Flattening Singular Values of Factorized Convolution for Medical Images

TL;DR

A Singular value equalization generalizer-induced Factorized Convolution (SFConv) is proposed to improve the expressive power of factorized convolutional layers in MIP models and yields competitive expressiveness over vanilla convolutions while reducing complexity.

Abstract

Convolutional neural networks (CNNs) have long been the paradigm of choice for robust medical image processing (MIP). Therefore, it is crucial to effectively and efficiently deploy CNNs on devices with different computing capabilities to support computer-aided diagnosis. Many methods employ factorized convolutional layers to alleviate the burden of limited computational resources at the expense of expressiveness. To this end, given weak medical image-driven CNN model optimization, a Singular value equalization generalizer-induced Factorized Convolution (SFConv) is proposed to improve the expressive power of factorized convolutions in MIP models. We first decompose the weight matrix of convolutional filters into two low-rank matrices to achieve model reduction. Then minimize the KL divergence between the two low-rank weight matrices and the uniform distribution, thereby reducing the number of singular value directions with significant variance. Extensive experiments on fundus and OCTA datasets demonstrate that our SFConv yields competitive expressiveness over vanilla convolutions while reducing complexity.

Figures (3)

• Figure 1: Measures of skewness and kurtosis of data distribution by calculating the histogram of pixel values. The first row is two natural images randomly selected from the VOC2012 dataset. The second row is two chest X-rays randomly sampled from the VIN-CXR dataset, and their corresponding lesion regions are in the third row.
• Figure 2: Schematic diagram of SFConv, where $s$ is stride and $p$ is padding. The other variables are described in section \ref{['sec:fc']}.
• Figure 3: Parameter distributions of the trained U-Net.