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

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Machine Learning

Background:

• Brain tumors represent abnormal cell growth, posing challenges in medical image analysis.
• Accurate detection of pituitary tumors from Magnetic Resonance Imaging (MRI) is crucial for patient outcomes.
• Developing robust algorithms for medical image processing remains a significant research area.

Purpose of the Study:

• To propose an expert algorithm for the automated detection of pituitary brain tumors from MRI scans.
• To evaluate the performance of various deep learning models and machine learning techniques for tumor classification.
• To compare the accuracy of the proposed computer-aided system against traditional image processing methods.

Main Methods:

• Preprocessing techniques including smoothing, edge detection, and filtering were applied.
• Segmentation was performed using the watershed technique.
• Transfer learning models (Resnet 50, Inception V3, VGG16, VGG19) and machine learning algorithms (SVM, PNN, kNN) with hybrid features (GLDS, GLCM) were employed for classification.

Main Results:

• The Inception V3 model achieved a classification accuracy of 96%.
• Hybrid GLDS and GLCM features with Support Vector Machine (SVM) yielded 95% accuracy.
• Probabilistic Neural Network (PNN) and k-Nearest Neighbor (kNN) techniques attained 93% accuracy.

Conclusions:

• Computer-aided systems offer superior speed and accuracy compared to conventional image processing techniques.
• The Inception V3 model demonstrated a 1.0% accuracy improvement over GLDS + GLCM + SVM.
• A 2.1% accuracy improvement was noted for GLDS + GLCM + SVM over GLDS + GLCM + kNN, highlighting the effectiveness of hybrid features.