Abstract
OBJECTIVE: To investigate a panel of variables using four machine learning based classifiers, i.e., support vector machine (SVM), random forest (RF), artificial neural network (ANN) and logistic regression (LR) to make a diagnosis of ovarian cancer, differentiating it from benign ovarian masses. MATERIALS AND METHODS: A prospective observational pilot study was done between November 2021 and June 2023. Following data pre-processing to ensure compatibility with ML models, four ML algorithms, i.e., support vector machine (SVM), logistic regression (LR), random forest (RF) and artificial neural network (ANN) were tested by multimodal parameters from the datasets of 50 patients presenting with suspected epithelial ovarian cancer (Group A) or benign ovarian tumour (Group B). Statistical analysis was done using STATA version 14.0. RESULTS: We found that the machine learning approach could predict malignant tumours with appreciably high accuracy similar to a few studies done so far in this field. All four ML algorithms showed high level of accuracy with a maximum AUROC of 0.92 in the RF model. Both RF and SVM had an accuracy of 85.87 and 83.05%. CONCLUSION: The ML algorithms can detect ovarian cancers with a high level of accuracy. Further, a large-volume prospective study on large volume data sets is required before inclusion of ML algorithms in clinical practice.