The accurate segmentation of breast ultrasound images is an important precondition for the lesion determination. The existing segmentation approaches embrace massive parameters, sluggish inference speed, and huge memory consumption. To tackle this problem, we propose T2KD Attention U-Net (dual-Teacher Knowledge Distillation Attention U-Net), a lightweight semantic segmentation method combined double-path joint distillation in breast ultrasound images. Primarily, we designed two teacher models to learn the fine-grained features from each class of images according to different feature representation and semantic information of benign and malignant breast lesions. Then we leveraged the joint distillation to train a lightweight student model. Finally, we constructed a novel weight balance loss to focus on the semantic feature of small objection, solving the unbalance problem of tumor and background. Specifically, the extensive experiments conducted on Dataset BUSI and Dataset B demonstrated that the T2KD Attention U-Net outperformed various knowledge distillation counterparts. Concretely, the accuracy, recall, precision, Dice, and mIoU of proposed method were 95.26%, 86.23%, 85.09%, 83.59%and 77.78% on Dataset BUSI, respectively. And these performance indexes were 97.95%, 92.80%, 88.33%, 88.40% and 82.42% on Dataset B, respectively. Compared with other models, the performance of this model was significantly improved. Meanwhile, compared with the teacher model, the number, size, and complexity of student model were significantly reduced (2.2×106 vs. 106.1×106, 8.4 MB vs. 414 MB, 16.59 GFLOPs vs. 205.98 GFLOPs, respectively). Indeedy, the proposed model guarantees the performances while greatly decreasing the amount of computation, which provides a new method for the deployment of clinical medical scenarios.
Accurate segmentation of pediatric echocardiograms is a challenging task, because significant heart-size changes with age and faster heart rate lead to more blurred boundaries on cardiac ultrasound images compared with adults. To address these problems, a dual decoder network model combining channel attention and scale attention is proposed in this paper. Firstly, an attention-guided decoder with deep supervision strategy is used to obtain attention maps for the ventricular regions. Then, the generated ventricular attention is fed back to multiple layers of the network through skip connections to adjust the feature weights generated by the encoder and highlight the left and right ventricular areas. Finally, a scale attention module and a channel attention module are utilized to enhance the edge features of the left and right ventricles. The experimental results demonstrate that the proposed method in this paper achieves an average Dice coefficient of 90.63% in acquired bilateral ventricular segmentation dataset, which is better than some conventional and state-of-the-art methods in the field of medical image segmentation. More importantly, the method has a more accurate effect in segmenting the edge of the ventricle. The results of this paper can provide a new solution for pediatric echocardiographic bilateral ventricular segmentation and subsequent auxiliary diagnosis of congenital heart disease.
Early screening based on computed tomography (CT) pulmonary nodule detection is an important means to reduce lung cancer mortality, and in recent years three dimensional convolutional neural network (3D CNN) has achieved success and continuous development in the field of lung nodule detection. We proposed a pulmonary nodule detection algorithm by using 3D CNN based on a multi-scale attention mechanism. Aiming at the characteristics of different sizes and shapes of lung nodules, we designed a multi-scale feature extraction module to extract the corresponding features of different scales. Through the attention module, the correlation information between the features was mined from both spatial and channel perspectives to strengthen the features. The extracted features entered into a pyramid-similar fusion mechanism, so that the features would contain both deep semantic information and shallow location information, which is more conducive to target positioning and bounding box regression. On representative LUNA16 datasets, compared with other advanced methods, this method significantly improved the detection sensitivity, which can provide theoretical reference for clinical medicine.
To accurately capture and effectively integrate the spatiotemporal features of electroencephalogram (EEG) signals for the purpose of improving the accuracy of EEG-based emotion recognition, this paper proposes a new method combining independent component analysis-recurrence plot with an improved EfficientNet version 2 (EfficientNetV2). First, independent component analysis is used to extract independent components containing spatial information from key channels of the EEG signals. These components are then converted into two-dimensional images using recurrence plot to better extract emotional features from the temporal information. Finally, the two-dimensional images are input into an improved EfficientNetV2, which incorporates a global attention mechanism and a triplet attention mechanism, and the emotion classification is output by the fully connected layer. To validate the effectiveness of the proposed method, this study conducts comparative experiments, channel selection experiments and ablation experiments based on the Shanghai Jiao Tong University Emotion Electroencephalogram Dataset (SEED). The results demonstrate that the average recognition accuracy of our method is 96.77%, which is significantly superior to existing methods, offering a novel perspective for research on EEG-based emotion recognition.
Existing classification methods for myositis ultrasound images have problems of poor classification performance or high computational cost. Motivated by this difficulty, a lightweight neural network based on a soft threshold attention mechanism is proposed to cater for a better IIMs classification. The proposed network was constructed by alternately using depthwise separable convolution (DSC) and conventional convolution (CConv). Moreover, a soft threshold attention mechanism was leveraged to enhance the extraction capabilities of key features. Compared with the current dual-branch feature fusion myositis classification network with the highest classification accuracy, the classification accuracy of the network proposed in this paper increased by 5.9%, reaching 96.1%, and its computational complexity was only 0.25% of the existing method. The obtained results support that the proposed method can provide physicians with more accurate classification results at a lower computational cost, thereby greatly assisting them in their clinical diagnosis.
