The Segmentation of Therapeutic Target Area in Glioma Cancer Patients by Transfer Learning

This study was conducted in order to investigate the power and efficiency of transfer learning in solving the problem of deep learning data volumes for automatic segmentation of the treatment target area in glioma cancer patients.

In this study, T1, T2 and Flair images of one hundred patients whose glioma cancer was confirmed were used. After quality review, all images were normalized and resized. Then the images were given to a model in two modes with and without transfer learning and their performance was evaluated with the degree of similarity, overlap, sensitivity and accuracy.

The results of our study show that transfer learning can increase the efficiency of automatic segmentation and increase the similarity of automatic segmentation with manual segmentation to more than 76% in Flair images. Also, this method has increased the speed of reaching the desired result in T2 images that could not improve the results.

Deep learning in automatic segmentation can overcome the limitations caused by data volume in glioma patients and improve their performance.