SI-ViT: Shuffle instance-based Vision Transformer for pancreatic cancer ROSE image classification.

BACKGROUND AND OBJECTIVE: The rapid on-site evaluation (ROSE) technique improves pancreatic cancer diagnosis by enabling immediate analysis of fast-stained cytopathological images. Automating ROSE classification could not only reduce the burden on pathologists but also broaden the application of this increasingly popular technique. However, this approach faces substantial challenges due to complex perturbations in color distribution, brightness, and contrast, which are influenced by various staining environments and devices. Additionally, the pronounced variability in cancerous patterns across samples further complicates classification, underscoring the difficulty in precisely identifying local cells and establishing their global relationships. METHODS: To address these challenges, we propose an instance-aware approach that enhances the Vision Transformer with a novel shuffle instance strategy (SI-ViT). Our approach presents a shuffle step to generate bags of shuffled instances and corresponding bag-level soft-labels, allowing the model to understand relationships and distributions beyond the limited original distributions. Simultaneously, combined with an un-shuffle step, the traditional ViT can model the relationships corresponding to the sample labels. This dual-step approach helps the model to focus on inner-sample and cross-sample instance relationships, making it potent in extracting diverse image patterns and reducing complicated perturbations. RESULTS: Compared to state-of-the-art methods, significant improvements in ROSE classification have been achieved. Aiming for interpretability, equipped with instance shuffling, SI-ViT yields precise attention regions that identifying cancer and normal cells in various scenarios. Additionally, the approach shows excellent potential in pathological image analysis through generalization validation on other datasets. CONCLUSIONS: By proposing instance relationship modeling through shuffling, we introduce a new insight in pathological image analysis. The significant improvements in ROSE classification leads to protential AI-on-site applications in pancreatic cancer diagnosis. The code and results are publicly available at https://github.com/sagizty/MIL-SI.

PPG-based Continuous BP Waveform Estimation Using Polarized Attention-guided Conditional Adversarial Learning Model.

The blood pressure (BP) waveform is a vital source of physiological and pathological information concerning the cardiovascular system. This study proposes a novel attention-guided conditional generative adversarial network (cGAN), named PPG2BP-cGAN, to estimate BP waveforms based on photoplethysmography (PPG) signals. The proposed model comprises a generator and a discriminator. Specifically, the UNet3+-based generator integrates a full-scale skip connection structure with a modified polarized self-attention module based on a spatial-temporal attention mechanism. Additionally, its discriminator comprises PatchGAN, which augments the discriminative power of the generated BP waveform by increasing the perceptual field through fully convolutional layers. We demonstrate the superior BP waveform prediction performance of our proposed method compared to state-of-the-art (SOTA) techniques on two independent datasets. Our approach first pre-trained on a dataset containing 683 subjects and then tested on a public dataset. Experimental results from the Multi-parameter Intelligent Monitoring in Intensive Care dataset show that the proposed method achieves a root mean square error of 3.54, mean absolute error of 2.86, and Pearson coefficient of 0.99 for BP waveform estimation. Furthermore, the estimation errors (mean error ± standard deviation error) for systolic BP and diastolic BP are 0.72 ± 4.34 mmHg and 0.41 ± 2.48 mmHg, respectively, meeting the American Association for the Advancement of Medical Instrumentation standard. Our approach exhibits significant superiority over SOTA techniques on independent datasets, thus highlighting its potential for future applications in continuous cuffless BP waveform measurement.

A Novel Feature Engineering Method Based on Latent Representation Learning for Radiomics: Application in NSCLC Subtype Classification.

