A multimodal feature fusion with deep representation learning approach for polycystic ovary syndrome diagnosis using ultrasound images.

Nowadays, polycystic ovary syndrome (PCOS) is among the very first causes of female infertility, which is a hormone imbalance that affects women of childbearing age. The earlier diagnosis of many cysts using ovarian ultrasound image scans is the most predictable technique to make a precise identification of PCOS and to make a proper treatment plan to cure patients with this disorder. Nevertheless, the analysis depends on Rotterdam conditions, comprising a higher level of androgen hormones, polycystic ovaries, and ovulation failure on the ultrasound images. Nowadays, radiologists and physicians manually execute PCOS identification using ovarian ultrasound by calculating the follicle counts and defining their volume in the ovaries, which is the most challenging PCOS diagnostic condition. Still, the conventional processes used for identifying PCOS using computer-based methods depend on numerous image processing methods and then classic machine learning (ML) tactics for image classification, which is a repetitive procedure with comparatively lower performance. Currently, some scholars have applied deep learning (DL) techniques to discover PCOS from ultrasound images. In this manuscript, a Multi-Model Feature Engineering and Deep Learning for Diagnosis of Polycystic Ovary Syndrome (MMFEDL-DPCOS) model is proposed. The MMFEDL-DPCOS model aims to analyse and diagnose PCOS employing ultrasound images for a precise and pre-diagnosis stage. Primarily, the Gaussian filtering (GF) method is utilized in the image pre-processing phase for enhancing image quality by reducing the noise. Furthermore, the fusion of InceptionResNetv2, EfficientNetV2B3, VGG16, ResNet-50, and Inception‐V3 methods is employed for feature extraction. Finally, the regularized stacked autoencoder (RSAE) methodology is utilized for classification. The comparison analysis of the MMFEDL-DPCOS methodology portrayed a superior accuracy value of 98.68% over existing models under the PCOS dataset.