ReLU Activation Research Articles

A Pixel-Based Machine Learning Atmospheric Correction for PeruSAT-1 Imagery

Atmospheric correction is essential in remote sensing, as it reduces the effects of light absorption and scattering by suspended particles and gases, enabling accurate surface reflectance computation from the observed Top-of-Atmosphere (TOA) reflectance. Each satellite sensor requires a customized atmospheric correction processor due to its unique system characteristics. Currently, PeruSAT-1, the first Peruvian remote sensing satellite, does not include this capability in its image processing pipeline, which poses challenges for its effectiveness in defense, security, and natural disaster management applications. This research investigated pixel-based machine learning methods for atmospheric correction of PeruSAT-1, using Sentinel-2 harmonized Bottom-of-Atmosphere (BOA) surface reflectance images as a benchmark, alongside additional atmospheric, terrain, and acquisition parameters. A robust dataset was developed to align data across temporal, spatial, geometric, and contextual conditions. Experimental results showed R2 values between 0.886 and 0.938, and RMSE values ranging from 0.009 to 0.025 compared to the benchmarks. Among the models tested, the Feedforward Neural Network (FFNN) using the Leaky ReLU activation function achieved the best overall performance. These findings confirm the robustness of this approach, offering a scalable methodology for satellites with similar characteristics and establishing a foundation for a customized atmospheric correction pipeline for PeruSAT-1. Future work will focus on diversifying the dataset across spectral and seasonal conditions and optimizing the modeling to address challenges in shorter wavelengths and high-reflectance surfaces.

CONVOLUTIONAL LSTM FOR EGG PRICES FORECASTING IN INDONESIA’S HIGH STUNTING PREVALENCE PROVINCE

The Sustainable Development Goals (SDGs) are a series of 17 goals fixed by the United Nations and adopted by 193 countries in 2015, including Indonesia. By 2030, to end all forms of malnutrition, targeting on stunting and wasting in children under 5 years of age is one of the targets from Goals Number 2. One affordable source of protein and nutrition used as a solution to overcome malnutrition problems such as stunting is eggs. Egg price modeling was carried out to see the affordability of prices for the community. Weekly dataset of egg price in NTT Province from 2018 to 2023 used to modeling with Convolutional LSTM. The Convolutional LSTM components are Adam optimizer, ReLU activation function, Huber loss function, with batch size and neurons of 32. The MAPE value obtained from the model is relatively small, with MAPE for training, validation, and testing of 1.97%, 1%, and 1.19% respectively. The results of egg price forecasting for December 11, 2023, to January 8, 2024, show that egg prices tend to continue to decline per week. Thus, a decrease in egg prices can be a good thing in providing more affordable nutrition for the community.

ENHANCING 〖PM〗_(2.5) PREDICTION IN KEMAYORAN DISTRICT, DKI JAKARTA USING DEEP BILSTM METHOD

Worldwide air pollution is a concern, and this is especially true in Indonesia, where most people breathe air that is more contaminated than recommended by the WHO. The concentration of presents notable health hazards. The respiratory system is the primary route of absorption for , allowing it to enter the lung alveoli and enter the bloodstream. Given the significant health risks associated with exposure, accurate forecasting methods are crucial to anticipate and mitigate its effects. Traditional forecasting methods like ARIMA have limitations in handling non-linear and complex patterns. Therefore, an accurate machine learning method is needed to improve forecasting performance. This research employs Deep Bidirectional Long-Short Term Memory (BiLSTM), a deep learning model particularly suited for time series forecasting due to its ability to capture both past and future dependencies in sequential data. To achieve accurate and precise forecasts for predicting concentration levels in Kemayoran District in November , 2023 (24 hours), this research utilized hourly concentration data from May until October , 2023, using Deep BiLSTM. The outcomes demonstrated the efficiency of the model, attaining a Mean Absolute Percentage Error (MAPE) of 17.1540% (training) and 14.2862% (testing) with an 80:20 data split. The optimal parameters, which comprised 24 timesteps, Adam optimizers with a learning rate of 0.001, 16 batch sizes, 1000 epochs, and ReLU activation functions across multiple BiLSTM layers, showcased the model’s effectiveness in forecasting the concentration in Kemayoran District, DKI Jakarta, on November , 2023.