Radiomics refers to the high-throughput extraction of quantitative features from medical images, and is widely used to construct machine learning models for the prediction of clinical outcomes, while feature engineering is the most important work in radiomics. However, current feature engineering methods fail to fully and effectively utilize the heterogeneity of features when dealing with different kinds of radiomics features. In this work, latent representation learning is first presented as a novel feature engineering approach to reconstruct a set of latent space features from original shape, intensity and texture features. This proposed method projects features into a subspace called latent space, in which the latent space features are obtained by minimizing a unique hybrid loss function including a clustering-like loss and a reconstruction loss. The former one ensures the separability among each class while the latter one narrows the gap between the original features and latent space features. Experiments were performed on a multi-center non-small cell lung cancer (NSCLC) subtype classification dataset from 8 international open databases. Results showed that compared with four traditional feature engineering methods (baseline, PCA, Lasso and L2,1-norm minimization), latent representation learning could significantly improve the classification performance of various machine learning classifiers on the independent test set (all p<0.001). Further on two additional test sets, latent representation learning also showed a significant improvement in generalization performance. Our research shows that latent representation learning is a more effective feature engineering method, which has the potential to be used as a general technology in a wide range of radiomics researches.

MSHT: Multi-stage Hybrid Transformer for the ROSE Image Analysis of Pancreatic Cancer.

Pancreatic cancer is one of the most malignant cancers with high mortality. The rapid on-site evaluation (ROSE) technique can significantly accelerate the diagnostic workflow of pancreatic cancer by immediately analyzing the fast-stained cytopathological images with on-site pathologists. However, the broader expansion of ROSE diagnosis has been hindered by the shortage of experienced pathologists. Deep learning has great potential for the automatic classification of ROSE images in diagnosis. But it is challenging to model the complicated local and global image features. The traditional convolutional neural network (CNN) structure can effectively extract spatial features, while it tends to ignore global features when the prominent local features are misleading. In contrast, the Transformer structure has excellent advantages in capturing global features and long-range relations, while it has limited ability in utilizing local features. We propose a multi-stage hybrid Transformer (MSHT) to combine the strengths of both, where a CNN backbone robustly extracts multi-stage local features at different scales as the attention guidance, and a Transformer encodes them for sophisticated global modeling. Going beyond the strength of each single method, the MSHT can simultaneously enhance the Transformer global modeling ability with the local guidance from CNN features. To evaluate the method in this unexplored field, a dataset of 4240 ROSE images is collected where MSHT achieves 95.68% in classification accuracy with more accurate attention regions. The distinctively superior results compared to the state-of-the-art models make MSHT extremely promising for cytopathological image analysis. The codes and records are available at: https://github.com/sagizty/ Multi-Stage-Hybrid-Transformer.

Radiomics feature analysis and model research for predicting histopathological subtypes of non-small cell lung cancer on CT images: A multi-dataset study.

PURPOSE: Classifying the subtypes of non-small cell lung cancer (NSCLC) is essential for clinically adopting optimal treatment strategies and improving clinical outcomes, but the histological subtypes are confirmed by invasive biopsy or post-operative examination at present. Based on multi-center data, this study aimed to analyze the importance of extracted CT radiomics features and develop the model with good generalization performance for precisely distinguishing major NSCLC subtypes: adenocarcinoma (ADC) and squamous cell carcinoma (SCC). METHODS: We collected a multi-center CT dataset with 868 patients from eight international databases on the cancer imaging archive (TCIA). Among them, patients from five databases were mixed and split to training and test sets (560:140). The remaining three databases were used as independent test sets: TCGA set (n = 97) and lung3 set (n = 71). A total of 1409 features containing shape, intensity, and texture information were extracted from tumor volume of interest (VOI), then the ℓ2,1 -norm minimization was used for feature selection and the importance of selected features was analyzed. Next, the prediction and generalization performance of 130 radiomics models (10 common algorithms and 120 heterogeneous ensemble combinations) were compared by the average AUC value on three test sets. Finally, predictive results of the optimal model were shown. RESULTS: After feature selection, 401 features were obtained. Features of intensity, texture GLCM, GLRLM, and GLSZM had higher classification weight coefficients than other features (shape, texture GLDM, and NGTDM), and the filtered image features exhibited significant importance than original image features (p-value = 0.0210). Moreover, five ensemble learning algorithms (Bagging, AdaBoost, RF, XGBoost, GBDT) had better generalization performance (p-value = 0.00418) than other non-ensemble algorithms (MLP, LR, GNB, SVM, KNN). The Bagging-AdaBoost-SVM model had the highest AUC value (0.815 ± 0.010) on three test sets. It obtained AUC values of 0.819, 0.823, and 0.804 on test set, TCGA set and lung3 set, respectively. CONCLUSION: Our multi-dataset study showed that intensity features, texture features (GLCM, GLRLM, and GLSZM) and filtered image features were more important for distinguishing ADCs from SCCs. The method of ensemble learning can improve the prediction and generalization performance on the complicated multi-center data. The Bagging-AdaBoost-SVM model had the strongest generalization performance, and it showed promising clinical value for non-invasively predicting the histopathological subtypes of NSCLC.

