CAD System Design for Pituitary Tumor Classification based on Transfer Learning Technique

Background: A brain tumor is an asymmetrical expansion by cells inevitably emulating amid them. Image processing is a vibrant research area where the handing out of the image in the medical field is an exceedingly tricky field. In this paper, an expert algorithm is suggested for the detection of pituitary brain tumors from MR images. Methods: The preprocessing techniques (smoothing, edge detection, filtering) and segmentation techniques (watershed) are applied to the online data set. The transfer learning technique is used as a classifier whose performance is measured in terms of classification accuracy. Resnet 50, Inception V3VGG16, and VGG19 models are used as classification algorithms. The proposed model is validated using different machine learning techniques considering hybrid features. Results: 96% accuracy was obtained employing the Inception V3 model & 95% accuracy was attained using hybrid GLDS and GLCM features employing Support Vector Machine algorithm while 93% was attained using Probabilistic Neural Network and k Nearest Neighbor techniques. Conclusion: Computer-aided systems gave much faster and more accurate results than image processing techniques.1.0% accuracy improvement was observed while using Inception V3 over GLDS + GLCM + SVM and 2.1% accuracy improvement using GLDS + GLCM + SVM over GLDS + GLCM + kNN.