Scalable Subsampling Inference for Deep Neural Networks

Deep neural networks (DNN) has received increasing attention in machine learning applications in the last several years. Recently, a non-asymptotic error bound has been developed to measure the performance of the fully connected DNN estimator with ReLU activation functions for estimating regression models. The paper at hand gives a small improvement on the current error bound based on the latest results on the approximation ability of (forward) DNN. More importantly, however, a non-random subsampling technique–scalable subsampling–is applied to construct a ‘subagged’ DNN estimator. Under regularity conditions, it is shown that the subagged DNN estimator is computationally efficient without sacrificing accuracy for either estimation or prediction tasks. Beyond point estimation/prediction, we propose different approaches to build confidence and prediction intervals based on the subagged DNN estimator. In addition to being asymptotically valid, the proposed confidence/prediction intervals appear to work well in finite samples. All in all, the scalable subsampling DNN estimator offers the complete package in terms of statistical inference, i.e., (a) computational efficiency; (b) point estimation/prediction accuracy; and (c) allowing for the construction of practically useful confidence and prediction intervals.

Application of Machine Learning to the Prediction of Surface Roughness in the Milling Process on the Basis of Sensor Signals

This article presents an investigation of the use of machine learning methodologies for the prediction of surface roughness in milling operations, using sensor data as the primary source of information. The sensors, which included current transformers, a microphone, and displacement sensors, captured comprehensive machining signals at a frequency of 10 kHz. The signals were subjected to preprocessing using the Savitzky–Golay filter, with the objective of isolating relevant moments of active material machining and reducing noise. Two machine learning models, namely Elastic Net and neural networks, were employed for the prediction purposes. The Elastic Net model demonstrated effective handling of multicollinearity and reduction in the data dimensionality, while the neural networks, utilizing the ReLU activation function, exhibited the capacity to capture complex, nonlinear patterns. The models were evaluated using the coefficient of determination (R²), which yielded values of 0.94 for Elastic Net and 0.95 for neural networks, indicating a high degree of predictive accuracy. These findings illustrate the potential of machine learning to optimize manufacturing processes by facilitating precise predictions of surface roughness. Elastic Net demonstrated its utility in reducing dimensionality and selecting features, while neural networks proved effective in modeling complex data. Consequently, these methods exemplify the efficacy of integrating data-driven approaches with robust predictive models to improve the quality and efficiency of surface.

FLNCVD-net: COVID-19 Detection Using Chest X-rays

This research work proposes faster alternative ways of predicting the Coronavirus Disease 2019 (COVID-19) infection in the patient’s body. Firstly, this research work proposes a transfer learning-based solution methodology that trains the following models, VGG16, VGG19, MobileNetV2, NASnet Moblie and ResNet-50 to find out an efficient base model for Novel Corona Virus Detect-net (NCVD-net) to detect COVID-19 infected patients using chest X-rays. Among these, VGG-19 is chosen as the best base model as it outperforms the other models. Afterward, the chosen VGG-19 is transformed to VGG-19 based NCVD-net by replacing its fully connected layer with an average max pooling layer, flattened layer, three dense layers (1 x 256, 1x128, 1 x 64) with leaky ReLU activation function and one output dense layer (1 x 2) with softmax activation. Following that, VGG-19-based NCVD-net is tested against unseen chest X-rays and successfully detects COVID-19-infected patients with 99.91% accuracy. Secondly, this paper employs Federated Learning-based Novel Corona Virus Detect-net (FLNCVD-net) by considering VGG-16, VGG-19 and ResNet101 as its base model. Later, these models are distributed securely to each hospital and trained on the chest X-rays available in that hospital. Finally, the model parameters from each FLNCVD-net are securely transmitted and combined to build a global FLNCVD-net. The proposed FLNCVD-net using the above-said three base models are validated against unseen COVID-19 +ve and COVID-19 –ve chest X-ray images from Dr. Joseph Cohen’s GitHub repository and Kaggle respectively. Surprisingly, it is observed that the VGG-16-based FLNCVD-net marginally outperforms VGG-19-based FLNCVD-net.

