A Comparative Evaluation of Deep Convolutional Neural Networks for Uterine Fibroid MRI Classification

Abstract

Uterine fibroids are among the most common benign gynaecological tumours affecting women of reproductive age and are associated with significant reproductive and gynaecological complications. Accurate classification of uterine fibroids from magnetic resonance imaging (MRI) is essential for effective diagnosis, treatment planning, and clinical decision-making. Although deep convolutional neural networks (CNNs) have shown promising performance in medical image analysis, limited studies have comparatively evaluated their effectiveness for uterine fibroid MRI classification. This study presents a comparative performance analysis of two pre-trained CNN architectures, GoogleNet and MobileNet, for an automated uterine fibroid MRI classification system. A total of 300 MRI images obtained from the University of Minnesota (UMD) dataset on Figshare were preprocessed through image resizing, normalisation, and data augmentation before deep feature extraction. Model performance was evaluated using accuracy, precision, recall, and F1-score. The experimental results showed that GoogleNet outperformed MobileNet, achieving an overall classification accuracy of 81.1% and a recall of 61.0%, compared with 59.6%, 59.8%, and 56.7%, respectively, for MobileNet. The class-wise evaluation further produced F1-scores of 0.59 and 0.60 for the No Fibroid and Fibroid classes, indicating balanced classification performance. The superior performance of GoogleNet is attributed to its Inception architecture, which effectively captures multi-scale image features from MRI images. The findings demonstrate that GoogleNet provides a more reliable approach for automated uterine fibroid MRI classification and has the potential to support computer-aided diagnosis in clinical practice.