Incorrect Classification Research Articles

Automated Glaucoma Detection Using Vision and Swin Transformers: Advancing Ophthalmic AI

Purpose:Glaucoma is one of the most common causes of permanent blindness in the world; early detection and precise diagnosis are essential to successful treatment.Convolutional Neural Networks (CNNs) are one of the deep learning techniques that have shown excellent results in the processing of medical images. Methodology:Using a dataset of 1,650 fundus pictures from the REFUGE, ORIGA, and ACRIMA databases600 glaucoma-positive and 1,050 glaucoma-negative samplesthis study assesses the effectiveness of three cutting-edge deep learning models for glaucoma classification. Pretrained models like ResNet-50, VGG16, GoogLeNet, Vision Transformer (ViT), and Swin Transformer are investigated, emphasizing on their capacity to extract key variables such as the optic cup-to-disc ratio, retinal nerve fiber layer thickness, and vascular patterns.Result:Swin Transformer outperformed other models, achieving 100% accuracy, precision, recall, and F1-score in perfect classification. ViT and GoogLeNet similarly showed remarkable performance, achieving 92.92% and 91.54% accuracy. ResNet-50 and VGG16, on the other hand, had lower accuracy percentages of 74.62% and 78.77%. A number of drawbacks were found in all the models. ResNet-50 suffered from underfitting, which resulted in incorrect classifications and lower validation accuracy. While VGG16 was effective in standard image classification tasks, it showed inadequate recall and substantial validation loss, especially in situations when the patient had glaucoma. GoogLeNet struggled with overfitting, which limited its ability to generalise to new data, whereas ViT needed a lot of processing power and initially had trouble correctly categorising cases of glaucoma.On the other hand, Swin Transformer used hierarchical feature maps and changing windows to efficiently capture both local and global picture information, such as blood vessel patterns and the structure of the optic nerve head. A confusion matrix verified that there were no misclassifications in the model's flawless generalisation. Conclusion:In summary, the study found that Swin Transformer was the most dependable and resilient model for glaucoma diagnosis, even though models like as ViT and GoogLeNet shown potential. This study highlights the promise of transformer-based topologies, in particular Swin Transformer, as a state-of-the-art remedy for ophthalmology-related medical image classification challenges.

Machine learning models for sequential motion recognition in human activity for logistics operations efficiency improvement

Abstract Human activity recognition (HAR) is a vital component of human–robot collaboration. Recognizing the operational elements involved in an operator’s task is essential for realizing this vision, and HAR plays a key role in achieving this. However, recognizing human activity in an industrial setting differs from recognizing daily living activities. An operator’s activity must be divided into fine elements to ensure efficient task completion. Despite this, there is relatively little related research in the literature. This study aims to develop machine learning models to classify the sequential movement elements of a task. To illustrate this, three logistic operations in an integrated circuit (IC) design house were studied, with participants wearing 13 inertial measurement units manufactured by XSENS to mimic the tasks. The kinematics data were collected to develop the machine learning models. The time series data preprocessing involved applying two normalization methods and three different window lengths. Eleven features were extracted from the processed data to train the classification models. Model validation was carried out using the subject-independent method, with data from three participants excluded from the training dataset. The results indicate that the developed model can efficiently classify operational elements when the operator performs the activity accurately. However, incorrect classifications occurred when the operator missed an operation or awkwardly performed the task. RGB video clips helped identify these misclassifications, which can be used by supervisors for training purposes or by industrial engineers for work improvement.

Enhancing Smart Grid Security Using BLS Privacy Blockchain With Siamese Bi‐LSTM for Electricity Theft Detection

ABSTRACTEnergy management inside a blockchain framework developed for smart grids is primarily concerned with improving intrusion detection to protect data privacy. The emphasis is on real‐time detection of cyberattacks and preemptive forecasting of possible risks, especially in the realm of electricity theft within smart grid systems. Existing Electricity Theft Detection techniques for smart grids have obstacles such as class imbalance, which leads to poor generalization, increased complexity due to large EC data aspects, and a high false positive rate in supervised models, resulting in incorrect classification of regular customers as abnormal. To provide security in the smart grid, a novel BLS Privacy Blockchain with Siamese Bi‐LSTM is proposed. Initially, the privacy‐preserving Boneh‐Lynn‐Shacham blockchain technique is built on BLS Short signature and hash algorithms, which mitigate misclassification rates and false positives in the detection of smart grid attacks. Then, a hybrid framework employs an intrusion detection algorithm based on Siamese Bidirectional Long Short‐Term Memory to semantically distinguish between harmful and authentic behaviors, thereby improving data quality and predictive capabilities. Furthermore, a Recurrent Neural Network‐Generative Adversarial Network is presented for detecting electricity fraud, which addresses the issue of class imbalance. This uses both supervised and unsupervised loss functions to produce synthetic theft samples that closely resemble actual theft incidents. From the experiment, it is showing that the proposed models perform with high accuracy and low error rates. The proposed model from the outcomes when compared to other existing models achieves high accuracy, detection rate, recall, and low computation time.

