KD_ConvNeXt: knowledge distillation-based image classification of lung tumor surgical specimen sections.

Introduction: Lung cancer is currently among the most prevalent and lethal cancers in the world in terms of incidence and fatality rates. In clinical practice, identifying the specific subtypes of lung cancer is essential in diagnosing and treating lung lesions. Methods: This paper aims to collect histopathological section images of lung tumor surgical specimens to construct a clinical dataset for researching and addressing the classification problem of specific subtypes of lung tumors. Our method proposes a teacher-student network architecture based on a knowledge distillation mechanism for the specific subtype classification of lung tumor histopathological section images to assist clinical applications, namely KD_ConvNeXt. The proposed approach enables the student network (ConvNeXt) to extract knowledge from the intermediate feature layers of the teacher network (Swin Transformer), improving the feature extraction and fitting capabilities of ConvNeXt. Meanwhile, Swin Transformer provides soft labels containing information about the distribution of images in various categories, making the model focused more on the information carried by types with smaller sample sizes while training. Results: This work has designed many experiments on a clinical lung tumor image dataset, and the KD_ConvNeXt achieved a superior classification accuracy of 85.64% and an F1-score of 0.7717 compared with other advanced image classification methods.

MVF-SleepNet: Multi-View Fusion Network for Sleep Stage Classification.

Sleep stage classification is of great importance in human health monitoring and disease diagnosing. Clinically, visual-inspected classifying sleep into different stages is quite time consuming and highly relies on the expertise of sleep specialists. Many automated models for sleep stage classification have been proposed in previous studies but their performances still exist a gap to the real clinical application. In this work, we propose a novel multi-view fusion network named MVF-SleepNet based on multi-modal physiological signals of electroencephalography (EEG), electrocardiography (ECG), electrooculography (EOG), and electromyography (EMG). To capture the relationship representation among multi-modal physiological signals, we construct two views of Time-frequency images (TF images) and Graph-learned graphs (GL graphs). To learn the spectral-temporal representation from sequentially timed TF images, the combination of VGG-16 and GRU networks is utilized. To learn the spatial-temporal representation from sequentially timed GL graphs, the combination of Chebyshev graph convolution and temporal convolution networks is employed. Fusing the spectral-temporal representation and spatial-temporal representation can further boost the performance of sleep stage classification. A large number of experiment results on the publicly available datasets of ISRUC-S1 and ISRUC-S3 show that the MVF-SleepNet achieves overall accuracy of 0.821, F1 score of 0.802 and Kappa of 0.768 on ISRUC-S1 dataset, and accuracy of 0.841, F1 score of 0.828 and Kappa of 0.795 on ISRUC-S3 dataset. The MVF-SleepNet achieves competitive results on both datasets of ISRUC-S1 and ISRUC-S3 for sleep stage classification compared to the state-of-the-art baselines. The source code of MVF-SleepNet is available on Github (https://github.com/YJPai65/MVF-SleepNet).

A novel EEG-based major depressive disorder detection framework with two-stage feature selection.

BACKGROUND: Major depressive disorder (MDD) is a common mental illness, characterized by persistent depression, sadness, despair, etc., troubling people's daily life and work seriously. METHODS: In this work, we present a novel automatic MDD detection framework based on EEG signals. First of all, we derive highly MDD-correlated features, calculating the ratio of extracted features from EEG signals at frequency bands between [Formula: see text] and [Formula: see text]. Then, a two-stage feature selection method named PAR is presented with the sequential combination of Pearson correlation coefficient (PCC) and recursive feature elimination (RFE), where the advantages lie in minimizing the feature searching space. Finally, we employ widely used machine learning methods of support vector machine (SVM), logistic regression (LR), and linear regression (LNR) for MDD detection with the merit of feature interpretability. RESULTS: Experiment results show that our proposed MDD detection framework achieves competitive results. The accuracy and [Formula: see text] score are up to 0.9895 and 0.9846, respectively. Meanwhile, the regression determination coefficient [Formula: see text] for MDD severity assessment is up to 0.9479. Compared with existing MDD detection methods with the best accuracy of 0.9840 and [Formula: see text] score of 0.97, our proposed framework achieves the state-of-the-art MDD detection performance. CONCLUSIONS: Development of this MDD detection framework can be potentially deployed into a medical system to aid physicians to screen out MDD patients.

