Improvement of Ovarian Detection from Ultrasound Images Using Ensemble Learning

Tumor detection is a challenging task that requires high accuracy and efficiency. In this paper, we propose a novel method that combines two state-of-the-art object detection models: Faster R-CNN and YOLOv5 to improve the accuracy of ovarian tumor detection. Faster R-CNN is a region-based convolutional neural network that can detect tumors with high precision but has lower performance in classification. YOLOv5, in contrast, is a single-stage detector that can classify tumors with high confidence but has lower performance in detection. To leverage the strengths of both models, we adopted an ensemble learning approach that fused their outputs. We evaluated our method on a dataset of ultrasound images with ovarian tumors. The results showed that our method improved the detection of tumors in terms of recall, AP50, and true positive images, compared to using each method alone.