Enhancing Endometriosis Prediction for Women Using Deep Learning Based Frame Work

Abstract

newline Endometriosis is a recurrence disorder that creates an impact on the global women population. During the menstrual cycle, the tissue that appears inside the uterus of women will be shed. In certain scenarios, the tissue spread across various reproductive organs including the ovary, fallopian tube, peritoneum, etc. The impact of endometriosis exists among women of age between 18 to 50. The traditional approach to recognize the occurrence and severity of endometriosis is by scanning procedures followed by laparoscopic surgery. Machine learning and deep learning algorithms is a cutting-edge technology in the diagnosis of diseases at earlier stages. newlineMachine learning and deep learning algorithm assist the gynecologist in the prediction of diseases. The carried research work employs machine learning and deep learning algorithms in the prediction and severity of endometriosis from the laparoscopic images. In the framed research work, endometriosis was recognized pathologically with the help of the transfer learning technique ResNet50. Here the laparoscopic images obtained are used for training and testing the model. The reported work performs well in categorizing the endometriosis and normal cysts using ResNet50 architecture with a training accuracy of 87.5% and test accuracy of 86.5%. newlineThe pathologically identified endometriosis is used as input for anatomically segmenting various regions affected by endometriosis. The pathologically proven images and their corresponding annotated images are used as input for execution. The carried research work invokes a semantic segmentation process to identify the various regions affected by endometriosis. The localization of endometriosis was effectively performed by implementing U-Net architecture. The approach was implemented with other architectures including a Fully newline newline newlineConvolution neural network. The U-net architecture executes well in segmenting the regions affected with an intersection over union of 0.72 and a dice coefficient of 0.74. newlineThe next carried approach is to predict the severity and likelihood of endometriosis using several symptoms that are associated with endometriosis. The internal symptoms were identified from laparoscopic images and the external symptoms were used as input for prediction. The internal symptoms include adnexal mass, lesion color, and size. Based on the severity of the symptoms corresponding weight value was assigned and evaluation was performed on the data obtained from a retrospective study. The framed research work internal factor-based severity analysis outperforms well with an AUC- ROC of 0.88. Also, the external factor-based severity analysis performs well with an AUC-ROC of 0.86. The represented work was validated with the LASSO algorithm, where the internal and external factors-based severity analysis performs quite well with an accuracy of 86% and 80% respectively newline