Convolution Neural Networks for Clinical Image Segmentation with Evolutionary Compression

Convolutional Neural Networks (CNN) have been very useful for image-related applications as a result of their high level of accuracy and their ability to analyze data in real time. Neural networks have a variety of potential applications, some of which include the classification of pictures, the fragmentation of images, and the identification of objects in photographs. Images are often used in the medical industry for the purposes of diagnosis as well as surgery. In this article, a study on the use of convolutional neural networks (CNNs) for clinical picture segmentation is presented, coupled with an approach for evolutionary model compression. On a collection of clinical pictures, four distinct models—UNet, ENet, ECO-UNet, and ECO-ENet—were assessed. The accuracy of their segmentation was gauged using the Dice coefficient, sensitivity, and specificity. The findings indicate that although the UNet and ENet models had high segmentation accuracy, the ECO-UNet and ECO-ENet models had even greater accuracy. The model size was subsequently decreased using the evolutionary compression technique, and even at high compression ratios, the compressed models' performance was comparable to that of the original models. These findings show how successful the evolutionary compression approach is in reducing model size as well as the potential of CNNs for clinical picture segmentation tasks. In environments with limited resources, this strategy may be very helpful. Overall, the research demonstrates that evolutionary compression and CNNs provide a potential method for clinical picture segmentation..