Hybrid Machine Learning (Ml)-Based System for Detection of Uterine Fibroids from Ultrasound Images Using Convolutional Neural Network (CNN) and Attention Mechanism

Abstract

Uterine fibroids are among the most common benign tumors affecting women of reproductive age, and their timely detection is crucial for effective clinical management. Traditional diagnostic practices rely on expert interpretation of ultrasound images, which is often time-intensive and subject to variability. This study presents a hybrid machine learning system for the early detection of uterine fibroids using transabdominal and transvaginal ultrasound images. The proposed system integrates Convolutional Neural Networks (CNN) with advanced feature refinement techniques (Attention Mechenism) to improve diagnostic accuracy and reliability. A curated dataset obtained from the Kaggle repository was used, and preprocessing methods such as contrast normalization and noise reduction were applied to enhance image quality. Experimental results demonstrated strong performance, with an accuracy of 94%, precision of 92%, recall of 90%, and an F1-score of 91%. These balanced metrics highlight the robustness of the hybrid approach, offering consistent detection of fibroid-positive cases while minimizing false positives and negatives. The system shows promise as a clinical decision-support tool, particularly in resource-limited settings where radiological expertise is scarce. Future research will focus on expanding the dataset, incorporating explainable AI methods for greater transparency, and validating the model across diverse populations and imaging protocols.