Performance Of Trained Model Research Articles

Deep learning-based sleep stage classification with cardiorespiratory and body movement activities in individuals with suspected sleep disorders

Deep learning methods have gained significant attention in sleep science. This study aimed to assess the performance of a deep learning-based sleep stage classification model constructed using fewer physiological parameters derived from cardiorespiratory and body movement data. Overnight polysomnography (PSG) data from 123 participants (age: 19–82 years) with suspected sleep disorders were analyzed. Multivariate time series data, including heart rate, respiratory rate, cardiorespiratory coupling, and body movement frequency, were input into a bidirectional long short-term memory (biLSTM) network model to train and predict five-class sleep stages. The trained model's performance was evaluated using balanced accuracy, Cohen's κ coefficient, and F1 scores on an epoch-per-epoch basis and compared with the ground truth using the leave-one-out cross-validation scheme. The model achieved an accuracy of 71.2 ± 5.8%, Cohen's κ of 0.425 ± 0.115, and an F1 score of 0.650 ± 0.083 across all sleep stages, and all metrics were negatively correlated with the apnea–hypopnea index, as well as age, but positively correlated with sleep efficiency. Moreover, the model performance varied for each sleep stage, with the highest F1 score observed for N2 and the lowest for N3. Regression and Bland–Altman analyses between sleep parameters of interest derived from deep learning and PSG showed substantial correlations (r = 0.33–0.60) with low bias. The findings demonstrate the efficacy of the biLSTM deep learning model in accurately classifying sleep stages and in estimating sleep parameters for sleep structure analysis using a reduced set of physiological parameters. The current model without using EEG information may expand the application of unobtrusive in-home monitoring to clinically assess the prevalence of sleep disorders outside of a sleep laboratory.

ANALISIS PENGARUH DIMENSI GAMBAR PADA KLASIFIKASI MOTIF BATIK DENGAN MENGGUNAKAN CONVOLUTIONAL NEURAL NETWORK

This study focuses on the classification of batik patterns using image data, specifically with three classes: kawung, parang, and pekalongan batik patterns. A total of 180 images were used for the research, divided equally among the three motifs. The dataset was collected through observations from Google. Data preprocessing involved two stages. In the first stage, the data was split into training and validation sets, with a 70% - 30% ratio, respectively. This separation allowed for evaluating the model's performance and generalization ability on unseen data. In the second stage, the Image Data Generator library was utilized to enhance data diversity and improve the model's ability to generalize patterns. Data augmentation techniques, such as rotation, shear, zoom, horizontal flip, and shift, were applied to the training data. The augmented data was then fed into a pre-defined Convolutional Neural Network (CNN) architecture. The CNN model processed the input data through convolutional layers with max pooling and ReLU activation functions. The outputs from the first convolutional layer were used as inputs for subsequent convolutional layers. The resulting feature maps were flattened and passed through fully connected layers for classification. The model's performance was assessed by evaluating accuracy measures. The CNN model achieved the highest accuracy of 98% on the training data and 100% on the validation data after 7 epochs.. The implementation was done using Python programming language and the Google Colab platform, along with required libraries. Model evaluation involved assessing the trained model's performance by inputting test data and computing evaluation metrics, including accuracy, precision, recall, and F1-score. Confusion matrix analysis provided insights into true positive, true negative, false positive, and false negative predictions. The classification report summarized the performance metrics of the model. In conclusion, the CNN method proved effective in classifying batik patterns. The size of the input images significantly influenced the accuracy of the model. A 148x148 pixel image size yielded 100% accuracy. As a suggestion, future research should consider using a larger dataset to maximize the accuracy of the classification model.

Walking Speed Learning and Generalization Using Seq2Seq Gated and Adaptive Continuous-Time Recurrent Neural Network (S2S-GACTRNN) for a Hip Exoskeleton

The objective of this article is to provide a new approach for walking speed learning and generalization in the speed-adaptation control of exoskeletons. By combining the gated and adaptive continuous-time recurrent neural network (GACTRNN), which has the potential to process periodic signals, and the “sequence to sequence” structure, the S2S-GACTRNN model is proposed for walking gait generation at different speeds. The “proactive loop” and “reactive loop” were presented for learning and generalization capability evaluation, respectively, in the experiments of sinusoidal signals and walking gait signals. First, simulation experiments were carried out to evaluate the S2S-GACTRNN's learning and generalization capabilities of sinusoidal signals with different frequencies. Second, in offline experiments, the model was applied to deal with walking gait signals to evaluate the capabilities of learning and generalization of walking speeds. Third, a client-server system was constructed and an online prediction method was proposed for online experiments to further evaluate the trained model's performance in walking speed generalization. The mean absolute errors of S2S-GACTRNN trained using walking data at three speeds were reduced by 24%, 38%, and 24% compared with that trained using walking data at one speed. The results show that the model has learning and generalization potential for gait signals of different walking speeds and may be utilized as a new approach to the adaptive control of walking speeds in the field of exoskeletons.

MLcps: machine learning cumulative performance score for classification problems.

