The role of eye movement signals in non-invasive brain-computer interface typing system.

Brain-Computer Interfaces (BCIs) have shown great potential in providing communication and control for individuals with severe motor disabilities. However, traditional BCIs that rely on electroencephalography (EEG) signals suffer from low information transfer rates and high variability across users. Recently, eye movement signals have emerged as a promising alternative due to their high accuracy and robustness. Eye movement signals are the electrical or mechanical signals generated by the movements and behaviors of the eyes, serving to denote the diverse forms of eye movements, such as fixations, smooth pursuit, and other oculomotor activities like blinking. This article presents a review of recent studies on the development of BCI typing systems that incorporate eye movement signals. We first discuss the basic principles of BCI and the recent advancements in text entry. Then, we provide a comprehensive summary of the latest advancements in BCI typing systems that leverage eye movement signals. This includes an in-depth analysis of hybrid BCIs that are built upon the integration of electrooculography (EOG) and eye tracking technology, aiming to enhance the performance and functionality of the system. Moreover, we highlight the advantages and limitations of different approaches, as well as potential future directions. Overall, eye movement signals hold great potential for enhancing the usability and accessibility of BCI typing systems, and further research in this area could lead to more effective communication and control for individuals with motor disabilities.

Tetherless Optical Neuromodulation: Wavelength from Orange-red to Mid-infrared.

Optogenetics, a technique that employs light for neuromodulation, has revolutionized the study of neural mechanisms and the treatment of neurological disorders due to its high spatiotemporal resolution and cell-type specificity. However, visible light, particularly blue and green light, commonly used in conventional optogenetics, has limited penetration in biological tissue. This limitation necessitates the implantation of optical fibers for light delivery, especially in deep brain regions, leading to tissue damage and experimental constraints. To overcome these challenges, the use of orange-red and infrared light with greater tissue penetration has emerged as a promising approach for tetherless optical neuromodulation. In this review, we provide an overview of the development and applications of tetherless optical neuromodulation methods with long wavelengths. We first discuss the exploration of orange-red wavelength-responsive rhodopsins and their performance in tetherless optical neuromodulation. Then, we summarize two novel tetherless neuromodulation methods using near-infrared light: upconversion nanoparticle-mediated optogenetics and photothermal neuromodulation. In addition, we discuss recent advances in mid-infrared optical neuromodulation.

Classification of self-limited epilepsy with centrotemporal spikes by classical machine learning and deep learning based on electroencephalogram data.

Electroencephalogram (EEG) has been widely utilized as a valuable assessment tool for diagnosing epilepsy in hospital settings. However, clinical diagnosis of patients with self-limited epilepsy with centrotemporal spikes (SeLECTS) is challenging due to the presence of similar abnormal discharges in EEG displays compared to other types of epilepsy (non-SeLECTS) patients. To assist the diagnostic process of epilepsy, a comprehensive classification study utilizing machine learning or deep learning techniques is proposed. In this study, clinical EEG was collected from 33 patients diagnosed with either SeLECTS or non-SeLECTS, aged between 3 and 11 years. In the realm of classical machine learning, sharp wave features (including upslope, downslope, and width at half maximum) were extracted from the EEG data. These features were then combined with the random forest (RF) and extreme random forest (ERF) classifiers to differentiate between SeLECTS and non-SeLECTS. Additionally, deep learning was employed by directly inputting the EEG data into a deep residual network (ResNet) for classification. The classification results were evaluated based on accuracy, F1-score, area under the curve (AUC), and area under the precision-recall curve (AUPRC). Following a 10-fold cross-validation, the ERF classifier achieved an accuracy of 73.15 % when utilizing sharp wave feature extraction for classification. The F1-score obtained was 0.72, while the AUC and AUPRC values were 0.75 and 0.63, respectively. On the other hand, the ResNet model achieved a classification accuracy of 90.49 %, with an F1-score of 0.90. The AUC and AUPRC values for ResNet were found to be 0.96 and 0.92, respectively. These results highlighted the significant potential of deep learning methods in SeLECTS classification research, owing to their high accuracy. Moreover, feature extraction-based methods demonstrated good reliability and could assist in identifying relevant biological features of SeLECTS within EEG data.