ANNZ+: an enhanced photometric redshift estimation algorithm with applications on the PAU survey

annz is a fast and simple algorithm which utilises artificial neural networks (ANNs), it was known as one of the pioneers of machine learning approaches to photometric redshift estimation decades ago. We enhanced the algorithm by introducing new activation functions like tanh, softplus, SiLU, Mish and ReLU variants; its new performance is then vigorously tested on legacy samples like the Luminous Red Galaxy (LRG) and Stripe-82 samples from SDSS, as well as modern galaxy samples like the Physics of the Accelerating Universe Survey (PAUS). This work focuses on testing the robustness of activation functions with respect to the choice of ANN architectures, particularly on its depth and width, in the context of galaxy photometric redshift estimation. Our upgraded algorithm, which we named annz+, shows that the tanh and Leaky ReLU activation functions provide more consistent and stable results across deeper and wider architectures with > 1 per cent improvement in root-mean-square error (σ RMS) and 68th percentile error (σ 68) when tested on SDSS data sets. While assessing its capabilities in handling high dimensional inputs, we achieved an improvement of 11 per cent in σ RMS and 6 per cent in σ 68 with the tanh activation function when tested on the 40-narrowband PAUS dataset; it even outperformed annz2, its supposed successor, by 44 per cent in σ RMS. This justifies the effort to upgrade the 20-year-old annz, allowing it to remain viable and competitive within the photo-z community today. The updated algorithm annz+ is publicly available at https://github.com/imdadmpt/ANNzPlus.

Super-Resolution Reconstruction of UAV Images for Maize Tassel Detection

HighlightsHigh resolution images are found to play an important role in observing maize growth and detecting maize tassels.Our super-resolution model (IESRGAN) produces higher-quality images compared to other SR models.IESRGAN largely improves the detection accuracy of maize tassels, increasing by 8.77%~25.71% in different tests.Abstract. The Unmanned Aerial Vehicle (UAV) technology provides essential technical support for maize (Zea mays) cultivation in smart agriculture. To address the challenge of low resolution in UAV maize images due to factors such as flight altitude and instrument parameters, which impact the accurate monitoring and assessment of maize crop growth and further tassel detection tasks, this study proposes an improved enhanced super-resolution generative adversarial network for super-resolution reconstruction of UAV maize images. Firstly, Iterative Attention Feature Fusion is introduced in the generation network, effectively integrating the features of maize and restoring high-frequency details in maize images. Subsequently, the Iterative Attention Feature Fusion is optimized by removing batch normalization and replacing the ReLU activation function with the PReLU activation function, thereby enhancing the model's representation capability. Furthermore, to address the complex features and structures of tassels in maize images, the VGG discriminator network is replaced with a Unet network, thereby improving the model's ability to handle image details and textures. Finally, the model was validated and analyzed on different maize image datasets. Experimental results show that compared with other mainstream super-resolution models, the reconstructed images of our model proposed exhibit higher quality and clarity; for maize tassel detection, the average accuracy of tassel detection in the images reconstructed by our model compared with the low-resolution images was increased by 8.77%~25.71%. This provides a more accurate and clear UAV monitoring method for maize agricultural production and provides more reliable image data support for related decisions and applications. Keywords: Generative adversarial network, Maize image, Super-resolution, Tassel detection, UAV.

