Framework for Uterine Fibroids Segmentation from Ultrasonography

• Main Authors: S, Reshma Shree, J, Anitha, D, Sujitha Juliet
• Format: Conference Proceeding
• Language: English
• Subjects: Accuracy; Analytical models; Computational modeling; Convolutional neural networks; Custom Convolutional Neural Networks; Fibroid Segmentation; Fibroids Detection; Image segmentation; Inception V3; Medical diagnostic imaging; Medical services; MobileNetV2; Noise; Testing; Ultrasonography; Ultrasound Images; Uterine Fibroids
• ISSN: 2469-5556
• DOI: 10.1109/ICACCS60874.2024.10717013

Abnormal growths in the uterine wall, called Uterine fibroids can lead to infertility. Ultrasound imaging has been used more frequently in the past few years to diagnose and treat uterine fibroids. Furthermore, proper identification of fibroids might be challenging due to their diverse textures and potential closeness to other structures. Currently, medical personnel frequently rely on manual observations for diagnosis. The principal aim of the framework that has been presented for the segmentation of uterine fibroid tissue is to accurately identify and delineate the boundaries of fibroid tissues in ultrasonography images. The first step in the framework's multiple phases is a preprocessing stage that enhances image quality. Next, a region of interest (ROI) is determined, and an initial seed is generated to begin the segmentation process. In order to identify fibroid tissues among noise and nearby structures, deep learning models are used for feature extraction. These methods enable the framework to effectively address difficulties related to speckles, surrounding tissues, and image noise, resulting in a more accurate fibroids boundary segmentation. In addition, these algorithms help with precise fibroid boundary delineation. In this work, models such as Custom Convolutional Neural Network (Custom CNN), InceptionV3, and MobileNetV2 are analyzed. With an accuracy rate of 96%, the custom CNN model performs better than the other models.