Empirical Analysis of Learnable Image Resizer for Large-Scale Medical Image Classification and Segmentation

Deep Convolutional Neural Networks (DCNN) demonstrate state-of-the-art performance in computer vision and medical imaging tasks. Learning with large-scale images is still a challenging task that usually deals with resizing or patching approaches to embed in the lower dimensional space. Recently, Learnable Resizer (LR) has been proposed to analyze large-scale images for computer vision tasks. In this study, we propose two DCNN models for classification and segmentation tasks constructed with LR in combination with successful classification and segmentation architectures. The performance of the proposed models is evaluated for the Diabetic Retinopathy (DR) analysis and skin cancer segmentation tasks. The proposed model demonstrated better performance than the existing methods for both classification and segmentation tasks. For classification tasks, the proposed architectures achieved a 5.34% and 7.39% improvement in accuracy compared to the base ResNet50 model for two different input resolutions. The segmentation model yielded around 0.62% accuracy over the base model and 0.28% in Intersection-over-Union (IoU) from state-of-the-art performance. The proposed model with the resizer network enhances the capability of the existing R2U-Net for medical image segmentation tasks. Moreover, the proposed methods enable a significant advantage in learning better with a comparatively lower number of training examples. The experimental results reveal that the proposed models are better than the current approaches.