used diagnostic tool and is readily accessible to radiologists, the evaluation process is often time-consuming and potentially increases the risk of misdiagnosis. Furthermore, raw X-ray images often have suboptimal quality due to digital noise and low contrast, which can complicate the process of detecting object edges in the image. Previous studies have used traditional approaches such as selective search, which often take time to generate region proposals. Therefore, this study develops an automatic detection system using the Faster R-CNN method to accelerate and improve the accuracy of cardiomegaly detection in CXR images. The preprocessing stage includes resizing the image to 512x512 pixels, enhancing contrast using CLAHE, and intensity thresholding to improve object clarity. Feature extraction was performed using the VGG-16 backbone architecture adapted to grayscale images and using pretrained weights from ImageNet. Object region proposals were generated using a Region Proposal Network (RPN) with adaptive anchor boxes from Mean Shift Clustering results, which were then further processed by ROI Align and a fully connected network for classification and regression. The model was trained using a combination of cardiomegaly and normal images for 30 epochs. The training results showed a significant decrease in total loss from 0.40 to 0.0966. Evaluation of the system showed an accuracy of 88.5%, a precision of 86.67%, a recall of 91.0%, and an F1-score of 88.78%. In addition, the Cardiothoracic Ratio (CTR) calculation on one of the image samples produced a value of 0.576, which indicates the presence of cardiomegaly

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Cardiomegaly, Chest X-ray (CXR), Faster R-CNN Deteksi otomatis, Convolutional Neural Network, VGG-16, ROI; Cardiomegaly, Chest X-ray (CXR), Faster R-CNN Automatic Detection, Convolutional Neural Network, VGG-16, ROI
Subjects:	R Medicine > R Medicine (General) > R858 Deep Learning
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Biomedical Engineering > 11410-(S1) Undergraduate Thesis
Depositing User:	Ratna Indriani
Date Deposited:	05 Aug 2025 03:46
Last Modified:	11 Aug 2025 03:01
URI:	http://repository.its.ac.id/id/eprint/126387

Actions (login required)

View Item