Enhancing PCOS Detection: A Hybrid Approach Utilizing Blood Profiles and Ultrasound Imaging

• Sandosh S

  Author

Polycystic Ovary Syndrome (PCOS) is a common condition affecting women’s ovaries, leading to hormonal imbalances during their reproductive years. This imbalance results in various issues such as irregular menstrual cycles, acne, hair loss, thinning of hair, cramps, insulin resistance, metabolic syndrome, infertility, and mood swings. Despite its widespread occurrence, many cases go undetected, and PCOS is increasingly being recognized as a psychosomatic disorder. This research is a comprehensive investigation into the diagnosis of PCOS, employing an integrated approach utilizing both blood parameters and ultrasound images. For blood-based diagnostics, machine learning algorithms like random forest and hyperparameter tuning using RandomizedSearchCV and GridSearchCV, Decision Tree with cost complexity pruning, Logistic regression, KNN, Na¨ıve Bayes, Neural network, LDA, QDA, Nearest centroid classifier, Gaussian process classifier, Full grown tree, Voting classifier (with logistic regression, Decision Tree, and SVM), Bagging classifier, Extratrees classifier, Adaboost classifier, XGBoost have been used which revealed patterns emphasizing the significance of hormonal imbalances and metabolic parameters. Concurrently, ultrasonographic evaluations provided insights into ovarian morphology, emphasizing the presence of follicles and other structural abnormalities. For differentiating between the images of infected and non-infected patients CNN with softmax activation in the output layer and ResNet50 have been used. This offers a deeper insight into the multi-faced nature of PCOS. There have been reports where PCOS is not detected in the ultrasound images but shows its presence in the patient’s body by deranging the hormones. Thus, we integrated both the findings to find a model that can detect PCOS with more sensitivity and specificity.

• PDF