| dc.description.abstract | Emphysema, as a form of Chronic Obstructive Pulmonary Disease (COPD), is
characterized by damage to the alveoli that leads to breathing difficulties and a reduced
quality of life. Conventional diagnosis using X-ray images often suffers from low
sensitivity and inter-observer variability, highlighting the need for a more reliable
diagnostic method. This study proposes a method for the automatic quantification of
emphysema lesions in X-ray images using the Mask Region-Convolutional Neural
Network (Mask R-CNN) with a ResNet50 backbone. The system is designed to perform
lung segmentation in the initial stage, followed by emphysema lesion segmentation to
calculate the percentage of Low Attenuation Areas (LAA%), which serves as the basis
for determining the severity of emphysema (Minimal, Mild, Moderate). A total of 800
data samples were used, consisting of 560 for training, 200 for validation, and 40 for
testing. The lung segmentation model achieved a mean Average Precision (mAP) of
0.9900, Intersection over Union (IoU) of 0.8734, and Area Under the Curve (AUC) of
0,9968. Meanwhile, the emphysema lesion segmentation model achieved an mAP of
0.9400, IoU of 0.8310, and AUC of 0.8563. The optimal hyperparameter configuration
for both models was obtained with a batch size of 1, learning rate of 0.001, and weight
decay of 0.0001. | en_US |
