RESUMO
Deep neural networks (DNNs) involve advanced image processing but depend on large quantities of high-quality labeled data. The presence of noisy data significantly degrades the DNN model performance. In the medical field, where model accuracy is crucial and labels for pathological images are scarce and expensive to obtain, the need to handle noisy data is even more urgent. Deep networks exhibit a memorization effect, they tend to prioritize remembering clean labels initially. Therefore, early stopping is highly effective in managing learning with noisy labels. Previous research has often concentrated on developing robust loss functions or implementing training constraints to mitigate the impact of noisy labels; however, such approaches have frequently resulted in underfitting. We propose using knowledge distillation to slow the learning process of the target network rather than preventing late-stage training from being affected by noisy labels. In this paper, we introduce a data sample self-selection strategy based on early stopping to filter out most of the noisy data. Additionally, we employ the distillation training method with dual teacher networks to ensure the steady learning of the student network. The experimental results show that our method outperforms current state-of-the-art methods for handling noisy labels on both synthetic and real-world noisy datasets. In particular, on the real-world pathological image dataset Chaoyang, the highest classification accuracy increased by 2.39 %. Our method leverages the model's predictions based on training history to select cleaner datasets and retrains them using these cleaner datasets, significantly mitigating the impact of noisy labels on model performance.
Assuntos
Redes Neurais de Computação , Humanos , Aprendizado Profundo , Processamento de Imagem Assistida por Computador/métodos , AlgoritmosRESUMO
Despite the fact that digital pathology has provided a new paradigm for modern medicine, the insufficiency of annotations for training remains a significant challenge. Due to the weak generalization abilities of deep-learning models, their performance is notably constrained in domains without sufficient annotations. Our research aims to enhance the model's generalization ability through domain adaptation, increasing the prediction ability for the target domain data while only using the source domain labels for training. To further enhance classification performance, we introduce nuclei segmentation to provide the classifier with more diagnostically valuable nuclei information. In contrast to the general domain adaptation that generates source-like results in the target domain, we propose a reversed domain adaptation strategy that generates target-like results in the source domain, enabling the classification model to be more robust to inaccurate segmentation results. The proposed reversed unsupervised domain adaptation can effectively reduce the disparities in nuclei segmentation between the source and target domains without any target domain labels, leading to improved image classification performance in the target domain. The whole framework is designed in a unified manner so that the segmentation and classification modules can be trained jointly. Extensive experiments demonstrate that the proposed method significantly improves the classification performance in the target domain and outperforms existing general domain adaptation methods.
Assuntos
Núcleo Celular , Processamento de Imagem Assistida por ComputadorRESUMO
Since pathological images have some distinct characteristics that are different from natural images, the direct application of a general convolutional neural network cannot achieve good classification performance, especially for fine-grained classification problems (such as pathological image grading). Inspired by the clinical experience that decomposing a pathological image into different components is beneficial for diagnosis, in this paper, we propose a Divide-and-Attention Network (DANet) for Hematoxylin-and-Eosin (HE)-stained pathological image classification. The DANet utilizes a deep-learning method to decompose a pathological image into nuclei and non-nuclei parts. With such decomposed pathological images, the DANet first performs feature learning independently in each branch, and then focuses on the most important feature representation through the branch selection attention module. In this way, the DANet can learn representative features with respect to different tissue structures and adaptively focus on the most important ones, thereby improving classification performance. In addition, we introduce deep canonical correlation analysis (DCCA) constraints in the feature fusion process of different branches. The DCCA constraints play the role of branch fusion attention, so as to maximize the correlation of different branches and ensure that the fused branches emphasize specific tissue structures. The experimental results of three datasets demonstrate the superiority of the DANet, with an average classification accuracy of 92.5% on breast cancer classification, 95.33% on colorectal cancer grading, and 91.6% on breast cancer grading tasks.
RESUMO
BACKGROUND: Due to the recent advances in deep learning, this model attracted researchers who have applied it to medical image analysis. However, pathological image analysis based on deep learning networks faces a number of challenges, such as the high resolution (gigapixel) of pathological images and the lack of annotation capabilities. To address these challenges, we propose a training strategy called deep-reverse active learning (DRAL) and atrous DenseNet (ADN) for pathological image classification. The proposed DRAL can improve the classification accuracy of widely used deep learning networks such as VGG-16 and ResNet by removing mislabeled patches in the training set. As the size of a cancer area varies widely in pathological images, the proposed ADN integrates the atrous convolutions with the dense block for multiscale feature extraction. RESULTS: The proposed DRAL and ADN are evaluated using the following three pathological datasets: BACH, CCG, and UCSB. The experiment results demonstrate the excellent performance of the proposed DRAL + ADN framework, achieving patch-level average classification accuracies (ACA) of 94.10%, 92.05% and 97.63% on the BACH, CCG, and UCSB validation sets, respectively. CONCLUSIONS: The DRAL + ADN framework is a potential candidate for boosting the performance of deep learning models for partially mislabeled training datasets.