EEG Emotion Recognition using Deep Neural Network (DNN) in Virtual Reality Environments

Purpose: The purpose of this study is to explore the integration of EEG technology with virtual reality (VR) systems to enhance therapeutic interventions, improve cognitive state recognition, and develop personalized immersive experiences. Specifically, it investigates the classification of EEG signals in a VR environment using machine learning models and identifies the most effective methods for individual-level analysis.Methods: The study utilized EEG data collected from 31 participants using the Muse 2016 headset, with electrodes positioned according to the 10-20 international system. EEG signals were analyzed for features such as statistical metrics (mean, median, standard deviation, skewness, and kurtosis) and Hjorth parameters (activity, mobility, complexity). Machine learning models, including K-Nearest Neighbors (KNN), Random Forest (RF), and Support Vector Machine (SVM), were evaluated for their performance in classifying emotional and cognitive states in a VR environment. Result: The results indicate that the Deep Neural Network (DNN) outperformed SVM and KNN models, achieving the highest average classification accuracy. SVM demonstrated consistent performance, with accuracy values consistently above 0.8 across subjects, while KNN showed greater variability and lower overall performance. DNN's architecture, incorporating two hidden layers with ReLU activation and a softmax output layer, demonstrated superior capability in modeling complex EEG patterns. The findings emphasize the effectiveness of DNN in handling high-dimensional and non-linear data, particularly for multi-class classification tasks.Novelty: This study is novel in its focus on personalized machine learning model performance in a VR-EEG setup. Instead of a one-size-fits-all approach, it emphasizes individualized analysis, identifying the most effective model for each participant.

Time Facial Expression Recognition Using Optimized CNN Models for Behavioral and Emotional Analysis

Facial expression recognition is a significant field in human-computer interaction, aiming to analyze emotions such as happiness, sadness, anger, and fear. This study develops a facial expression detection system using Convolutional Neural Networks (CNN) to address challenges like lighting variations and facial angles. The research begins with collecting and preprocessing datasets, including FER-2013, to normalize, augment, and label images for seven emotion classes. The CNN model is designed with convolutional, pooling, and fully connected layers, optimized using ReLU activation, Adam optimizer, and categorical crossentropy loss function. Training is conducted on 80% of the dataset, with 20% for validation, achieving a validation accuracy of 91.7%. System performance is evaluated using precision, recall, F1-score, and real-time testing integrated with cameras, achieving an average detection accuracy of 90%. Results demonstrate the system's robustness in detecting emotions under varying conditions, highlighting its potential for applications in security, education, and emotional therapy. Future research recommends incorporating larger datasets and advanced transfer learning methods to improve system efficiency and accuracy.

Acne Severity Classification Study Using Convolutional Neural Network Algorithm with MobileNetV2 Architecture

Data classification is a key technique in machine learning that maps patterns and features of input data into a target class. Significant developments in data classification occur in deep learning with neural networks and Convolutional Neural Networks (CNN) that are able to extract image features automatically. CNN can classify the level of a condition based on image data, one of which is the severity of acne. Acne (acne vulgaris) is a common skin disease with varying severity. This study aims to apply the CNN MobileNetV2 model to classify acne severity based on acne input images. The data consists of 1457 acne images at 4 severity levels divided into 80% training data and 20% test data. MobileNetV2 was used as a feature extractor through transfer learning. Fine-tuning and classification were performed using fully connected layers with ReLU and softmax activation functions. The model was evaluated with a confusion matrix and classification report. The model with a combination of hyperparameter batch size 16 and a learning rate of 0.00001 was the best model that achieved 87.29% accuracy with 89% precision, 84% recall, and 86% F1 score for classifying acne severity.