Institute of Technical Medicine (ITeM), Hochschule Furtwangen University (HFU), Germany

Abstract Machine learning algorithms and neural networks have recently been used for the classification of middle ear disorders using wideband acoustic immittance and wideband tympanometry data. This study applies the extreme gradient boosting (XGB) classifier, trained on simulated WAI data, to classify real measured data for normal, otosclerotic, and disarticulated ears. The achieved macro recall of 82 % is comparable to other approaches trained with real measurement data. The interpretability methods LIME and SHAP are used to quantify each feature’s contribution, both revealing energy reflectance between 600-800 Hz as a key feature for all classes. The key feature identified matches the differences that can be visually observed in the training and test data. However, the obtained feature contributions don’t provide enough distinguishable information to recognise incorrect or uncertain classifications.

Multifunctional adversarial examples: A novel mechanism for authenticatable privacy protection of images

With the rapid development of network technology, more and more images containing personal identity characteristics are being released by users on open network platforms. However, these images are easily collected by malicious users, leading to problems such as privacy leakage, infringement, and tampering, thus harming users’ legitimate interests. Recent studies have found that adversarial examples generated by adding tiny perturbations to an image can mislead image classifiers, causing incorrect classifications. Therefore significant privacy protection against deep neural networks is achieved while the visual quality remains indistinguishable to human eyes. However, these methods cannot protect the authenticity and integrity of the image simultaneously, failing to address infringement and tampering issues, which are also neglectable in the open network platforms. To solve this problem, we propose a novel authentication-enabled privacy protection method. The meaningful information used for authentication, instead of the meaningless perturbations, is embedded into the host image to generate adversarial examples, thereby achieving both authentication and privacy protection simultaneously. This scheme combines attention mechanisms with generative adversarial networks to adaptively select and weight features between different channels, achieving significant improvements in both aggressiveness and authentication capability. Experimental results show that our method outperforms recent similar methods in overall performance.

Integrating agent-based models and clustering methods for improving image segmentation

Image segmentation through clustering is a widely used technique in computer vision that partitions an image into multiple segments by grouping pixels based on feature similarity. Although effective for certain applications, this approach often struggles with the complexity of real-world images, where noise and random variations can significantly affect feature homogeneity, leading to incorrect pixel classifications. To address these limitations, this paper introduces a novel hybrid image segmentation method that combines an agent-based model with a clustering technique to enhance segmentation accuracy and robustness. The method starts with an agent-based model as a preprocessing step aimed at homogenizing pixel intensities within each region. In this model, pixels adjust their intensities based on a consensus reached within their neighborhood, promoting a more uniform feature distribution. Subsequently, the Firefly metaheuristic clustering method is applied to segment the preprocessed image into distinct regions. Metaheuristic techniques, distinguished from classical clustering methods, possess the capability to adaptively navigate through a broad solution space to discover optimal clustering configurations. This adaptability makes them suitable for complex image datasets. The efficacy of the proposed hybrid segmentation method has been tested on various images, employing key quality indices for evaluation. Experimental outcomes demonstrate that this approach yields superior segmented images, showcasing enhanced quality and robustness compared to other segmentation methods.

Towards explainable evaluation: Explaining predicted performance using local performance regions

Identifying which instances in a learning problem are difficult to be predicted by a model is important to avoid critical errors at deployment time as well as to plan how to learn an improved model (e.g., by training data cleaning or augmentation). Previous works have been mainly devoted on measuring instance hardness or developing meta-learners (e.g., assessors) to predict a base model performance based on the instances’ features while neglecting interpretability. In this paper, we propose a method to explain the performance of learned models in a problem based on the induction of meta-rules. Each meta-rule identifies a local region of instances, called Local Performance Region (LPR), where the base model has a predictable performance. The meta-rules are induced using a reduced number of attributes, in such a way that each LPR can be more easily inspected (e.g., by an attribute plot). The proposed method combines assessors, data augmentation and rule induction procedures. Initially, given a dataset of interest and a base model, we build an assessor model, which will be able to predict the base model’s performance for new instances. The assessor is trained based on the test results obtained when the base model is evaluated, thus generalizing the observed errors across instances in the dataset. In our work, we built an assessor in a case study to predict the probability of incorrect classifications of a Random Forest (RF) base model, achieving a mean absolute error of 0.05 in a hold out experiment. Once learned, the assessor is used to predict the model’s errors for new instances in an augmented dataset, covering a variety of features. Finally, meta-rules are learned to approximate the assessor’s predictions in local regions of instances. Experiments show the usefulness of the proposal by finding 18 local regions of bad RF performance, demonstrating a special case of LPRs, called Local Hard Regions (LHRs). By explaining the (in)correctness of model predictions, LPRs constitute a novel application in explainable AI, but focusing on explaining model performance, which can be adopted to different ML contexts.