ChromosomeNet: A massive dataset enabling benchmarking and building basedlines of clinical chromosome classification.

Chromosome karyotyping analysis is a vital cytogenetics technique for diagnosing genetic and congenital malformations, analyzing gestational and implantation failures, etc. Since the chromosome classification as an essential stage in chromosome karyotype analysis is a highly time-consuming, tedious, and error-prone task, which requires a large amount of manual work of experienced cytogenetics experts. Many deep learning-based methods have been proposed to address the chromosome classification issues. However, two challenges still remain in current chromosome classification methods. First, most existing methods were developed by different private datasets, making these methods difficult to compare with each other on the same base. Second, due to the absence of reproducing details of most existing methods, these methods are difficult to be applied in clinical chromosome classification applications widely. To address the above challenges in the chromosome classification issue, this work builds and publishes a massive clinical dataset. This dataset enables the benchmarking and building chromosome classification baselines suitable for different scenarios. The massive clinical dataset consists of 126,453 privacy preserving G-band chromosome instances from 2763 karyotypes of 408 individuals. To our best knowledge, it is the first work to collect, annotate, and release a publicly available clinical chromosome classification dataset whose data size scale is also over 120,000. Meanwhile, the experimental results show that the proposed dataset can boost performance of existing chromosome classification models at a varied range of degrees, with the highest accuracy improvement by 5.39 % points. Moreover, the best baseline with 99.33 % accuracy reports state-of-the-art classification performance. The clinical dataset and state-of-the-art baselines can be found at https://github.com/CloudDataLab/BenchmarkForChromosomeClassification.

Predicting Prognostic Effects of Acupuncture for Depression Using the Electroencephalogram.

Depression is considered to be a major public health problem with significant implications for individuals and society. Patients with depression can be with complementary therapies such as acupuncture. Predicting the prognostic effects of acupuncture has a big significance in helping physicians make early interventions for patients with depression and avoid malignant events. In this work, a novel framework of predicting prognostic effects of acupuncture for depression based on electroencephalogram (EEG) recordings is presented. Specifically, EEG, as a widely used measurement to evaluate the therapeutic effects of acupuncture, is utilized for predicting prognostic effects of acupuncture. Max-relevance and min-redundancy (mRMR), with merits of removing redundant information among selected features and remaining high relevance between selected features and response variable, is employed to select important lead-rhythm features extracted from EEG recordings. Then, according to the subject Hamilton Depression Rating Scale (HAMD) scores before and after acupuncture for eight weeks, the reduction rate of HAMD score is calculated as a measure of the prognostic effects of acupuncture. Finally, five widely used machine learning methods are utilized for building the predicting models of prognostic effects of acupuncture for depression. Experimental results show that nonlinear machine learning methods have better performance than linear ones on predicting prognostic effects of acupuncture using EEG recordings. Especially, the support vector machine with Gaussian kernel (SVM-RBF) can achieve the best and most stable performance using the mRMR with both evaluating criteria of FCD and FCQ for feature selection. Both mRMR-FCD and mRMR-FCQ obtain the same best performance, where the accuracy and F 1 score are 84.61% and 86.67%, respectively. Moreover, lead-rhythm features selected by mRMR-FCD and mRMR-FCQ are analyzed. The top seven selected lead-rhythm features have much higher mRMR evaluating scores, which guarantee the good predicting performance for machine learning methods to some degree. The presented framework in this work is effective in predicting the prognostic effects of acupuncture for depression. It can be integrated into an intelligent medical system and provide information on the prognostic effects of acupuncture for physicians. Informed prognostic effects of acupuncture for depression in advance and taking interventions can greatly reduce the risk of malignant events for patients with mental disorders.

CIR-Net: Automatic Classification of Human Chromosome Based on Inception-ResNet Architecture.