Assessing the performance of machine learning (ML) models requires careful consideration of the evaluation metrics used. It is often necessary to utilize multiple metrics to gain a comprehensive understanding of a trained model's performance, as each metric focuses on a specific aspect. However, comparing the scores of these individual metrics for each model to determine the best-performing model can be time-consuming and susceptible to subjective user preferences, potentially introducing bias. We propose the Machine Learning Cumulative Performance Score (MLcps), a novel evaluation metric for classification problems. MLcps integrates several precomputed evaluation metrics into a unified score, enabling a comprehensive assessment of the trained model's strengths and weaknesses. We tested MLcps on 4 publicly available datasets, and the results demonstrate that MLcps provides a holistic evaluation of the model's robustness, ensuring a thorough understanding of its overall performance. By utilizing MLcps, researchers and practitioners no longer need to individually examine and compare multiple metrics to identify the best-performing models. Instead, they can rely on a single MLcps value to assess the overall performance of their ML models. This streamlined evaluation process saves valuable time and effort, enhancing the efficiency of model evaluation. MLcps is available as a Python package at https://pypi.org/project/MLcps/.

High-quality PET image synthesis from ultra-low-dose PET/MRI using bi-task deep learning.

Lowering the dose for positron emission tomography (PET) imaging reduces patients' radiation burden but decreases the image quality by increasing noise and reducing imaging detail and quantifications. This paper introduces a method for acquiring high-quality PET images from an ultra-low-dose state to achieve both high-quality images and a low radiation burden. We developed a two-task-based end-to-end generative adversarial network, named bi-c-GAN, that incorporated the advantages of PET and magnetic resonance imaging (MRI) modalities to synthesize high-quality PET images from an ultra-low-dose input. Moreover, a combined loss, including the mean absolute error, structural loss, and bias loss, was created to improve the trained model's performance. Real integrated PET/MRI data from 67 patients' axial heads (each with 161 slices) were used for training and validation purposes. Synthesized images were quantified by the peak signal-to-noise ratio (PSNR), normalized mean square error (NMSE), structural similarity (SSIM), and contrast noise ratio (CNR). The improvement ratios of these four selected quantitative metrics were used to compare the images produced by bi-c-GAN with other methods. In the four-fold cross-validation, the proposed bi-c-GAN outperformed the other three selected methods (U-net, c-GAN, and multiple input c-GAN). With the bi-c-GAN, in a 5% low-dose PET, the image quality was higher than that of the other three methods by at least 6.7% in the PSNR, 0.6% in the SSIM, 1.3% in the NMSE, and 8% in the CNR. In the hold-out validation, bi-c-GAN improved the image quality compared to U-net and c-GAN in both 2.5% and 10% low-dose PET. For example, the PSNR using bi-C-GAN was at least 4.46% in the 2.5% low-dose PET and at most 14.88% in the 10% low-dose PET. Visual examples also showed a higher quality of images generated from the proposed method, demonstrating the denoising and improving ability of bi-c-GAN. By taking advantage of integrated PET/MR images and multitask deep learning (MDL), the proposed bi-c-GAN can efficiently improve the image quality of ultra-low-dose PET and reduce radiation exposure.

Detection of Fake Indian Currency Notes using Deep Learning

A normal human being can easily see and distinguish any banknote, however doing the same job is extremely difficult for someone who is visually challenged or blind. Because money plays such an essential part in our everyday lives and is required for any commercial transaction, real-time detection and recognition of banknotes is a must for anyone who is blind or visually impaired. The mobilenet based CNN model-based Indian currency detection and identification system is presented for this purpose, and it is quick and accurate. To make the system more resilient, pictures of various denominations and situations were collected first, and then these images were supplemented with various geometric and image modifications. These augmented pictures are then manually tagged, and training and validation image sets are created from them. Later, the trained model's performance was assessed on a real-time scenario as well as a test dataset. The suggested mobile net model-based technique exhibits detection accuracy of 91.33% according to the test results. This standalone system operates in real-time.

SEM-Net: Deep features selections with Binary Particle Swarm Optimization Method for classification of scanning electron microscope images

Materials Science is increasingly handling artificial intelligence methods to address the complexity in the field of everyday life necessities. Researchers in both academia and industry are interested in imaging techniques used in the characterization of nanomaterial with designed properties to meet the needs of applications in the literature. However, the increase in image size and complexity in its content restricts the use of traditional methods. Recent advances in machine learning have been used to benefit computers' potential to make sense of these images. The approach proposed in this paper aims for the feature reduction with the Binary Particle Swarm Optimization method to execute the classification process on SEM images by concatenating the deeper layers of pre-trained CNN models AlexNet and ResNet-50. The feature vectors were used as input to support vector machine classifier (SVMC) after dimension reduction to obtain the final model. Finally, the trained model's performance was tested using SEM images of Ag-doped SnO2 nanoparticles, which were prepared by the author using the low-temperature hydrothermal method. To the best of the author knowledge, these images were not available in the databases. The best accuracy value was observed with 3112 features for the SEM dataset with optimized vectors as 99.3 %. An example was illustrated where the feature selection with the BPSO technique could provide novel insight into nanoscience research and test the model with the SEM images of Ag-doped SnO2 particles that are obtained by the hydrothermal method.

