Segmentation on brain MR images by using deep learning network and 3D modelling

In the last few years, utilizing deep learning techniques for predicting tumor presence on brain MR images became quite common.We offer a semantic segmentation method that uses a convolutional neural network to autonomously segment brain tumors on 3D Brain Tumor Segmentation (BraTS) image data sets using four different imaging modalities in this research (T1, T1C, T2, and Flair). In addition, our research incorporates whole-brain 3D imaging and a comparison of ground truth and anticipated labels in 3D. This method was effectively applied to acquire specific tumor regions and measurements such as height, width, and depth, and images were presented in various planes including sagittal, coronal, and axial. In terms of tumor prediction, the evaluation findings of semantic segmentation performed by a deep learning network are extremely promising. The average prediction ratio was found to be 91.718. The mean IoU (Intersection over Union) score was 86.946 and the mean BF score was 92.938. Finally, the test images' dice scores revealed a considerable resemblance between the ground truth and predicted labels. As a result, semantic segmentation metrics and 3D imaging can both be viewed as useful for effectively diagnosing brain tumors. Calculating the surface areas of the brain, real and predicted labels, and applying this process to all slices is very useful in terms of comparing tumor volume information. As a result, the predicted volume values are very close to the ground truth volume values, showing that this study can determine the presence of tumors by the deep learning method and tumor size by surface area algorithm. In addition, the calculation of the brain tumor volume and 3D modeling will facilitate clear visualization of the tumor site and understanding the size of the tumor.