BACKGROUND: In medicine, karyotyping chromosomes is important for medical diagnostics, drug development, and biomedical research. Unfortunately, chromosome karyotyping is usually done by skilled cytologists manually, which requires experience, domain expertise, and considerable manual efforts. Therefore, automating the karyotyping process is a significant and meaningful task. METHOD: This paper focuses on chromosome classification because it is critical for chromosome karyotyping. In recent years, deep learning-based methods are the most promising methods for solving the tasks of chromosome classification. Although the deep learning-based Inception architecture has yielded state-of-the-art performance in the 2015 ILSVRC challenge, it has not been used in chromosome classification tasks so far. Therefore, we develop an automatic chromosome classification approach named CIR-Net based on Inception-ResNet which is an optimized version of Inception. However, the classification performance of origin Inception-ResNet on the insufficient chromosome dataset still has a lot of capacity for improvement. Further, we propose a simple but effective augmentation method called CDA for improving the performance of CIR-Net. RESULTS: The experimental results show that our proposed method achieves 95.98 percent classification accuracy on the clinical G-band chromosome dataset whose training dataset is insufficient. Moreover, the proposed augmentation method CDA improves more than 8.5 percent (from 87.46 to 95.98 percent) classification accuracy comparing to other methods. In this paper, the experimental results demonstrate that our proposed method is recent the most effective solution for solving clinical chromosome classification problems in chromosome auto-karyotyping on the condition of the insufficient training dataset. Code and Dataset are available at https://github.com/CloudDataLab/CIR-Net.

A Clinical Dataset and Various Baselines for Chromosome Instance Segmentation.

BACKGROUND: In medicine, chromosome karyotyping analysis plays a crucial role in prenatal diagnosis for diagnosing whether a fetus has severe defects or genetic diseases. However, chromosome instance segmentation is the most critical obstacle to automatic chromosome karyotyping analysis due to the complicated morphological characteristics of chromosome clusters, restricting chromosome karyotyping analysis to highly depend on skilled clinical analysts. METHOD: In this paper, we build a clinical dataset and propose multiple segmentation baselines to tackle the chromosome instance segmentation problem of various overlapping and touching chromosome clusters. First, we construct a clinical dataset for deep learning-based chromosome instance segmentation models by collecting and annotating 1,655 privacy-removal chromosome clusters. After that, we design a chromosome instance labeled dataset augmentation (CILA) algorithm for the clinical dataset to improve the generalization performance of deep learning-based models. Last, we propose a chromosome instance segmentation framework and implement multiple baselines for the proposed framework based on various instance segmentation models. RESULTS AND CONCLUSIONS: Experiments evaluated on the clinical dataset show that the best baseline of the proposed framework based on the Mask-RCNN model yields an outstanding result with 77% mAP, 97.5% AP50, and 95.5% AP75 segmentation precision, and 95.38% accuracy, which exceeds results reported in current chromosome instance segmentation methods. The quantitative evaluation results demonstrate the effectiveness and advancement of the proposed method for the chromosome instance segmentation problem. The experimental code and privacy-removal clinical dataset can be found at Github.

EEG-based Major Depressive Disorder Detection Using Data Mining Techniques.

Major depressive disorder (MDD) is a common mental illness characterized by a persistent feeling of low mood, sadness, fatigue, despair, etc.. In a serious case, patients with MDD may have suicidal thoughts or even suicidal behaviors. In clinical practice, a widely used method of MDD detection is based on a professional rating scale. However, the scale-based diagnostic method is highly subjective, and requires a professional assessment from a trained staff. In this work, 92 participants were recruited to collect EEG signals in the Shenzhen Traditional Chinese Medicine Hospital, assessing MDD severity with the HAMD-17 rating scale by a trained physician. Two data mining methods of logistic regression (LR) and support vector machine (SVM) with derived EEG-based beta-alpha-ratio features, namely LR-DF and SVM-DF, are employed to screen out patients with MDD. Experimental results show that the presented the LR-DF and SVM-DF achieved F 1 scores of 0:76 0:30 and 0:92 0:18, respectively, which have obvious superiority to the LR and SVM without derived EEG-based beta-alpha-ratio features.

MDD-TSVM: A novel semisupervised-based method for major depressive disorder detection using electroencephalogram signals.

