Ovarian Malignance Risk Prediction from Medical Notes: a case study in Vietnam

Timely and accurate assessment of ovarian tumor malignancy is critical for early intervention and effective treatment planning. This study investigates the effectiveness of artificial intelligence (AI) models in predicting malignancy risk based on Vietnamese ultrasound medical reports. A real-world dataset of 805 patients from the National Hospital of Obstetrics and Gynecology was collected. Textual data were extracted using an Optical Character Recognition (OCR) technique and processed, then regressed by various machine learning algorithms. Besides, to enhance the performance of models, we extracted key clinical findings and represented them in a tabular structure. ADASYN oversampling was applied to address class imbalance. Experimental results show that models using structured data, particularly RF, significantly outperform free text approaches, achieving an MSE of 0.032, R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of 0.609, and accuracy of binary classification of 93.2%. This case study demonstrates the effectiveness of combining structured AI pipelines with clinical domain knowledge for ovarian cancer risk stratification in low-resource settings.