Computer Vision Tools for Tumor Segmentation

In High Energy Experiments there is extensive use of Machine Learning (ML) and Deep Learning (DL) algorithms. These well-established algorithms extract complex features from the data and are used for event and particle identification, energy estimation, and pile-up suppression. We present the application of these tools in the domain of pituitary tumor identification in MRI and PET-CT scans. In this proceeding, we present the use of deep convolutional network architectures such as UNET, Mask R-CNN, and YOLO for the initial studies for localization of the pituitary gland. We use Transfer Learning and data augmentation for improving the performance of the networks. Two different types of datasets (MRI and CT) available freely on Kaggle are used for this purpose. We provide an explanation of the algorithms used, their performance, and a comparison with different backbones such as ResNet and VGG on the MRI image dataset. We also provide the results of using data augmentation on the MRI dataset.