Empirical Study of Ovarian Tumors Detection and Classification in Ultrasound Images

  • Le hung MICA Institute, HUST - CNRS/UMI-2954 - GRENOBLE INP, Vietnam
Keywords: MMOTU dataset, Ovarian tumor detection and classification, Data augmentation, Ultrasound images, YOLOv5 and YOLOv7


Ovarian cancer is one of the leading causes of death in women. Ultrasound images are often used to provide initial findings and diagnosis before further tests are performed. Although a number of works have been done for medical image analysis, ovarian cancer detection, and recognition are still quite limited. Most of the existing works tried to detect or classify two tumor categories (benign or benign and malignant lesions). In this paper, we perform an initial comprehensive study of ovarian tumor detection, and at the same time classify eight classes of ovarian tumors in ultrasound images using deep-learning models. Due to the lack of training data, we conducted a data augmentation process and investigated the improvement of performance with and without data augmentation. Experiments carried out on the OTU-2D set of MMOTU dataset with YOLOv5, YOLOv7, and YOLOv7 variants, show promising results of detection and classification.