Detection of brain tumors using a transfer learning-based optimized ResNet152 model in MR images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Machine Learning

Background:

• Brain tumors significantly impact life expectancy and are typically detected using magnetic resonance (MR) scans.
• Artificial intelligence (AI) and deep learning methods offer enhanced diagnostic accuracy through efficient data processing in medical imaging.

Purpose of the Study:

• To investigate the effectiveness of deep transfer learning techniques for accurate brain tumor diagnosis using MR images.
• To compare the performance of pretrained deep learning models (ResNet152, GoogleNet) in conjunction with conventional machine learning classifiers for brain tumor detection.

Main Methods:

• A preprocessing pipeline involving morphological operations, Gaussian blurring, and thresholding was applied to enhance MR image quality.
• Principal Component Analysis (PCA) for dimensionality reduction and data augmentation were utilized, followed by feature extraction using ResNet152 and GoogleNet.
• Extracted features were classified using Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Classification and Regression Trees (CART), and Gaussian Naive Bayes (GNB).

Main Results:

• ResNet152 demonstrated superior performance over GoogleNet in brain tumor classification.
• ResNet152 achieved a high accuracy of 98.53%, an F1 score of 97.4%, and a sensitivity of 96.52%.

Conclusions:

• The integration of deep transfer learning (ResNet152) with traditional machine learning classifiers provides an effective approach for accurate brain tumor detection.
• This hybrid methodology enhances diagnostic capabilities in medical image analysis, potentially improving patient outcomes.