Fine-Tuned Faster R-CNN for Universal Lesion Detection

Abstract

Universal Lesion Detection in computed tomography scan images is an important task for clinical diagnosis. Manuel and conventional medical methods are time-consuming and error-prone. Automated detection of lesions at an early phase is important to increment the chances of treatment. In this paper, we proposes a fine-tuned deep learning-based method using Faster R-CNN and 1cycle training for universal lesion detection more accurately. The learning rate is one of the most significant hyperparameter to improve deep neural networks performance. Training with the well learning rate strategy can make a significant contribution to improve the model. The experiments showed that the proposed approach improves significantly Faster R-CNN performance for universal lesion detection. © 2022 IEEE.