Predictive AI platform on thin film evaporation in hierarchical structures

The trend in miniaturization and enhanced functional performance of integrated circuits and power electronics and photonics has amplified the generated thermal energy in these devices making thermal management a bottleneck for further advancement in these fields. A range of geometries of hierarchical structures are developed and examined to address this challenge. However, the numerous form factors and dimension of hierarchical structures in addition to cost and time-consuming synthesis and test procedures make it unfeasible to explore bountiful variations of hierarchical geometries through experimental methods. Here, we introduce a general Artificial Intelligence (AI) platform to address this challenge and guide discovery of hierarchical structures for extreme thermal management of high-performance photonics/electronics. The AI platform is based on Random Forest (RF) algorithm, a robust AI method, and was trained using a large collected experimental data set corresponding to thin film evaporation in various forms of hierarchical structures. Four geometrical dimensions of the hierarchical structures and two dimensionless numbers governing heat transfer and fluid dynamics in these structures were used as independent variables to predict heat flux in these structures. The trained model's performance was analyzed using statistical metrics and showed an excellent prediction of heat flux for all the structures with various working fluids. The performance of predictive AI platform was further validated by two independent studies of different research groups. This predictive platform provides a foundation for rational discovery of hierarchical structures and working fluids to address the ongoing challenge of thermal management in broad spectrum of technologies including electronics, hypersonic aviation and electric vehicles.

Artificial intelligence and its applications in ophthalmology

The term artificial intelligence (AI) was proposed in 1956 by Dartmouth scholar John McCarthy, which refers to hardware or software that exhibits behavior which appears intelligent.1 During recent times, AI gained immense popularity as new algorithms, specialized hardware, huge data and cloud-based services were developed. Machine learning (ML), a subset of AI, originated in 1980 and is defined as a set of methods that automatically detect patterns in data and then incorporate this information to predict future data under uncertain conditions. Another escalating technology of ML called Deep learning (DL), launched in 2000s, is an escalating technology of ML and has revolutionized the world of AI. These technologies are powerful tools utilized by modern society for objects' recognition in images, real-time languages' translation, device manipulation via speech (such as Apple's Siri®, Amazon’s Alexa®, Microsoft’s Cortana®, etc.). The steps for AI model include preprocessing image data, train, validate and test the model, and evaluate the trained model's performance. To increase AI prediction efficiency, raw data need to be preprocessed. Data collected from different sources needs to be integrated and the most relevant features selected and extracted to improve the learning process performance. Data set is randomly partitioned into two independent subsets, one is for modeling and the other is for testing. The test set is used to evaluate the final performance of the trained model. The area under receiver operating characteristic curves (AUC) is most used evaluation metrics for quantitative assessment of a model in AI diagnosis. The AUCs effective models range from 0.5 to 1; higher the value of AUC, better the performance of the model.2 In the medical field, AI gained popularity by visualization of input images of highly potential abnormal sites which can be reviewed and analyzed in future.
 AI and DL algorithms or systems are also widely used in field of ophthalmology. More intensively studied fields are diabetic retinopathy, age related macular degeneration, and cataract and glaucoma. Various ophthalmic imaging modalities used for AI diagnosis include fundus image, optical coherence tomography (OCT), ocular ultrasound, slit-lamp image and visual field. Diabetic retinopathy (DR), a diabetic complication, is a vasculopathy that affects one-third of diabetic patients leading to irreversible blindness. AI has been in use to predict DR risk and its progression. Gulshan and colleague were the first to report the application of DL for DR identification.3 They used large fundus image data sets in supervised manner for DR detection. Other studies applied DL to identify and stage DR. DL-based computer-aided system was introduced to detect DR through OCT images, achieving a specificity of 0.98.4 A computer-aided diagnostic (CAD) system based on CML algorithms using optical coherence tomography angiography images to automatically diagnose non-proliferative DR (NPDR) also achieved high accuracy and AUC.5 Age-related macular degeneration (AMD) is the leading cause of irreversible blindness among old people in the developed world. ML algorithms are being used to identify AMD lesions and prompt early treatment with accuracy usually over 80%.6 Using ML to predict treatment of retinal neovascularity in AMD and DR by anti-vascular endothelial growth factor (Anti VEGF) injection requirements can manage patients' economic burden and resource management. ML algorithms have been applied to diagnose and grade cataract using fundus images, ultrasounds images, and visible wavelength eye images.7 Glaucoma is the third largest sight-threatening eye disease around the world. Glaucoma patients suffered from high intraocular pressure, damage of the optic nerve head, retina nerve fiber layer defect, and gradual vision loss. Studies using DL methods to diagnose glaucoma are few. So far, fundus images and wide-field OCT scans have all been used to construct DL-based glaucomatous diagnostic models. Mostly, the DL-based methods show excellent results.8
 In this era of “evidence-based medicine,” clinicians and patients find it difficult to trust a mysterious machine to diagnose yet cannot provide explanations of why the patient has certain disease. In future, advanced AI interpreters will be launched which will contribute significantly to revolutionize current disease diagnostic pattern.