Fine-Tuning based Deep Transfer Learning System used to Identify the Stage of Brain Tumour from MR-Images

Brain tumour detection based on the automation of 3D images helps the medical field perform deep analysis. Traditional approaches use 2D images in medical applications. In some cases, the system to identify the blocks or tumour regions is essential to measure depth along with length and breadth, which is impossible with 2D images. The proposed method has designed a fine-tuned CNN containing the convolution 3D layers with global pooling layers, which has the advantage of creating a single feature map for performing classification rather than multiple feature maps. These maps help the model interpret each class label's confidence level. In the proposed model, the layers implemented for feature extraction are frozen, and weights are trained. The fine-tuning process helps the model acquire similarities between the elements through mitigation learning. The proposed model uses VGG-16 to retrain it because it accurately classifies the images. The first layers of VGG are stacked with odd-numbered size layers to get all the structural parameters. The last layers of VGG-16 are customized for identifying the stage of the tumour. The weights obtained in the initial layers are used to train the model to learn about the complex structures.