Deep learning-based ovarian cyst classification and abnormality detection using convolutional neural networks

The goal of this study was to design a computer-aided diagnosis system using deep learning network for ovarian cyst classification and abnormality detection. An ovarian cyst is a sac filled with fluid or semisolid material that forms on or within one or both the ovaries. One in every 4–5 women in India suffers from different types of ovarian cysts (20–25% incidence). The doctor-population ratio of 0.62:1000 further worsens this situation for the most populous country on the planet. The main motivation for this work was to ease this situation by providing an automated system which can act as an aid to doctors in diagnosing ovarian cysts for different complications requiring immediate medical intervention. It can also be used as teaching aid for the gynecology/radiology students to identify the sonographic variations being exhibited by different ovarian cyst types. A new deep learning convolutional neural network dedicated for Ovarian cysts classification and abnormality detection, OvarianCystNet with 87 layers is designed using Deep Learning Toolbox of MATLAB R 2022b. Most of deep learning ovarian cyst classification systems are either using only pre-trained networks like VGG16 or inception V3. We have designed a dedicated deep learning network specifically for ovarian cyst classification. The classification results for OvarianCystNet show promising results and validate its use as tool for ovarian cyst diagnosis. A validation accuracy of 92.06% is achieved. It is 18.68% more accurate in comparison to VGG19, 7.65% more accurate in comparison with VGG16, 3.47% more accurate in comparison with MobileNetV2 and 1.49% more accurate in comparison with Resnet50. The training accuracy, validation loss, confusion matrix, receiver operating characteristics, and area under the curve is compared with four pre-trained deep learning networks namely MobilenetV2, Resnet50, VGG16 and VGG19. A dataset consisting of around 1319 images of ovarian cyst ultrasound images was collected from a leading obstetrics and gynecological hospital by several operators on Voluson-S8 a 4D ultrasound machine. Voluson-S8 is manufactured by general electric which is specifically used for obstetrics and gynecology related scans. All images in the dataset were manually labeled by an expert obstetrics and gynecology doctor with over 15 years of experience. Images were divided into five classes. © The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2024.