Abstract:Background Deep learning technology has been widely applied to medical image analysis. But due to the limitations of its own imaging principle, ultrasound image has the disadvantages of low resolution and high Speckle Noise density, which not only hinder the diagnosis of patients’ conditions but also affect the extraction of ultrasound image features by computer technology. Objective In this study, we investigate the robustness of deep convolutional neural network (CNN) for classification, segmentation, and target detection of breast ultrasound image through random Salt & Pepper Noise and Gaussian Noise. Methods We trained and validated 9 CNN architectures in 8617 breast ultrasound images, but tested the models with noisy test set. Then, we trained and validated 9 CNN architectures with different levels of noise in these breast ultrasound images, and tested the models with noisy test set. Diseases of each breast ultrasound image in our dataset were annotated and voted by three sonographers based on their malignancy suspiciousness. we use evaluation indexes to evaluate the robustness of the neural network algorithm respectively. Results There is a moderate to high impact (The accuracy of the model decreased by about 5%-40%) on model accuracy when Salt and Pepper Noise, Speckle Noise, or Gaussian Noise is introduced to the images respectively. Consequently, DenseNet, UNet++ and Yolov5 were selected as the most robust model based on the selected index. When any two of these three kinds of noise are introduced into the image at the same time, the accuracy of the model will be greatly affected. Conclusions Our experimental results reveal new insights: The variation trend of accuracy with the noise level in Each network used for classification tasks and object detection tasks has some unique characteristics. This finding provides us with a method to reveal the black-box architecture of computer-aided diagnosis (CAD) systems. On the other hand, the purpose of this study is to explore the impact of adding noise directly to the image on the performance of neural networks, which is different from the existing articles on robustness in the field of medical image processing. Consequently, it provides a new way to evaluate the robustness of CAD systems in the future.

Improving Robustness of Medical Image Diagnosis with Denoising Convolutional Neural Networks

Spatial-Adaptive Network for Single Image Denoising

CNN and Multi-Feature Extraction Based Denoising of CT Images

CNN Denoising for Medical Image Based on Wavelet Domain

Robust Convolutional Neural Networks Against Adversarial Attacks on Medical Images

Feature-Guided CNN for Denoising Images From Portable Ultrasound Devices

Feature Denoising for Improving Adversarial Robustness

A Robust Deformed Convolutional Neural Network (CNN) for Image Denoising.

Cross-denoising Network against Corrupted Labels in Medical Image Segmentation with Domain Shift

Dilated Residual Encode-Decode Networks for Image Denoising

Removing Adversarial Noise in X-ray Images via Total Variation Minimization and Patch-Based Regularization for Robust Deep Learning-based Diagnosis

Two-Stage Convolutional Neural Network for Medical Noise Removal Via Image Decomposition.

Do Noises Bother Human and Neural Networks In the Same Way? A Medical Image Analysis Perspective

Research on denoising and enhancing methods of medical images based on convolutional neural networks

Convolutional Neural Networks For Image Denoising And Restoration

Medical image denoising using convolutional neural network: a residual learning approach

Digital Radiography Image Denoising Using a Generative Adversarial Network

A Robust System for Noisy Image Classification Combining Denoising Autoencoder and Convolutional Neural Network

A Variational Network for Biomedical Images Denoising using Bayesian model and Auto-Encoder

A Novel Medical Image Denoising Method Based on Conditional Generative Adversarial Network

Noise-robustness test for ultrasound breast nodule neural network models as medical devices