Integration GSTARIMA with deep neural network to enhance prediction accuracy on rainfall data

This study aimed to improve rainfall prediction accuracy by integrating spatio-temporal Generalized Autoregressive Integrated Moving Average (GSTARIMA) with Deep Neural Network (DNN) techniques. Accurate rainfall forecasting is vital for regions like West Java, which face risks of flooding and landslides during heavy rains in December, January and February. The research focused on enhancing monthly rainfall prediction using GSTARIMA model residuals, transformed from daily data, with a spatial resolution of 0.5° × 0.625°. The GSTARIMA-DNN model was developed to reduce prediction errors and closely align with actual rainfall values. The DNN architecture utilized a Multilayer Perceptron (MLP) with eight input nodes for spatio-temporal residuals, two hidden layers with Leaky ReLU (LReLU) activation functions and a linear output layer for regression. The integration of GSTARIMA and DNN significantly improved prediction accuracy compared to traditional methods. Specifically, the GSTARIMA(0,1,1)-DNN model achieved a 7.28% Mean Absolute Percentage Error (MAPE) at four key locations (Tasikmalaya, Cirebon, Majalengka and Garut) among ten locations with the same lag order. This hybrid model effectively managed complex temporal and spatial data, underscoring the benefits of combining spatio-temporal models with DNN for enhanced rainfall forecasting. This approach offers valuable insights for advancing precise spatio-temporal models in various research fields.

Innovative Convolutional Neural Network Hybrid for Brain Tumor Segmentation

Brain tumor segmentation and classification play vital roles in the treatment, diagnosis, therapy planning, and evaluation of responses to cancer treatment. Magnetic resonance imaging (MRI) is a prevalent method for analyzing brain tumors, utilizing various acquisition protocols, including both conventional and advanced techniques. This study introduces a fully automated model designed to segment abnormal tissues related to brain tumors from multimodal MRI images, thereby assisting radiologists in diagnosis and treatment planning. The proposed model, termed "CNN-FCM," combines a hybrid convolutional neural network with fuzzy c-means clustering to tackle the challenges posed by inaccurate segmentation and diverse input image dimensions. The architecture of the model consists of three layers: a convolutional layer, a ReLU activation layer, and a clustering layer. Comprehensive experiments conducted on the BRATS 2017 dataset reveal that the model delivers competitive outcomes, achieving performance metrics such as accuracy, sensitivity, specificity, overall Dice coefficient, and recall of 96.5%, 95.21%, 94.8%, 93.7%, and 94.1%, respectively. These findings suggest that the hybrid CNN-FCM model effectively resolves issues related to segmentation accuracy and variability in input images within the context of brain tumor analysis is.

Radiomics for differentiating adenocarcinoma and squamous cell carcinoma in non-small cell lung cancer beyond nodule morphology in chest CT

Distinguishing between primary adenocarcinoma (AC) and squamous cell carcinoma (SCC) within non-small cell lung cancer (NSCLC) tumours holds significant management implications. We assessed the performance of radiomics-based models in distinguishing primary there is from SCC presenting as lung nodules on Computed Tomography (CT) scans. We studied individuals with histopathologically proven adenocarcinoma or SCC type NSCLC tumours, detected as lung nodules on Chest CT. The workflow comprised manual nodule segmentation, regions of interest creation, preprocessing data, feature extraction, and nodule classification using machine learning algorithms. The dataset comprised 46 adenocarcinoma and 28 SCC cases. For feature extraction, 101 radiomic features were extracted from the tumour regions using the ‘pyradiomics’ module in Python. After extensive experimentation with various feature importance techniques, the top 10 most significant radiomic features for differentiating between adenocarcinoma and squamous cell carcinoma (SCC) were identified. The Synthetic Minority Over-Sampling Technique was used to achieve a balanced distribution. Lung nodules were classified using 13 machine-learning algorithms, including Linear Discriminant Analysis, Random Forest, AdaBoost, and eXtreme Gradient Boosting. The Multilayer Perceptron (MLP) Classifier with Rectified Linear Unit (ReLu) activation was the most accurate (83% accuracy) with 83% precision and 86% sensitivity in distinguishing SCC from adenocarcinoma. It achieved a balanced F1 score of 83%, indicating well-rounded performance in both precision and sensitivity. The average Area Under the Curve score was 88%, representing good discrimination between the two classes of lung nodules. Radiomics is a powerful non-invasive tool that could potentially add to visual information obtained on CT. The MLP Classifier with ReLu activation showed good accuracy in distinguishing primary lung adenocarcinoma from SCC nodules. However, widespread multicentre trials are required to realize the full potential of radiomics in lung nodules.