Robust reconstruction of fluorescence molecular tomography based on adaptive adversarial learning strategy.

Objective.Fluorescence molecular tomography (FMT) is a promising molecular imaging modality for quantifying the three-dimensional (3D) distribution of tumor probes in small animals. However, traditional deep learning reconstruction methods that aim to minimize the mean squared error (MSE) and iterative regularization algorithms that rely on optimal parameters are typically influenced by strong noise, resulting in poor FMT reconstruction robustness.Approach.In this letter, we propose an adaptive adversarial learning strategy (3D-UR-WGAN) to achieve robust FMT reconstructions. Unlike the pixel-based MSE criterion in traditional CNNs or the regularization strategy in iterative solving schemes, the reconstruction strategy can greatly facilitate the performance of the network models through alternating loop training of the generator and the discriminator. Second, the reconstruction strategy combines the adversarial loss in GANs with the L1 loss to significantly enhance the robustness and preserve image details and textual information.Main results.Both numerical simulations and physical phantom experiments demonstrate that the 3D-UR-WGAN method can adaptively eliminate the effects of different noise levels on the reconstruction results, resulting in robust reconstructed images with reduced artifacts and enhanced image contrast. Compared with the state-of-the-art methods, the proposed method achieves better reconstruction performance in terms of target shape recovery and localization accuracy.Significance.This adaptive adversarial learning reconstruction strategy can provide a possible paradigm for robust reconstruction in complex environments, and also has great potential to provide an alternative solution for solving the problem of poor robustness encountered in other optical imaging modalities such as diffuse optical tomography, bioluminescence imaging, and Cherenkov luminescence imaging.

KD-Informer: A Cuff-Less Continuous Blood Pressure Waveform Estimation Approach Based on Single Photoplethysmography.

Ambulatory blood pressure (BP) monitoring plays a critical role in the early prevention and diagnosis of cardiovascular diseases. However, cuff-based inflatable devices cannot be used for continuous BP monitoring, while pulse transit time or multi-parameter-based methods require more bioelectrodes to acquire electrocardiogram signals. Thus, estimating the BP waveforms only based on photoplethysmography (PPG) signals for continuous BP monitoring has essential clinical values. Nevertheless, extracting useful features from raw PPG signals for fine-grained BP waveform estimation is challenging due to the physiological variation and noise interference. For single PPG analysis utilizing deep learning methods, the previous works depend mainly on stacked convolution operation, which ignores the underlying complementary time-dependent information. Thus, this work presents a novel Transformer-based method with knowledge distillation (KD-Informer) for BP waveform estimation. Meanwhile, we integrate the prior information of PPG patterns, selected by a novel backward elimination algorithm, into the knowledge transfer branch of the KD-Informer. With these strategies, the model can effectively capture the discriminative features through a lightweight architecture during the learning process. Then, we further adopt an effective transfer learning technique to demonstrate the excellent generalization capability of the proposed model using two independent multicenter datasets. Specifically, we first fine-tuned the KD-Informer with a large and high-quality dataset (Mindray dataset) and then transferred the pre-trained model to the target domain (MIMIC dataset). The experimental test results on the MIMIC dataset showed that the KD-Informer exhibited an estimation error of 0.02 ± 5.93 mmHg for systolic BP (SBP) and 0.01 ± 3.87 mmHg for diastolic BP (DBP), which complied with the association for the advancement of medical instrumentation (AAMI) standard. These results demonstrate that the KD-Informer has high reliability and elegant robustness to measure continuous BP waveforms.