Major depressive disorder (MDD) is a common mental illness characterized by persistent feeling of depressed mood and loss of interest. It would cause, in a severe case, suicide behaviors. In clinical settings, automatic MDD detection is mainly based on electroencephalogram (EEG) signals with supervised learning techniques. However, supervised-based MDD detection methods encounter two ineviTable bottlenecks: firstly, such methods rely heavily on an EEG training dataset with MDD labels annotated by a physical therapist, leading to subjectivity and high cost; secondly, most of EEG signals are unlabeled in a real scenario. In this paper, a novel semisupervised-based MDD detection method named MDD-TSVM is presented. Specifically, the MDD-TSVM utilizes the semisupervised method of transductive support vector machine (TSVM) as its backbone, further dividing the unlabeled penalty item of the TSVM objective function into two pseudo-labeled penalty items with or without MDD. By such improvement, the MDD-SVM can make full use of labeled and unlabeled datasets as well as alleviate the class imbalance problem. Experiment results showed that our proposed MDD-TSVM achieved F1 score of 0.85 ± 0.05 and accuracy of 0.89 ± 0.03 on identifying MDD patients, which is superior to the state-of-the-art methods.

A novel approach to diagnose sleep apnea using enhanced frequency extraction network.

Sleep apnea-hypopnea syndrome (SAHS), as a widespread respiratory sleep disorder, if left untreated, can lead to a series of pathological changes. By using Polysomnography (PSG), traditional SAHS diagnosis tends to be complex and costly. Nasal airflow (NA) is the most direct reflection of the severity of SAHS. Therefore, we try to take advantage of NA signals that can be easily recorded by wearable devices. In this paper, we present an automatic detection approach of SAH events based on single-channel signal. Through this approach, an enhanced frequency extraction network is designed, which factorizes the mixed feature maps by their frequencies. And the spatial resolution of low-frequency components is reduced so as to save spending. Besides, in our research, the vanilla convolution block of the high-frequency components are replaced by residual blocks and smaller groups of filters with bigger size kernels. And we use the spatial attention module to facilitate feature extraction. Compared with state-of-the-art networks in this field, the promising results reveal that the proposed network for SAH events multiclass classification shows outstanding performance with accuracy of 91.23%, sensitivity of 90.81% and specificity of 90.59%. Thus, we believe that our approach, as a low-cost and high-efficiency solution, shows a great potential for detecting SAH events.

Deep Learning for Diagnosis and Classification of Obstructive Sleep Apnea: A Nasal Airflow-Based Multi-Resolution Residual Network.

PURPOSE: This study evaluated a novel approach for diagnosis and classification of obstructive sleep apnea (OSA), called Obstructive Sleep Apnea Smart System (OSASS), using residual networks and single-channel nasal pressure airflow signals. METHODS: Data were collected from the sleep center of the First Affiliated Hospital, Sun Yat-sen University, and the Integrative Department of Guangdong Province Traditional Chinese Medical Hospital. We developed a new model called the multi-resolution residual network (Mr-ResNet) based on a residual network to detect nasal pressure airflow signals recorded by polysomnography (PSG) automatically. The performance of the model was assessed by its sensitivity, specificity, accuracy, and F1-score. We built OSASS based on Mr-ResNet to estimate the apnea‒hypopnea index (AHI) and to classify the severity of OSA, and compared the agreement between OSASS output and the registered polysomnographic technologist (RPSGT) score, assessed by two technologists. RESULTS: In the primary test set, the sensitivity, specificity, accuracy, and F1-score of Mr-ResNet were 90.8%, 90.5%, 91.2%, and 90.5%, respectively. In the independent test set, the Spearman correlation for AHI between OSASS and the RPSGT score determined by two technologists was 0.94 (p < 0.001) and 0.96 (p < 0.001), respectively. Cohen's Kappa scores for classification between OSASS and the two technologists' scores were 0.81 and 0.84, respectively. CONCLUSION: Our results indicated that OSASS can automatically diagnose and classify OSA using signals from a single-channel nasal pressure airflow, which is consistent with polysomnographic technologists' findings. Thus, OSASS holds promise for clinical application.

A novel chromosome cluster types identification method using ResNeXt WSL model.