A Snapshot Multi-Spectral Demosaicing Method for Multi-Spectral Filter Array Images Based on Channel Attention Network.

Multi-spectral imaging technologies have made great progress in the past few decades. The development of snapshot cameras equipped with a specific multi-spectral filter array (MSFA) allow dynamic scenes to be captured on a miniaturized platform across multiple spectral bands, opening up extensive applications in quantitative and visualized analysis. However, a snapshot camera based on MSFA captures a single band per pixel; thus, the other spectral band components of pixels are all missed. The raw images, which are captured by snapshot multi-spectral imaging systems, require a reconstruction procedure called demosaicing to estimate a fully defined multi-spectral image (MSI). With increasing spectral bands, the challenge of demosaicing becomes more difficult. Furthermore, the existing demosaicing methods will produce adverse artifacts and aliasing because of the adverse effects of spatial interpolation and the inadequacy of the number of layers in the network structure. In this paper, a novel multi-spectral demosaicing method based on a deep convolution neural network (CNN) is proposed for the reconstruction of full-resolution multi-spectral images from raw MSFA-based spectral mosaic images. The CNN is integrated with the channel attention mechanism to protect important channel features. We verify the merits of the proposed method using 5 × 5 raw mosaic images on synthetic as well as real-world data. The experimental results show that the proposed method outperforms the existing demosaicing methods in terms of spatial details and spectral fidelity.

Spectral encoder to extract the efficient features of Raman spectra for reliable and precise quantitative analysis.

Raman spectroscopy has become a powerful analytical tool highly demanded in many applications such as microorganism sample analysis, food quality control, environmental science, and pharmaceutical analysis, owing to its non-invasiveness, simplicity, rapidity and ease of use. Among them, quantitative research using Raman spectroscopy is a crucial application field of spectral analysis. However, the entire process of quantitative modeling largely relies on the extraction of effective spectral features, particularly for measurements on complex samples or in environments with poor spectral signal quality. In this paper, we propose a method of utilizing a spectral encoder to extract effective spectral features, which can significantly enhance the reliability and precision of quantitative analysis. We built a latent encoded feature regression model; in the process of utilizing the autoencoder for reconstructing the spectrometer output, the latent feature obtained from the intermediate bottleneck layer is extracted. Then, these latent features are fed into a deep regression model for component concentration prediction. Through detailed ablation and comparative experiments, our proposed model demonstrates superior performance to common methods on single-component and multi-component mixture datasets, remarkably improving regression precision while without needing user-selected parameters and eliminating the interference of irrelevant and redundant information. Furthermore, in-depth analysis reveals that latent encoded feature possesses strong nonlinear feature representation capabilities, low computational costs, wide adaptability, and robustness against noise interference. This highlights its effectiveness in spectral regression tasks and indicates its potential in other application fields. Sufficient experimental results show that our proposed method provides a novel and effective feature extraction approach for spectral analysis, which is simple, suitable for various methods, and can meet the measurement needs of different real-world scenarios.

Rapid Determination of Positive-Negative Bacterial Infection Based on Micro-Hyperspectral Technology.