D2-RecST: Dual-domain joint reconstruction strategy for fluorescence molecular tomography based on image domain and perception domain.

BACKGROUND AND OBJECTIVE: Fluorescence molecular tomography (FMT) is a promising molecular imaging modality for quantifying the three-dimensional (3D) distribution of fluorescent probes in small animals. Over the past few years, learning-based FMT reconstruction methods have achieved promising results. However, these methods typically attempt to minimize the mean-squared error (MSE) between the reconstructed image and the ground truth. Although signal-to-noise ratios (SNRs) are improved, they are susceptible to non-uniform artifacts and loss of structural detail, making it extremely challenging to obtain accurate and robust FMT reconstructions under noisy measurements. METHODS: We propose a novel dual-domain joint strategy based on the image domain and perception domain for accurate and robust FMT reconstruction. First, we formulate an explicit adversarial learning strategy in the image domain, which greatly facilitates training and optimization through two enhanced networks to improve anti-noise ability. Besides, we introduce a novel transfer learning strategy in the perceptual domain to optimize edge details by providing perceptual priors for fluorescent targets. Collectively, the proposed dual-domain joint reconstruction strategy can significantly eliminate the non-uniform artifacts and effectively preserve the structural edge details. RESULTS: Both numerical simulations and in vivo mouse experiments demonstrate that the proposed method markedly outperforms traditional and cutting-edge methods in terms of positioning accuracy, image contrast, robustness, and target morphological recovery. CONCLUSIONS: The proposed method achieves the best reconstruction performance and has great potential to facilitate precise localization and 3D visualization of tumors in in vivo animal experiments.

Multi-branch attention prior based parameterized generative adversarial network for fast and accurate limited-projection reconstruction in fluorescence molecular tomography.

Limited-projection fluorescence molecular tomography (FMT) allows rapid reconstruction of the three-dimensional (3D) distribution of fluorescent targets within a shorter data acquisition time. However, the limited-projection FMT is severely ill-posed and ill-conditioned due to insufficient fluorescence measurements and the strong scattering properties of photons in biological tissues. Previously, regularization-based methods, combined with the sparse distribution of fluorescent sources, have been commonly used to alleviate the severe ill-posed nature of the limited-projection FMT. Due to the complex iterative computations, time-consuming solution procedures, and less stable reconstruction results, the limited-projection FMT remains an intractable challenge for achieving fast and accurate reconstructions. In this work, we completely discard the previous iterative solving-based reconstruction themes and propose multi-branch attention prior based parameterized generative adversarial network (MAP-PGAN) to achieve fast and accurate limited-projection FMT reconstruction. Firstly, the multi-branch attention can provide parameterized weighted sparse prior information for fluorescent sources, enabling MAP-PGAN to effectively mitigate the ill-posedness and significantly improve the reconstruction accuracy of limited-projection FMT. Secondly, since the end-to-end direct reconstruction strategy is adopted, the complex iterative computation process in traditional regularization algorithms can be avoided, thus greatly accelerating the 3D visualization process. The numerical simulation results show that the proposed MAP-PGAN method outperforms the state-of-the-art methods in terms of localization accuracy and morphological recovery. Meanwhile, the reconstruction time is only about 0.18s, which is about 100 to 1000 times faster than the conventional iteration-based regularization algorithms. The reconstruction results from the physical phantoms and in vivo experiments further demonstrate the feasibility and practicality of the MAP-PGAN method in achieving fast and accurate limited-projection FMT reconstruction.