Robust text-dependent speaker verification system using gender aware Siamese-Triplet Deep Neural Network

ABSTRACT Speaker verification in text-dependent scenarios is critical for high-security applications but faces challenges such as voice quality variations, linguistic diversity, and gender-related pitch differences, which affect authentication accuracy. This paper introduces a Gender-Aware Siamese-Triplet Network-Deep Neural Network (ST-DNN) architecture to address these challenges. The Gender-Aware Network utilizes Convolutional 2D layers with ReLU activation for initial feature extraction, followed by multi-fusion dense skip connections and batch normalization to integrate features across different depths, enhancing discrimination between male and female speakers. A bottleneck layer compresses feature maps to capture gender-related characteristics effectively. For enhanced speaker verification, separate male and female ST-DNN models are used, each incorporating Individual, Siamese, and Triplet Networks. The Individual Network extracts unique utterance characteristics, the Siamese Network compares speech sample pairs for speaker identity, and the Triplet Network ensures closely grouped embeddings of samples from the same speaker, facilitating precise verification. Experimental results on RSR2015 and RedDots Challenge 2016 datasets demonstrate significant improvements, with reductions in Equal Error Rate (EER) ranging from 32.31% to 54.55% for males and 33.73% to 38.98% for females, and reductions in MinDCF from 53.47% to 86.36% and 39.46% to 71.19%, respectively, validating the efficacy of the ST-DNN in real-world applications.

Classification of diabetic retinopathy grades using CNN feature extraction to segment the lesion

Diabetes's microvascular aftereffect, diabetic retinopathy (DR), is the primary cause of eyesight loss in the globe. In order to prevent vision impairment and to intervene promptly, early detection and precise classification of DR severity are essential. Using standard methods for diagnosing DR requires ophthalmologists to grade cases by hand, a process that can be laborious, subjective, and subject to observer error. In supervised learning task of classification, data instances are classified into predefined classes based on features. The relation between the traits and the classes can be found from the labelled data. After the training is completed, the classes of the unseen data. The frequent reason found for the loss of vision in diabetic retinopathy (DR) is found to be diabetes. Visual damage can be prevented by identifying the degree of DR at right time. For the grading of the DR, deep learning techniques are found to be very effective with maximum possible accuracy. The proposed model is useful in accurately classifying the DR images using the feature extraction with lesion segmentation, by implementing the patterns in the DR images. ReLU activation function is used in the proposed model. CNN feature extraction is used for the important feature extraction by applying the Convolution layers, and edges, textures, and forms are identified. As the model proceeds layer by layer, complicated patterns in the photos can be learned by the model, and can be analysed better. The features of the photos were extracted and found useful in segmentation and classification. ReLU is helpful in improving the convergence and also found useful in learning the patterns. Among the other activation functions, ReLU has higher computational efficiency and therefore is used in the model, which suits well for the DR application. A strong framework is proposed for the classification of the DR grade, for the lesion segmentation and CNN feature extraction. DR categorization using the proposed model is evaluated by data visualization of the important calculated metrics and found to be very effective.

ANN-based software cost estimation with input from COCOMO: CANN model

ANN-based software cost estimation with input from COCOMO: CANN model

Advanced brain age prediction using 3D convolutional neural network on structural MRI

