Early Detection of Benign Ovarian Tumor Classification Using U-NET+ with Hybrid Deep Learning Techniques

Ovarian cysts are common gynecological ailments that, if undiagnosed or ignored, can lead to substantial medical issues. Recognizing and categorizing ovarian cysts is critical for optimal treatment planning and patient care. Screening for ovarian cysts on time is critical for efficient treatment and better patient outcomes. Despite the fact that cysts often damage female follicles, a large ovarian cyst can induce torsion and infertility. As a result, it is critical to obtain a diagnosis as quickly as feasible. An ovarian cyst is diagnosed with an ultrasound (US) screening. Most cysts have been detected at an early stage due to the enhancement of these concepts and technologies in medical imaging. As a consequence, with this study, US images from several women's ovaries were collected to find out the kinds of ovarian cysts that would be identified. Existing attempts, however, have flaws such as slow convergence and prolonged training time. This study attempts to further develop deep learning-based segmentation of ovarian US cyst images using an active database to address this issue. The present investigation examines five types of ovarian cysts: serous tumor (ST), mucinous tumor (MT), endometrioid tumor (ET), clear cell tumor (CCT), and No tumor (NT). The input image was first pre-processed using the Diagonal bilinear interpolation (DBI) preprocessing approach, designed to enhance borders and reduce noise while retaining crucial image attributes. It's extremely helpful for medical images like USs. The pre-processed image is then segmented using U-NET +, which offers a new “deep supervision” technique in which intermediate segmentations are created at each level of the hierarchy and combined to form the final segmentation. The deep features are then extracted using the VGG19 model, which is fine-tuned on a dataset to attain high accuracy with minimal training data. A hybrid Equilibrium and Jarratt Butterfly optimization algorithm (EJBOA) is then employed for feature selection. The hybrid (Equilibrium Optimization algorithm) EOA and Jarratt-Butterfly optimization algorithm (JBOA) may be used to optimize several image quality measures or constraints at the same time. Long Short-Term Memory (LSTM) networks combined with fully linked layers are then employed for classification. The Python tool is used for the simulation of this approach. According to simulation results, the proposed approach achieves more powerful classification accuracy than established technologies, demonstrating its efficiency in cyst recognition and classification.