To meet the demand for rapid bacterial detection in clinical practice, this study proposed a joint determination model based on spectral database matching combined with a deep learning model for the determination of positive-negative bacterial infection in directly smeared urine samples. Based on a dataset of 8124 urine samples, a standard hyperspectral database of common bacteria and impurities was established. This database, combined with an automated single-target extraction, was used to perform spectral matching for single bacterial targets in directly smeared data. To address the multi-scale features and the need for the rapid analysis of directly smeared data, a multi-scale buffered convolutional neural network, MBNet, was introduced, which included three convolutional combination units and four buffer units to extract the spectral features of directly smeared data from different dimensions. The focus was on studying the differences in spectral features between positive and negative bacterial infection, as well as the temporal correlation between positive-negative determination and short-term cultivation. The experimental results demonstrate that the joint determination model achieved an accuracy of 97.29%, a Positive Predictive Value (PPV) of 97.17%, and a Negative Predictive Value (NPV) of 97.60% in the directly smeared urine dataset. This result outperformed the single MBNet model, indicating the effectiveness of the multi-scale buffered architecture for global and large-scale features of directly smeared data, as well as the high sensitivity of spectral database matching for single bacterial targets. The rapid determination solution of the whole process, which combines directly smeared sample preparation, joint determination model, and software analysis integration, can provide a preliminary report of bacterial infection within 10 min, and it is expected to become a powerful supplement to the existing technologies of rapid bacterial detection.

Event-related potentials reveal visual episodic memory deficits in patients with temporal lobe epilepsy.

OBJECTIVE: Temporal lobe epilepsy (TLE) patients usually suffer from impaired episodic memory (EM), but its underlying electrophysiologic mechanism and impacted cognitive performance are unclear. We aim to investigate the association between episodic memory reserve and physiological measures of memory workload in TLE patients using Event-related potentials (ERP). METHODS: A change detection task with image stimuli assesses visual episodic memory. During the memory encoding and decoding phases, the ERP signals were analyzed from twenty-nine TLE patients (twelve with left TLE patients, seventeen with TLE), and thirty healthy controls. Given that EM is a complex process involving many fundamental cognitive processes, the amplitudes and latencies of EM-related ERP (FN400, late positive potential (LPC), and late posterior negativity (LPN)), and the ERP reflecting the fundamental processes (P100, N100, P200, and P300) were calculated. Then we used a three-by-two factorial design on the ERP metrics for interaction and main effects. The correlation analysis among Wechsler Memory Scales-Chinese Revision (WMS-RC) results, behavioral data, and the ERPs was carried out. RESULTS: The TLE patients performed worse in WMS-RC and the memory task. The increased P200 and decreased P300 amplitudes were observed in the TLE patients, and LPN was abnormal in only LTLE patients. For EM-related components, differences were observed in both the LTLE and RTLE patients: the lack of the FN400 effect, the lack of the reversed LPC effect, and the reduced FN400. No significant inter-group difference was detected for the latencies of all the ERPs. Additionally, there were significant correlations among WMS-RC scores, behaviors, and some ERP amplitudes. CONCLUSIONS: The impaired EM is linked to the increased P200 and decreased P300 amplitudes. LPN seems to be sensitive to left temporal lobe dysfunction. More importantly, the abnormal old or new effects of the FN400 and LPC, and the reduced FN400 amplitude might be associated with the visual EM deficit in the TLE patients. These findings may assist in the deep understanding of the EM disorder and the evaluation of the side effects of antiepileptic drugs.

An eye tracking investigation of attention mechanism in driving behavior under emotional issues and cognitive load.

Emotions have specific effects on behavior. At present, studies are increasingly interested in how emotions affect driving behavior. We designed the experiment by combing driving tasks and eye tracking. DSM-V assessment scale was applied to evaluate the depression and manic for participants. In order to explore the dual impacts of emotional issues and cognitive load on attention mechanism, we defined the safety-related region as the area of interest (AOI) and quantified the concentration of eye tracking data. Participants with depression issues had lower AOI sample percentage and shorter AOI fixation duration under no external cognitive load. During our experiment, the depression group had the lowest accuracy in arithmetic quiz. Additionally, we used full connected network to detect the depression group from the control group, reached 83.33%. Our experiment supported that depression have negative influences on driving behavior. Participants with depression issues reduced attention to the safety-related region under no external cognitive load, they were more prone to have difficulties in multitasking when faced with high cognitive load. Besides, participants tended to reallocate more attention resources to the central area under high cognitive load, a phenomenon we called "visual centralization" in driving behavior.