Prior Attention Network for Multi-Lesion Segmentation in Medical Images.

The accurate segmentation of multiple types of lesions from adjacent tissues in medical images is significant in clinical practice. Convolutional neural networks (CNNs) based on the coarse-to-fine strategy have been widely used in this field. However, multi-lesion segmentation remains to be challenging due to the uncertainty in size, contrast, and high interclass similarity of tissues. In addition, the commonly adopted cascaded strategy is rather demanding in terms of hardware, which limits the potential of clinical deployment. To address the problems above, we propose a novel Prior Attention Network (PANet) that follows the coarse-to-fine strategy to perform multi-lesion segmentation in medical images. The proposed network achieves the two steps of segmentation in a single network by inserting a lesion-related spatial attention mechanism in the network. Further, we also propose the intermediate supervision strategy for generating lesion-related attention to acquire the regions of interest (ROIs), which accelerates the convergence and obviously improves the segmentation performance. We have investigated the proposed segmentation framework in two applications: 2D segmentation of multiple lung infections in lung CT slices and 3D segmentation of multiple lesions in brain MRIs. Experimental results show that in both 2D and 3D segmentation tasks our proposed network achieves better performance with less computational cost compared with cascaded networks. The proposed network can be regarded as a universal solution to multi-lesion segmentation in both 2D and 3D tasks. The source code is available at https://github.com/hsiangyuzhao/PANet.

In vivo accurate detection of the liver tumor with pharmacokinetic parametric images from dynamic fluorescence molecular tomography.

SIGNIFICANCE: Pharmacokinetic parametric images in dynamic fluorescence molecular tomography (FMT) can describe three-dimensional (3D) physiological and pathological information inside biological tissues, potentially providing quantitative assessment tools for biological research and drug development. AIM: In vivo imaging of the liver tumor with pharmacokinetic parametric images from dynamic FMT based on the differences in metabolic properties of indocyanine green (ICG) between normal liver cells and tumor liver cells inside biological tissues. APPROACH: First, an orthotopic liver tumor mouse model was constructed. Then, with the help of the FMT/computer tomography (CT) dual-modality imaging system and the direct reconstruction algorithm, 3D imaging of liver metabolic parameters in nude mice was achieved to distinguish liver tumors from normal tissues. Finally, pharmacokinetic parametric imaging results were validated against in vitro anatomical results. RESULTS: This letter demonstrates the ability of dynamic FMT to monitor the pharmacokinetic delivery of the fluorescent dye ICG in vivo, thus, enabling the distinction between normal and tumor tissues based on the pharmacokinetic parametric images derived from dynamic FMT. CONCLUSIONS: Compared with CT structural imaging technology, dynamic FMT combined with compartmental modeling as an analytical method can obtain quantitative images of pharmacokinetic parameters, thus providing a more powerful research tool for organ function assessment, disease diagnosis and new drug development.

Prediction of EGFR Mutation Status in Non-Small Cell Lung Cancer Based on Ensemble Learning.

Objectives: We aimed to identify whether ensemble learning can improve the performance of the epidermal growth factor receptor (EGFR) mutation status predicting model. Methods: We retrospectively collected 168 patients with non-small cell lung cancer (NSCLC), who underwent both computed tomography (CT) examination and EGFR test. Using the radiomics features extracted from the CT images, an ensemble model was established with four individual classifiers: logistic regression (LR), support vector machine (SVM), random forest (RF), and extreme gradient boosting (XGBoost). The synthetic minority oversampling technique (SMOTE) was also used to decrease the influence of data imbalance. The performances of the predicting model were evaluated using the area under the curve (AUC). Results: Based on the 26 radiomics features after feature selection, the SVM performed best (AUCs of 0.8634 and 0.7885 on the training and test sets, respectively) among four individual classifiers. The ensemble model of RF, XGBoost, and LR achieved the best performance (AUCs of 0.8465 and 0.8654 on the training and test sets, respectively). Conclusion: Ensemble learning can improve the model performance in predicting the EGFR mutation status of patients with NSCLC, showing potential value in clinical practice.

