Less Complex U-Net (LCU-Net)
Mayank Singh, Indu Saini, Neetu Sood · 2024
The application of Artificial Intelligence-based sustainable systems in disease prediction and classification is critically required. Accurate segmentation of the area of interest plays a vital role in the decision-making process of a diagnosis system. It has a direct impact on the classification accuracy. Ultrasound (UD) image has all the qualities required for a sustainable system. However, segmenting the region of interest in noisy images like Ultrasound (UD) needs to be worked upon. UD provides the safest diagnosis mode, and an automated system designed using UD can be used repeatedly. Here, we propose a network for segmenting areas of interest, like tumors in UD. The UD suffers severely from noises, which takes the segmentation challenge to the next level. We have proposed a unique pre-processing block for the US image. Then, we have proposed a unique paradigm to train a Less Complex U-Net (LCU-Net) for tumor segmentation. The values of the Dice and Jaccard coefficients obtained in the results were unprecedented and prove that the amount of complexity is added and the corresponding performance improvement is not justifiable.