Temporal transformer-spatial graph convolutional network: an intelligent classification model for anti N-methyl-D-aspartate receptor encephalitis based on electroencephalogram signal.

Encephalitis is a disease typically caused by viral infections or autoimmunity. The most common type of autoimmune encephalitis is anti-N-methyl-D-aspartate receptor (NMDAR) antibody-mediated, known as anti-NMDA receptor encephalitis, which is a rare disease. Specific EEG patterns, including "extreme delta brush" (EDB), have been reported in patients with anti-NMDA receptor encephalitis. The aim of this study was to develop an intelligent diagnostic model for encephalitis based on EEG signals. A total of 131 Participants were selected based on reasonable inclusion criteria and divided into three groups: health control (35 participants), viral encephalitis (58 participants), and anti NMDAR receptor encephalitis (55 participants). Due to the low prevalence of anti-NMDAR receptor encephalitis, it took several years to collect participants' EEG signals while they were in an awake state. EEG signals were collected and analyzed following the international 10-20 system layout. We proposed a model called Temporal Transformer-Spatial Graph Convolutional Network (TT-SGCN), which consists of a Preprocess Module, a Temporal Transformer Module (TTM), and a Spatial Graph Convolutional Module (SGCM). The raw EEG signal was preprocessed according to traditional procedures, including filtering, averaging, and Independent Component Analysis (ICA) method. The EEG signal was then segmented and transformed using short-time Fourier transform (STFT) to produce concatenated power density (CPD) maps, which served as inputs for the proposed model. TTM extracted the time-frequency features of each channel, and SGCM fused these features using graph convolutional methods based on the location of electrodes. The model was evaluated in two experiments: classification of the three groups and pairwise classification among the three groups. The model was trained using two stages and achieved the performance, with an accuracy of 82.23%, recall of 80.75%, precision of 82.51%, and F1 score of 81.23% in the classification of the three groups. The proposed model has the potential to become an intelligent auxiliary diagnostic tool for encephalitis.

Deep neural network: As the novel pipelines in multiple preprocessing for Raman spectroscopy.

Raman spectroscopy is a kind of vibrational method that can rapidly and non-invasively gives chemical structural information with the Raman spectrometer. Despite its technical advantages, in practical application scenarios, Raman spectroscopy often suffers from interference, such as noises and baseline drifts, resulting in the inability to acquire high-quality Raman spectroscopy signals, which brings challenges to subsequent spectral analysis. The commonly applied spectral preprocessing methods, such as Savitzky-Golay smooth and wavelet transform, can only perform corresponding single-item processing and require manual intervention to carry out a series of tedious trial parameters. Especially, each scheme can only be used for a specific data set. In recent years, the development of deep neural networks has provided new solutions for intelligent preprocessing of spectral data. In this paper, we first creatively started from the basic mechanism of spectral signal generation and constructed a mathematical model of the Raman spectral signal. By counting the noise parameters of the real system, we generated a simulation dataset close to the output of the real system, which alleviated the dependence on data during deep learning training. Due to the powerful nonlinear fitting ability of the neural network, fully connected network model is constructed to complete the baseline estimation task simply and quickly. Then building the Unet model can effectively achieve spectral denoising, and combining it with baseline estimation can realize intelligent joint processing. Through the simulation dataset experiment, it is proved that compared with the classic method, the method proposed in this paper has obvious advantages, which can effectively improve the signal quality and further ensure the accuracy of the peak intensity. At the same time, when the proposed method is applied to the actual system, it also achieves excellent performance compared with the common method, which indirectly indicates the effectiveness of the Raman signal simulation model. The research presented in this paper offers a variety of efficient pipelines for the intelligent processing of Raman spectroscopy, which can adapt to the requirements of different tasks while providing a new idea for enhancing the quality of Raman spectroscopy signals.

