Bayesian statistics‐guided label refurbishment mechanism: Mitigating label noise in medical image classification

Abstract

Purpose Deep neural networks (DNNs) have been widely applied in medical image classification, benefiting from its powerful mapping capability among medical images. However, these existing deep learning‐based methods depend on an enormous amount of carefully labeled images. Meanwhile, noise is inevitably introduced in the labeling process, degrading the performance of models. Hence, it is significant to devise robust training strategies to mitigate label noise in the medical image classification tasks. Methods In this work, we propose a novel Bayesian statistics‐guided label refurbishment mechanism (BLRM) for DNNs to prevent overfitting noisy images. BLRM utilizes maximum a posteriori probability in the Bayesian statistics and the exponentially time‐weighted technique to selectively correct the labels of noisy images. The training images are purified gradually with the training epochs when BLRM is activated, further improving classification performance. Results Comprehensive experiments on both synthetic noisy images (public OCT & Messidor datasets) and real‐world noisy images (ANIMAL‐10N) demonstrate that BLRM refurbishes the noisy labels selectively, curbing the adverse effects of noisy data. Also, the anti‐noise BLRMs integrated with DNNs are effective at different noise ratio and are independent of backbone DNN architectures. In addition, BLRM is superior to state‐of‐the‐art comparative methods of anti‐noise. Conclusions These investigations indicate that the proposed BLRM is well capable of mitigating label noise in medical image classification tasks.

Attention-Based Self-Supervised Learning Monocular Depth Estimation With Edge Refinement

“…Notably, the noise rate was set as 8% for sample-selection style methods. Meanwhile, we report some performance of other popular anti-noise methods based on the ANIMAL10N given in original papers [13], [29]. We conducted our method on different backbones, including ResNet-18 and VGG-19.…”

Deep Learning for Automatic Vision-Based Recognition of Industrial Surface Defects: A Survey

“…• Label smoothing and refurbishment: the former transforms the hard label y (e.g., one-hot encoded for a K-class classification) to a soft target y s acting as a regularization technique [288]: y s = (1 − α)y + α K , where α parameter modulates the level of confidence during training thus avoiding over-fitting predictions. The latter consists in replacing the original given noisy label with a refurbished one; Gao et al [289] adopted a plug-and-play additional module with Bayesian statistic and a time-weighting module for optimal label selection.…”

Ensemble Meta-Learning using SVM for Improving Cardiovascular Disease Risk Prediction

“…Studies often focus on traditional ensemble methods [14], but the advantages of strategically analyzing predicted probabilities using a SVM meta-learner warrant further investigation. Meta-learning frameworks have shown the ability to surpass the accuracy of single-model or ensemble approaches in other domains by learning patterns of model performance across diverse tasks [29, 30]. Applying this concept to CVD risk prediction holds the potential to significantly improve accuracy, as individual models often demonstrate varying reliability when dealing with complex patient risk factors [31].…”