Investigating the Effect of Transfer Learning on Medical Image Segmentation Performance

Abstract

This paper investigates the effect of switch studying on clinical photo segmentation performance. Switch learning entails the usage of a pre-trained model as the basis for a new technique for a comparable project. By leveraging pre-educated models, the manner of schooling a version to perform a project can be made greener. This paper evaluates the effect of transfer getting to know on medical photograph segmentation performance in terms of accuracy and speed of schooling. Moreover, the paper compares the overall performance of transfer getting to know and non-switch gaining knowledge of tactics for segmenting the tumors in MRI and CT scans. Effects from the experiments display that transfer learning outperforms non-transfer mastering approaches in the challenge of scientific image segmentation. Further, the paper offers insights into the VGG16 and U-internet architectures and indicates feasible guidelines for in addition research.