An optimized deep neural network and transfer learning approaches for endometriosis classification
Year: 2026
Type: article
Abstract: The prevalence of endometriosis is underestimated because of the need for laparoscopy an invasive diagnostic method, which is considered the gold standard. Advanced stages of endometriosis may lead to endometrial cancer, infertility, psychological depression, leading to further complications. Endometriosis has multiple appearances; the lesions may be confused with other non-endometriotic lesions or endometriotic lesions that are non-endometriotic by appearance, or deep infiltrating ones may be missed on visual diagnosis. Therefore, this research aims to develop an endometriosis prediction by utilizing four different deep and transfer learning architecture including CNN, RestNet101V2, MobileNet, and VGG16 The proposed model employs Pelican Optimization Algorithm (POA) to extract predominant features for CNN, ResNet101V2, MobileNet, and VGG16 endometriosis classification. Image Dataset was obtained from Gynecologic Laparoscopy Endometriosis (GLENDA) repository containing 25,683 sample laparoscopic images of both pathological and non-pathological identified endometriosis regions. The experimental analysis revealed that POA_ResNet101V2, POA_MobileNet, and POA_VGG16 perform significantly better than CNN during the classification of endometriosis (pathology and non-pathology). Betterstill, MobileNet achieved a general accuracy of 100%, precision 99.5%, Recall 99.5%, and F1-score of 100%. The model demonstrates the effectiveness of transfer learning, MobileNet better than other transfer Learning in the existing studies. To address the diagnostic challenges of endometriosis, this study developed an optimized endometriosis prediction model with deep and transfer learning techniques, perform comparative analysis on the developed model and benchmark the results with existing ones. This model will assist health practitioners to early detect endometriosis and proffer appropriate solutions for patients.
DOI: 10.65752/eq1tse55
Institutions: Veritas University, Federal University of Technology Minna
Language: English
Subfield: Reproductive Medicine
Field: Medicine
Domain: Health Sciences
Open Access status: hybrid
OpenAlex ID: W7142847434