AbstractBackgroundPredicting brain age from neuroimaging data is an emerging field. The age gap (AG), the difference between chronological age (CA) and brain age (BA), is crucial for indicating individual neuroanatomical aging. Previous deep learning models faced challenges in generalizability and neuroanatomical interpretability. We incorporated patients with different dementia types, including dementia with Lewy bodies (DLB) and Alzheimer’s disease (AD), alongside mild cognitive impairment (MCI) and cognitive normal (CN) individuals. This inclusive strategy is essential for comprehensive mapping of neurocognitive trajectories and understanding distinct aging patterns across various cognitive conditions.MethodUtilizing T1‐weighted MRI images of n = 3,859 subjects (Table 1) from the CamCAN, NACC, and ADNI databases, this study aimed to predict brain age in four groups (CN, MCI, AD, and DLB). Structural MRI data were spatial normalized and skull‐striped. Then a 3D Convolutional Neural Network (CNN) based on the skull‐striped data was used for age prediction. The model’s architecture includes three convolutional layers with ReLU activation, max‐pooling, batch normalization, and dropout for regularization, ending with global average pooling and dense layers. The model was trained and validated on CN subjects. The trained model was used to predict age in MCI, DLB, and AD patients as well as the test set of CN subjects.ResultThe 3D CNN model accurately predicted brain age in the CN test set with an AG of 0.64 ± 2.74 years and an absolute AG of 1.86 ± 2.11 years (Figure 1 and Table 1). In DLB and AD patients, the average AG was 3.81 and 2.90 years, respectively, and significantly larger than 0 (P < 10‐5), indicating accelerated aging patterns in these groups. The average AG of MCI was 0.09 years which was significantly smaller than that of both DLB and AD (P < 10‐3), indicating the early stage of impairment in MCI patients.ConclusionOur 3D CNN model accurately predicted brain age in cognitively normal individuals and identified accelerated aging in DLB and AD patients. The model's precision highlights its potential for early detection and understanding of neurocognitive trajectories, contributing to advancements in neurological research and clinical diagnostics.

Simple fully convolutional network to detect Alzheimer’s Disease

AbstractBackgroundDeep Neural Networks (DNNs) have emerged as powerful tools in the biomedical field, particularly for the early detection of neurodegenerative diseases. Despite their complexity and data‐intensive nature, simpler fully connected Convolutional Neural Network (SFCN) architectures have shown effectiveness in accurately discerning between subjects affected by Alzheimer’s disease (AD) and healthy controls (HC). This model draws inspiration from the work of Peng H et al. and is trained on the openly available Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset.MethodIn this study, RAW 3T T1‐weighted MPRAGE MRI images with a sagittal acquisition plane and 3D acquisition type were carefully selected, amounting to 807 MRI images categorized into AD and CN classes. These images underwent brain extraction and linear registration (12 degrees of freedom) to a standard 2mm MNI template using FSL, resulting in uniform‐shaped scans of dimensions 1 × 91 × 109 × 91. The model comprises 5 blocks, each incorporating 2 × 3D CNN blocks, 3D batch normalization, and ReLU activation functions. A single MaxPooling layer doubles the channels at each CNN block, followed by a 3D pointwise CNN block reducing channels from 1024 to 256 while introducing non‐linearity. The final block includes 3D adaptive average pooling with a kernel size of 1, dropout set to 0.5, and 2 × 3D point‐ wise CNN layers acting as fully connected layers. This 13‐layer deep network is trained using Weighted Cross Entropy as the loss function and the Adam optimizer with a learning rate initialized to 0.00001. Training involved 100 epochs with a batch size of 16 and a 70‐10‐20 train‐validation‐test split.ResultThe voxel‐based DNN, trained on 278 AD and 288 CN MRI images, achieves a remarkable 95% accuracy on a test dataset of 161, exhibiting precision and recall values of 0.96 and 0.94, respectively.ConclusionThe promising 95% classification accuracy on the ADNI dataset, encompass‐ ing over 800 subjects, underscores the efficacy of the SFCN in distinguishing between AD and NC subjects based on T1‐weighted MRI scans. This ap‐ proach holds potential for robust early detection of Alzheimer’s disease, con‐ tributing to advancements in neuroimaging‐based diagnostic methodologies.

Artificial neural network assisted omnidirectional strain sensors for human motion perception

Artificial neural network assisted omnidirectional strain sensors for human motion perception