Attention-guided lightweight deep learning architecture for classification of polycystic ovary syndrome from ultrasound images

Polycystic ovary syndrome is one of the major causes of endometrial cancer,cardiovascular risk and irregular menstruation. Early detection of the disease can enable a better diagnostic approach for patients. Hence the medical diagnosis setup contain the light weight architecture, usually performs better then the weighted architectures, due to requirement of the low Computational resource and faster inference for real time diagnosis. This work has embedded Attention based mechanism into Convnet, VGG19, DenseNet-121, and EfficientNet-B0 based CNN architectures containing transfer learning approach. These architectures have shown the excellent performance in terms of accuracy, precision and recall. The DenseNet-121 + Attention provides the best prediction performance with an accuracy of 0.9913, while EfficientNet-B0 + Attention offers the most parameter-efficient solution and achieves a competitive accuracy of 0.9913.