Automatic Seizure Detection and Prediction Based on Brain Connectivity Features and a CNNs Meet Transformers Classifier.

(1) Background: Epilepsy is a neurological disorder that causes repeated seizures. Since electroencephalogram (EEG) patterns differ in different states (inter-ictal, pre-ictal, and ictal), a seizure can be detected and predicted by extracting various features. However, the brain connectivity network, a two-dimensional feature, is rarely studied. We aim to investigate its effectiveness for seizure detection and prediction. (2) Methods: Two time-window lengths, five frequency bands, and five connectivity measures were used to extract image-like features, which were fed into a support vector machine for the subject-specific model (SSM) and a convolutional neural networks meet transformers (CMT) classifier for the subject-independent model (SIM) and cross-subject model (CSM). Finally, feature selection and efficiency analyses were conducted. (3) Results: The classification results on the CHB-MIT dataset showed that a long window indicated better performance. The best detection accuracies of SSM, SIM, and CSM were 100.00, 99.98, and 99.27%, respectively. The highest prediction accuracies were 99.72, 99.38, and 86.17%, respectively. In addition, Pearson Correlation Coefficient and Phase Lock Value connectivity in the ß and γ bands showed good performance and high efficiency. (4) Conclusions: The proposed brain connectivity features showed good reliability and practical value for automatic seizure detection and prediction, which expects to develop portable real-time monitoring equipment.

Adaptive convolution kernel network for change detection in hyperspectral images.

Feature extraction is a key step in hyperspectral image change detection. However, many targets with great various sizes, such as narrow paths, wide rivers, and large tracts of cultivated land, can appear in a satellite remote sensing image at the same time, which will increase the difficulty of feature extraction. In addition, the phenomenon that the number of changed pixels is much less than unchanged pixels will lead to class imbalance and affect the accuracy of change detection. To address the above issues, based on the U-Net model, we propose an adaptive convolution kernel structure to replace the original convolution operations and design a weight loss function in the training stage. The adaptive convolution kernel contains two various kernel sizes and can automatically generate their corresponding weight feature map during training. Each output pixel obtains the corresponding convolution kernel combination according to the weight. This structure of automatically selecting the size of the convolution kernel can effectively adapt to different sizes of targets and extract multi-scale spatial features. The modified cross-entropy loss function solves the problem of class imbalance by increasing the weight of changed pixels. Study results on four datasets indicate that the proposed method performs better than most existing methods.

Recent advances of lanthanide nanomaterials in Tumor NIR fluorescence detection and treatment.

Lanthanide nanomaterials have garnered significant attention from researchers among the main near-infrared (NIR) fluorescent nanomaterials due to their excellent chemical and fluorescence stability, narrow emission band, adjustable luminescence color, and long lifetime. In recent years, with the preparation, functional modification, and fluorescence improvement of lanthanide materials, great progress has been made in their application in the biomedical field. This review focuses on the latest progress of lanthanide nanomaterials in tumor diagnosis and treatment, as well as the interaction mechanism between fluorescence and biological tissues. We introduce a set of efficient strategies for improving the fluorescence properties of lanthanide nanomaterials and discuss some representative in-depth research work in detail, showcasing their superiority in early detection of ultra-small tumors, phototherapy, and real-time guidance for surgical resection. However, lanthanide nanomaterials have only realized a portion of their potential in tumor applications so far. Therefore, we discuss promising methods for further improving the performance of lanthanide nanomaterials and their future development directions.

Joint Texture Search and Histogram Redistribution for Hyperspectral Image Quality Improvement.