A multi-classification model for non-small cell lung cancer subtypes based on independent subtask learning.

PURPOSE: The non-small cell lung cancer (NSCLC) can be divided into adenocarcinoma (ADC), squamous cell carcinoma (SCC), large cell carcinoma (LCC), and not otherwise specified (NOS), which is crucial for clinical decision-making. However, current related researches are rare for the complex multi-classification of NSCLC, mainly due to the serious data imbalance, the difficulty to unify the feature space, and the complicated decision boundary among multiple subtypes. The machine learning method of traditional "one-vs-one" (OVO) is difficult to solve these problems and achieve good results. METHODS: To this end, we propose a novel independent subtask learning (ISTL) method to better carry out the multi-classification task. Specifically, it includes four pertinent strategies: (1) independent data expansion; (2) independent feature selection (IFS); (3) independent model construction; and (4) a novel voting strategy: majority voting combined with Bayesian prior. We performed experiments using 1036 CT scans (ADC:SCC:LCC:NOS = 600:268:105:63) collected from eight international databases, and the data acquisition was highly complex and diverse. RESULTS: The experimental results showed that the ISTL method obtained an accuracy of 0.812 on the independent test cohort, which significantly improved the performance of multi-classification compared with the traditional OVO-support vector machine (0.691) and OVO-random forest (0.710) models. After the IFS, six selected feature sets of six binary tasks are obviously different, indicating that the ISTL method has better interpretability to distinguish the multiple NSCLC subtypes. The results of a further auxiliary contrast experiment showed that four pertinent strategies were all effective. CONCLUSION: Our work indicates that the ISTL method can effectively perform multi-classification of NSCLC subtypes with better interpretability for clinical computer-aided detection and has the potential to be applied in a wide range of multi-classification studies.

A Multi-Classification Model for Predicting the Invasiveness of Lung Adenocarcinoma Presenting as Pure Ground-Glass Nodules.

Objectives: To establish a multi-classification model for precisely predicting the invasiveness (pre-invasive adenocarcinoma, PIA; minimally invasive adenocarcinoma, MIA; invasive adenocarcinoma, IAC) of lung adenocarcinoma manifesting as pure ground-glass nodules (pGGNs). Methods: By the inclusion and exclusion criteria, this retrospective study enrolled 346 patients (female, 297, and male, 49; age, 55.79 ± 10.53 (24-83)) presenting as pGGNs from 1292 consecutive patients with pathologically confirmed lung adenocarcinoma. A total of 27 clinical were collected and 1409 radiomics features were extracted by PyRadiomics package on python. After feature selection with L2,1-norm minimization, logistic regression (LR), extra w(ET) and gradient boosting decision tree (GBDT) were used to construct the three-classification model. Then, an ensemble model of the three algorithms based on model ensemble strategy was established to further improve the classification performance. Results: After feature selection, a hybrid of 166 features consisting of 1 clinical (short-axis diameter, ranked 27th) and 165 radiomics (4 shape, 71 intensity and 90 texture) features were selected. The three most important features are wavelet-HLL_firstorder_Minimum, wavelet-HLL_ngtdm_Busyness and square_firstorder_Kurtosis. The hybrid-ensemble model based on hybrid clinical-radiomics features and the ensemble strategy showed more accurate predictive performance than other models (hybrid-LR, hybrid-ET, hybrid-GBDT, clinical-ensemble and radiomics-ensemble). On the training set and test set, the model can obtain the accuracy values of 0.918 ± 0.022 and 0.841, and its F1-scores respectively were 0.917 ± 0.024 and 0.824. Conclusion: The multi-classification of invasive pGGNs can be precisely predicted by our proposed hybrid-ensemble model to assist patients in the early diagnosis of lung adenocarcinoma and prognosis.