Lung cancer is the most threatening tumor disease to human health. Early detection is crucial to improve the survival rate and recovery rate of lung cancer patients. Existing methods use the two-dimensional multi-view framework to learn lung nodules features and simply integrate multi-view features to achieve the classification of benign and malignant lung nodules. However, these methods suffer from the problems of not capturing the spatial features effectively and ignoring the variability of multi-views. Therefore, this paper proposes a three-dimensional (3D) multi-view convolutional neural network (MVCNN) framework. To further solve the problem of different views in the multi-view model, a 3D multi-view squeeze-and-excitation convolution neural network (MVSECNN) model is constructed by introducing the squeeze-and-excitation (SE) module in the feature fusion stage. Finally, statistical methods are used to analyze model predictions and doctor annotations. In the independent test set, the classification accuracy and sensitivity of the model were 96.04% and 98.59% respectively, which were higher than other state-of-the-art methods. The consistency score between the predictions of the model and the pathological diagnosis results was 0.948, which is significantly higher than that between the doctor annotations and the pathological diagnosis results. The methods presented in this paper can effectively learn the spatial heterogeneity of lung nodules and solve the problem of multi-view differences. At the same time, the classification of benign and malignant lung nodules can be achieved, which is of great significance for assisting doctors in clinical diagnosis.
Accurate segmentation of whole slide images is of great significance for the diagnosis of pancreatic cancer. However, developing an automatic model is challenging due to the complex content, limited samples, and high sample heterogeneity of pathological images. This paper presented a multi-tissue segmentation model for whole slide images of pancreatic cancer. We introduced an attention mechanism in building blocks, and designed a multi-task learning framework as well as proper auxiliary tasks to enhance model performance. The model was trained and tested with the pancreatic cancer pathological image dataset from Shanghai Changhai Hospital. And the data of TCGA, as an external independent validation cohort, was used for external validation. The F1 scores of the model exceeded 0.97 and 0.92 in the internal dataset and external dataset, respectively. Moreover, the generalization performance was also better than the baseline method significantly. These results demonstrate that the proposed model can accurately segment eight kinds of tissue regions in whole slide images of pancreatic cancer, which can provide reliable basis for clinical diagnosis.
The processing mechanism of the human brain for speech information is a significant source of inspiration for the study of speech enhancement technology. Attention and lateral inhibition are key mechanisms in auditory information processing that can selectively enhance specific information. Building on this, the study introduces a dual-branch U-Net that integrates lateral inhibition and feedback-driven attention mechanisms. Noisy speech signals input into the first branch of the U-Net led to the selective feedback of time-frequency units with high confidence. The generated activation layer gradients, in conjunction with the lateral inhibition mechanism, were utilized to calculate attention maps. These maps were then concatenated to the second branch of the U-Net, directing the network’s focus and achieving selective enhancement of auditory speech signals. The evaluation of the speech enhancement effect was conducted by utilising five metrics, including perceptual evaluation of speech quality. This method was compared horizontally with five other methods: Wiener, SEGAN, PHASEN, Demucs and GRN. The experimental results demonstrated that the proposed method improved speech signal enhancement capabilities in various noise scenarios by 18% to 21% compared to the baseline network across multiple performance metrics. This improvement was particularly notable in low signal-to-noise ratio conditions, where the proposed method exhibited a significant performance advantage over other methods. The speech enhancement technique based on lateral inhibition and feedback-driven attention mechanisms holds significant potential in auditory speech enhancement, making it suitable for clinical practices related to artificial cochleae and hearing aids.
The brain-computer interface (BCI) based on motor imagery electroencephalography (MI-EEG) enables direct information interaction between the human brain and external devices. In this paper, a multi-scale EEG feature extraction convolutional neural network model based on time series data enhancement is proposed for decoding MI-EEG signals. First, an EEG signals augmentation method was proposed that could increase the information content of training samples without changing the length of the time series, while retaining its original features completely. Then, multiple holistic and detailed features of the EEG data were adaptively extracted by multi-scale convolution module, and the features were fused and filtered by parallel residual module and channel attention. Finally, classification results were output by a fully connected network. The application experimental results on the BCI Competition IV 2a and 2b datasets showed that the proposed model achieved an average classification accuracy of 91.87% and 87.85% for the motor imagery task, respectively, which had high accuracy and strong robustness compared with existing baseline models. The proposed model does not require complex signals pre-processing operations and has the advantage of multi-scale feature extraction, which has high practical application value.
Motor imagery electroencephalogram (EEG) signals are non-stationary time series with a low signal-to-noise ratio. Therefore, the single-channel EEG analysis method is difficult to effectively describe the interaction characteristics between multi-channel signals. This paper proposed a deep learning network model based on the multi-channel attention mechanism. First, we performed time-frequency sparse decomposition on the pre-processed data, which enhanced the difference of time-frequency characteristics of EEG signals. Then we used the attention module to map the data in time and space so that the model could make full use of the data characteristics of different channels of EEG signals. Finally, the improved time-convolution network (TCN) was used for feature fusion and classification. The BCI competition IV-2a data set was used to verify the proposed algorithm. The experimental results showed that the proposed algorithm could effectively improve the classification accuracy of motor imagination EEG signals, which achieved an average accuracy of 83.03% for 9 subjects. Compared with the existing methods, the classification accuracy of EEG signals was improved. With the enhanced difference features between different motor imagery EEG data, the proposed method is important for the study of improving classifier performance.