Due to optical noise, electrical noise, and compression error, data hyperspectral remote sensing equipment is inevitably contaminated by various noises, which seriously affect the applications of hyperspectral data. Therefore, it is of great significance to enhance hyperspectral imaging data quality. To guarantee the spectral accuracy during data processing, band-wise algorithms are not suitable for hyperspectral data. This paper proposes a quality enhancement algorithm based on texture search and histogram redistribution combined with denoising and contrast enhancement. Firstly, a texture-based search algorithm is proposed to improve the accuracy of denoising by improving the sparsity of 4D block matching clustering. Then, histogram redistribution and Poisson fusion are used to enhance spatial contrast while preserving spectral information. Synthesized noising data from public hyperspectral datasets are used to quantitatively evaluate the proposed algorithm, and multiple criteria are used to analyze the experimental results. At the same time, classification tasks were used to verify the quality of the enhanced data. The results show that the proposed algorithm is satisfactory for hyperspectral data quality improvement.

Rapid Identification of Infectious Pathogens at the Single-Cell Level via Combining Hyperspectral Microscopic Images and Deep Learning.

Identifying infectious pathogens quickly and accurately is significant for patients and doctors. Identifying single bacterial strains is significant in eliminating culture and speeding up diagnosis. We present an advanced optical method for the rapid detection of infectious (including common and uncommon) pathogens by combining hyperspectral microscopic imaging and deep learning. To acquire more information regarding the pathogens, we developed a hyperspectral microscopic imaging system with a wide wavelength range and fine spectral resolution. Furthermore, an end-to-end deep learning network based on feature fusion, called BI-Net, was designed to extract the species-dependent features encoded in cell-level hyperspectral images as the fingerprints for species differentiation. After being trained based on a large-scale dataset that we built to identify common pathogens, BI-Net was used to classify uncommon pathogens via transfer learning. An extensive analysis demonstrated that BI-Net was able to learn species-dependent characteristics, with the classification accuracy and Kappa coefficients being 92% and 0.92, respectively, for both common and uncommon species. Our method outperformed state-of-the-art methods by a large margin and its excellent performance demonstrates its excellent potential in clinical practice.

Analytical method for calculating internal stray radiation from an IR spectrometer based on the view factor.

Optomechanical components such as the lens barrels and frames of IR spectrometers produce strong internal stray radiation, which reduces the instrument's SNR and dynamic range. An IR internal stray radiation calculation method based on an analytical model of the view factor is proposed. The mathematical model of the view factor calculation method of typical optomechanical components is established. For any IR optical systems, the internal stray radiation can be quickly and accurately calculated by adjusting the coordinate systems in the calculation method. Based on the proposed method, the internal stray radiation of a double-pass long-wave IR spectrometer was calculated. The calculation results are consistent with the simulation results. The RMS value of the relative error between the calculated value and the simulated value is around 11%. To verify the proposed method, an experiment was conducted to test the internal stray radiation of the long-wave IR spectrometer. The internal stray radiation test results agree with the calculated and simulated results, and the relative error between the test results and the calculation results is within 9%.

CMID: Crossmodal Image Denoising via Pixel-Wise Deep Reinforcement Learning.

Removing noise from acquired images is a crucial step in various image processing and computer vision tasks. However, the existing methods primarily focus on removing specific noise and ignore the ability to work across modalities, resulting in limited generalization performance. Inspired by the iterative procedure of image processing used by professionals, we propose a pixel-wise crossmodal image-denoising method based on deep reinforcement learning to effectively handle noise across modalities. We proposed a similarity reward to help teach an optimal action sequence to model the step-wise nature of the human processing process explicitly. In addition, We designed an action set capable of handling multiple types of noise to construct the action space, thereby achieving successful crossmodal denoising. Extensive experiments against state-of-the-art methods on publicly available RGB, infrared, and terahertz datasets demonstrate the superiority of our method in crossmodal image denoising.

Transformer-based factorized encoder for classification of pneumoconiosis on 3D CT images.

