A Comparative Performance of U-net and Mask R-CNN for Lung Segmentation Across Public Chest X-ray Datasets

Lung segmentation is critical for detecting and monitoring respiratory conditions and abnormalities in the lungs, making it a useful diagnostic tool. This study compares the performance of two deep learning models, U-net convolutional neural network (U-net) and Mask Region-based Convolutional Neural Network (R-CNN), for segmenting lungs using publicly available datasets. The U-net model was trained and validated on the Montgomery dataset, while the Mask R-CNN model was evaluated after being pre-trained without fine-tuning. Both models were tested on both the Montgomery and Shenzhen datasets to assess generalizability. Mask R-CNN was found to have the best performance with a Dice Coefficient of 0.9302 and IoU of 0.8696 on the Shenzhen dataset. Although Mask R-CNN showed stronger performance on the unseen Shenzhen dataset, the comparison is limited since the two models are trained on different datasets. This study highlights strengths and limitations of each model and outlines future work to make the study more fair.