Early Detection and Segmentation of Ovarian Tumor Using Convolutional Neural Network with Ultrasound Imaging

• Published in: Revue d'Intelligence Artificielle Vol. 37; no. 6; p. 1503
• Main Authors: Ohood Fadil Alwan, Qusay Kanaan Kadhim, Rasha Basim Issa, Shaymaa Taha Ahmed
• Format: Journal Article
• Language: English; French
• Published: Edmonton International Information and Engineering Technology Association (IIETA) 01.12.2023
• Subjects: Abdomen; Algorithms; Artificial intelligence; Artificial neural networks; Biopsy; Cancer; Classification; Cysts; Datasets; Deep learning; Differentiation; Efficiency; Fuzzy sets; Gynecology; Image enhancement; Image quality; Image segmentation; Machine learning; Medical imaging; Neural networks; Ovarian cancer; Physiology; Surgery; Thyroid gland; Tumors; Ultrasonic imaging; Ultrasonic testing
• ISSN: 0992-499X; 1958-5748
• DOI: 10.18280/ria.370614

The clinical differentiation between benign and malignant ovarian tumors represents a substantial challenge in the fields of obstetrics and gynecology, particularly following the detection of ovarian cysts through ultrasound. Ovarian cancers, diverse in type, often exhibit overlapping characteristics that complicate diagnosis. In this study, a deep learning-based methodology was developed to aid in the rapid and accurate differentiation of ovarian cancer types using ultrasound imaging. A deep learning approach, utilizing transfer learning with Convolutional Neural Network (CNN) models, was employed. To ensure the stability and robustness of the solution, ten iterations of training and validation were executed, with data randomly sampled for each iteration. The mean of the ten iterations' outcomes constituted the final evaluation metric. Initially, ultrasound images were enhanced to augment the quality of the training dataset, followed by the extraction of low-level texture features for the segmentation of images. Subsequently, ten established CNN models were utilized for both training and transfer learning processes. In the culmination of the study, a multitask model was proposed, capable of concurrently executing detection and segmentation tasks. The conducted evaluations reveal that the deep learning models can classify ovarian tumors with an accuracy of 98.79%, a rate comparable to that of skilled medical practitioners.