Image denoising is an important pre-processing step in medical imageanalysis. Different algorithms have been proposed in past three decades withvarying denoising performances. More recently, having outperformed allconventional methods, deep learning based models have shown a great promise.These methods are however limited for requirement of large training sample sizeand high computational costs. In this paper we show that using small samplesize, denoising autoencoders constructed using convolutional layers can be usedfor efficient denoising of medical images. Heterogeneous images can be combinedto boost sample size for increased denoising performance. Simplest of networkscan reconstruct images with corruption levels so high that noise and signal arenot differentiable to human eye.
展开▼