Chromosome karyotyping analysis plays a crucial role in prenatal diagnosis for diagnosing whether a fetus has severe defects or genetic diseases. However, due to the complicated morphological characteristics of various types of chromosome clusters, chromosome instance segmentation is the most challenging stage of chromosome karyotyping analysis, leading chromosome karyotyping analysis to highly dependent on skilled clinical analysts. Since most of the chromosome instance segmentation efforts are currently devoted to segmenting chromosome instances from different types of chromosome clusters, type identification of chromosome clusters is a vital anterior task for chromosome instance segmentation. Firstly, this paper proposes an automatic approach for chromosome cluster identification using recent transfer learning techniques. The proposed framework is based on ResNeXt weakly-supervised learning (WSL) pre-trained backbone and a task-specific network header. Secondly, this paper proposes a fast training methodology that tunes our framework from coarse-to-fine gradually. Extensive evaluations on a clinical dataset consisting of 6592 clinical chromosome samples show that the proposed framework achieves 94.09%accuracy, 92.79%sensitivity, and 98.03%specificity. Such performance is superior to the best baseline model that we obtain 92.17%accuracy, 89.1%sensitivity, and 97.42%specificity. To foster research and application in the chromosome cluster type identification, we make our clinical dataset and code available via GitHub.

Clustering by Errors: A Self-Organized Multitask Learning Method for Acoustic Scene Classification.

Acoustic scene classification (ASC) tries to inference information about the environment using audio segments. The inter-class similarity is a significant issue in ASC as acoustic scenes with different labels may sound quite similar. In this paper, the similarity relations amongst scenes are correlated with the classification error. A class hierarchy construction method by using classification error is then proposed and integrated into a multitask learning framework. The experiments have shown that the proposed multitask learning method improves the performance of ASC. On the TUT Acoustic Scene 2017 dataset, we obtain the ensemble fine-grained accuracy of 81.4%, which is better than the state-of-the-art. By using multitask learning, the basic Convolutional Neural Network (CNN) model can be improved by about 2.0 to 3.5 percent according to different spectrograms. The coarse category accuracies (for two to six super-classes) range from 77.0% to 96.2% by single models. On the revised version of the LITIS Rouen dataset, we achieve the ensemble fine-grained accuracy of 83.9%. The multitask learning models obtain an improvement of 1.6% to 1.8% compared to their basic models. The coarse category accuracies range from 94.9% to 97.9% for two to six super-classes with single models.

Developing a Machine Learning System for Identification of Severe Hand, Foot, and Mouth Disease from Electronic Medical Record Data.

Children of severe hand, foot, and mouth disease (HFMD) often present with same clinical features as those of mild HFMD during the early stage, yet later deteriorate rapidly with a fulminant disease course. Our goal was to: (1) develop a machine learning system to automatically identify cases with high risk of severe HFMD at the time of admission; (2) compare the effectiveness of the new system with the existing risk scoring system. Data on 2,532 HFMD children admitted between March 2012 and July 2015, were collected retrospectively from a medical center in China. By applying a holdout strategy and a 10-fold cross validation method, we developed four models with the random forest algorithm using different variable sets. The prediction system HFMD-RF based on the model of 16 variables from both the structured and unstructured data, achieved 0.824 sensitivity, 0.931 specificity, 0.916 accuracy, and 0.916 area under the curve in the independent test set. Most remarkably, HFMD-RF offers significant gains with respect to the commonly used pediatric critical illness score in clinical practice. As all the selected risk factors can be easily obtained, HFMD-RF might prove to be useful for reductions in mortality and complications of severe HFMD.

A Two-Phase Space Resection Model for Accurate Topographic Reconstruction from Lunar Imagery with PushbroomScanners.

Exterior orientation parameters' (EOP) estimation using space resection plays an important role in topographic reconstruction for push broom scanners. However, existing models of space resection are highly sensitive to errors in data. Unfortunately, for lunar imagery, the altitude data at the ground control points (GCPs) for space resection are error-prone. Thus, existing models fail to produce reliable EOPs. Motivated by a finding that for push broom scanners, angular rotations of EOPs can be estimated independent of the altitude data and only involving the geographic data at the GCPs, which are already provided, hence, we divide the modeling of space resection into two phases. Firstly, we estimate the angular rotations based on the reliable geographic data using our proposed mathematical model. Then, with the accurate angular rotations, the collinear equations for space resection are simplified into a linear problem, and the global optimal solution for the spatial position of EOPs can always be achieved. Moreover, a certainty term is integrated to penalize the unreliable altitude data for increasing the error tolerance. Experimental results evidence that our model can obtain more accurate EOPs and topographic maps not only for the simulated data, but also for the real data from Chang'E-1, compared to the existing space resection model.