Optimization Function Selection and Its Impact on Unet’s Efficiency: A Comprehensive Analysis
摘要
Automatic segmentation of thoracic organs in lung CT images is essential in the computer-aided diagnosis (CAD) process. Several powerful models have been proposed over the last few decades to facilitate this work and set up computer-aided diagnosis systems more suited to the needs; Unet is one of them, which has been proposed explicitly for medical images. In this work, we analyzed Unet-based segmentation and experimented with the impact that the choice of optimization function can have on the performance of Unet-based models. The model is evaluated based on Kaggle’s “Chest CT-Scan images Dataset.” First, we manually provide the masks associated with each image, using a set of processes such as grayscale conversion, binarization, and ROI highlighting. Then we try to predict these masks through our trained Unet network before evaluating them. Experiments show that the performance of the proposed model is higher for the case where RMSprop is the optimization function and lower for SGD.