A review of advances in imaging methodology in fluorescence molecular tomography.

Objective.Fluorescence molecular tomography (FMT) is a promising non-invasive optical molecular imaging technology with strong specificity and sensitivity that has great potential for preclinical and clinical studies in tumor diagnosis, drug development and therapeutic evaluation. However, the strong scattering of photons and insufficient surface measurements make it very challenging to improve the quality of FMT image reconstruction and its practical application for early tumor detection. Therefore, continuous efforts have been made to explore more effective approaches or solutions in the pursuit of high-quality FMT reconstructions.Approach.This review takes a comprehensive overview of advances in imaging methodology for FMT, mainly focusing on two critical issues in FMT reconstructions: improving the accuracy of solving the forward physical model and mitigating the ill-posed nature of the inverse problem from a methodological point of view. More importantly, numerous impressive and practical strategies and methods for improving the quality of FMT reconstruction are summarized. Notably, deep learning methods are discussed in detail to illustrate their advantages in promoting the imaging performance of FMT thanks to large datasets, the emergence of optimized algorithms and the application of innovative networks.Main results.The results demonstrate that the imaging quality of FMT can be effectively promoted by improving the accuracy of optical parameter modeling, combined with prior knowledge, and reducing dimensionality. In addition, the traditional regularization-based methods and deep neural network-based methods, especially end-to-end deep networks, can enormously alleviate the ill-posedness of the inverse problem and improve the quality of FMT image reconstruction.Significance.This review aims to illustrate a variety of effective and practical methods for the reconstruction of FMT images that may benefit future research. Furthermore, it may provide some valuable research ideas and directions for FMT in the future, and could promote, to a certain extent, the development of FMT and other methods of optical tomography.

Global hybrid multi-scale convolutional network for accurate and robust detection of atrial fibrillation using single-lead ECG recordings.

BACKGROUND AND OBJECTIVE: Atrial fibrillation (AF) is the most common persistent cardiac arrhythmia in clinical practice, and its accurate screening is of great significance to avoid cardiovascular diseases (CVDs). Electrocardiogram (ECG) is considered to be the most commonly used technique for detecting AF abnormalities. However, previous ECG-based deep learning algorithms did not take into account the complementary nature of inter-layer information, which may lead to insufficient AF screening. This study reports the first attempt to use hybrid multi-scale information in a global space for accurate and robust AF detection. METHODS: We propose a novel deep learning classification method, namely, global hybrid multi-scale convolutional neural network (i.e., GH-MS-CNN), to implement binary classification for AF detection. Unlike previous deep learning methods in AF detection, an ingenious hybrid multi-scale convolution (HMSC) module, for the advantage of automatically aggregating different types of complementary inter-layer multi-scale features in the global space, is introduced into all dense blocks of the GH-MS-CNN model to implement sufficient feature extraction, and achieve much better overall classification performance. RESULTS: The proposed GH-MS-CNN method has been fully validated on the CPSC 2018 database and tested on the independent PhysioNet 2017 database. The experimental results show that the global and hybrid multi-scale information has tremendous advantages over local and single-type multi-scale information in AF screening. Furthermore, the proposed GH-MS-CNN method outperforms the state-of-the-art methods and achieves the best classification performance with an accuracy of 0.9984, a precision of 0.9989, a sensitivity of 0.9965, a specificity of 0.9998 and an F1 score of 0.9954. In addition, the proposed method has achieved comparable and considerable generalization capability on the PhysioNet 2017 database. CONCLUSIONS: The proposed GH-MS-CNN method has promising capabilities and great advantages in accurate and robust AF detection. It is assumed that this research has made significant improvements in AF screening and has great potential for long-term monitoring of wearable devices.