In the past decade, deep learning methods have been implemented in the medical image fields and have achieved good performance. Recently, deep learning algorithms have been successful in the evaluation of diagnosis on lung images. Although chest radiography (CR) is the standard data modality for diagnosing pneumoconiosis, computed tomography (CT) typically provides more details of the lesions in the lung. Thus, a transformer-based factorized encoder (TBFE) was proposed and first applied for the classification of pneumoconiosis depicted on 3D CT images. Specifically, a factorized encoder consists of two transformer encoders. The first transformer encoder enables the interaction of intra-slice by encoding feature maps from the same slice of CT. The second transformer encoder explores the inter-slice interaction by encoding feature maps from different slices. In addition, the lack of grading standards on CT for labeling the pneumoconiosis lesions. Thus, an acknowledged CR-based grading system was applied to mark the corresponding pneumoconiosis CT stage. Then, we pre-trained the 3D convolutional autoencoder on the public LIDC-IDRI dataset and fixed the parameters of the last convolutional layer of the encoder to extract CT feature maps with underlying spatial structural information from our 3D CT dataset. Experimental results demonstrated the superiority of the TBFE over other 3D-CNN networks, achieving an accuracy of 97.06%, a recall of 89.33%, precision of 90%, and an F1-score of 93.33%, using 10-fold cross-validation.

A Deep-Learning Based System for Rapid Genus Identification of Pathogens under Hyperspectral Microscopic Images.

Infectious diseases have always been a major threat to the survival of humanity. Additionally, they bring an enormous economic burden to society. The conventional methods for bacteria identification are expensive, time-consuming and laborious. Therefore, it is of great importance to automatically rapidly identify pathogenic bacteria in a short time. Here, we constructed an AI-assisted system for automating rapid bacteria genus identification, combining the hyperspectral microscopic technology and a deep-learning-based algorithm Buffer Net. After being trained and validated in the self-built dataset, which consists of 11 genera with over 130,000 hyperspectral images, the accuracy of the algorithm could achieve 94.9%, which outperformed 1D-CNN, 2D-CNN and 3D-ResNet. The AI-assisted system we developed has great potential in assisting clinicians in identifying pathogenic bacteria at the single-cell level with high accuracy in a cheap, rapid and automatic way. Since the AI-assisted system can identify the pathogenic genus rapidly (about 30 s per hyperspectral microscopic image) at the single-cell level, it can shorten the time or even eliminate the demand for cultivating. Additionally, the system is user-friendly for novices.

An Attention-Based CoT-ResNet With Channel Shuffle Mechanism for Classification of Alzheimer's Disease Levels.

Detection of early morphological changes in the brain and early diagnosis are important for Alzheimer's disease (AD), and high-resolution magnetic resonance imaging (MRI) can be used to help diagnose and predict the disease. In this paper, we proposed two improved ResNet algorithms that introduced the Contextual Transformer (CoT) module, group convolution, and Channel Shuffle mechanism into the traditional ResNet residual blocks. The CoT module is used to replace the 3 × 3 convolution in the residual block to enhance the feature extraction capability of the residual block, while the Channel Shuffle mechanism is used to reorganize the feature maps of different groups in the input layer to improve the communication between the feature maps from different groups. Images of 503 subjects, including 116 healthy controls (HC), 187 subjects with mild cognitive impairment (MCI), and 200 subjects with AD, were selected and collated from the ADNI database, and then, the data were pre-processed and sliced. After that, 10,060 slices were obtained and the three groups of AD, MCI and HC were classified using the improved algorithms. The experiments showed that the refined ResNet-18-based algorithm improved the top-1 accuracy by 2.06%, 0.33%, 1.82%, and 1.52% over the traditional ResNet-18 algorithm for four medical image classification tasks, namely AD: MCI, AD: HC, MCI: HC, and AD: MCI: HC, respectively. The enhanced ResNet-50-based algorithm improved the top-1 accuracy by 1.02%, 2.92%, 3.30%, and 1.31%, respectively, over the traditional ResNet-50 algorithm in four medical image classification tasks, demonstrating the effectiveness of the CoT module replacement and the inclusion of the channel shuffling mechanism, as well as the competitiveness of the improved algorithms.