Early Detection and Risk Stratification of Uterine Fibroids Using Deep Learning - A Review

Abstract

The term uterine fibroid refers to benign tumours that carry considerable morbidity in females, including symptoms like excessive menstrual bleeding, pelvic pain, and fertility problems. The capability of diagnosing and stratifying risk associated with uterine fibroids is of vital importance due to the possibility of undergoing early treatment by patients. Traditional techniques applied in the diagnosis of uterine fibroids include US and MRI, both of which exhibit low levels of sensitivity and specificity and require operator proficiency to produce reliable results. Modern technologies allow the utilisation of DL as an extremely useful tool in making medical imaging interpretation faster and more accurate. The current review demonstrates the potential of using DL algorithms, which may involve CNN, U-net, and hybrid approaches, in identifying and risk-stratifying uterine fibroids. A comparison of existing literature on the topic shows that DL models provide better predictions than their predecessors. VGG16, ResNet50, InceptionV3, DPCNN, EfficientNetB0, and MobileNetV2 + DCGAN have achieved impressive accuracies of 99.8%, 99.5%, and 97.45% respectively. There remain certain challenges, including insufficient data, lack of multimodal learning and non-generalisability issues. Here, the literature review emphasises that the growing importance of risk stratification lies in the fact that in such cases, deep learning models will not only be able to detect the presence of fibroids but also categorise the detected fibroids into low-risk, medium-risk and high-risk categories. Moreover, it emphasises the need for large-scale datasets and standardised evaluation metrics to design better and clinically relevant models. It concludes by providing some directions for future research.