Inventory classification with artificial intelligence: Conceptual framework

In today's world where competition is increasing harshly, it is important to achieve sustainable profits and keep costs competitive. Under these conditions, the importance of supply chain elements is increasing day by day. Management of inventory costs, which constitute a large volume among cost items, affects the performance of companies. The first step to focus on to keep inventory costs under control and improve is inventory classification. Inventory classification, which is at the top of the supply chain elements and closely affects the subsequent phases, it is critical in determining supply chain performance. Thanks to inventory classification, material groups are determined and stock strategies for these groups are clarified. Incorrect inventory classification causes materials to be assigned to incorrect groups, which negatively affects inventory costs and subsequent phases of the supply chain, causing an increase in costs. The most used methods for inventory classification are ABC analysis, multi-criteria inventory classification and optimization. However, the increasing momentum in artificial intelligence studies in recent years has also closely affected inventory classification. The advantages brought by artificial intelligence methods have also created distinctive contributions to inventory classification studies. This study provides a conceptual framework that examines artificial intelligence methods in the field of inventory classification.

Data Balancing Approach Using Combine Sampling on Sentiment Analysis With K-Nearest Neighbor

One of the topics that has been discussed on twitter is the rules regarding the removal of masks. However, there's a chance that the data from Twitter contains unequal data classes. An unequal amount of data can cause the classification process to malfunction. Combining under- and oversampling techniques is known as combine sampling, and it is a data-balancing strategy. The research's data consists of Indonesian tweets using the hashtag "The Policy of Removing Masks." In this study, the classification approach was K-Nearest Neighbor, while the oversampling and undersampling techniques were SMOTE and Tomek Links. The purpose of this research is to classify sentiment using the K-Nearest Neighbor algorithm and to use combine sampling to balance the amount of training data in the two classes that are not yet balanced. 234 training data with a positive sentiment and 652 training data with a negative sentiment were obtained after the data was divided. Due to an imbalance in the quantity of training data between the two classes, the positive class's data is minor and the negative class's data is major. The quantity of training data are 613 in the positive class and 613 in the negative class obtained following the combine sampling. Following the balancing of data between the two classes, sentiment classification was performed, yielding accuracy of 60.4%, precision of 78.5%, and recall of 65%. The reason for the accuracy number of 60.4% is because machine learning misinterpreted a tweet regarding Indonesia's mask removal policy, leading to incorrect classification.

Safe and Reliable Movement of Fast LiDAR-based Self-driving Vehicle

Classification of objects is an important technique for autonomous ground vehicles to identify a surrounding environment and execute safe path planning. In this paper, a method based on horizontal segmentation is proposed to detect cone-shaped objects in vehicle’s vicinity using a LiDAR sensor. A captured point cloud is divided into five layers based on height information, and the division of detected objects into two groups, cones and others, has been made using classifiers available in MATLAB toolboxes. To separate the classified conical objects into four types used to mark the route, an algorithm for their recognition was developed and used. The proposed solution, verified by navigation experiments in real conditions using an unmanned racing car, has gave good results, i.e., a high rate of cone-shaped objects classification, a short processing time and a low computational load. The performed tests have allowed also to diagnose the causes of incorrect classification of objects. Thus, the experimental results indicated that the approach presented in this work can be used in real time for autonomous, collision-free driving along marked routes.

Towards universal and transferable adversarial attacks against network traffic classification

In recent years, deep learning technology has shown astonishing potential in many fields, but at the same time, it also hides serious vulnerabilities. In the field of network traffic classification, attackers exploit this vulnerability to add designed perturbations to normal traffic, causing incorrect network traffic classification to implement adversarial attacks. The existing network traffic adversarial attack methods mainly target specific models or sample application scenarios, which have many problems such as poor transferability, high time cost, and low practicality. Therefore, this article proposes a method towards universal and transferable adversarial attacks against network traffic classification, which can not only perform universal adversarial attacks on all samples in the network traffic dataset, but also achieve cross data and cross model transferable adversarial attacks, that is, it has transferable attack effects at both the network traffic data and classification model levels. This method utilizes the geometric characteristics of the network model to design the target loss function and optimize the generation of universal perturbations, resulting in biased learning of features at each layer of the network model, leading to incorrect classification results. Meanwhile, this article conducted universality and transferability adversarial attack verification experiments on standard network traffic datasets of three different classification applications, USTC-TFC2016, ISCX2016, and CICIoT2023, as well as five common network models such as LeNet5. The results show that the proposed method performs universal adversarial attacks on five network models on three datasets, USTC-TFC2016, ISCX2016, and CICIoT2023, with an average attack success rate of over 80 %, 85 %, and 88 %, respectively, and an average time cost of about 0–0.3 ms; And the method proposed in this article has shown good transferable attack performance between five network models and on three network traffic datasets, with transferable attack rates approaching 100 % across different models and datasets, which is more closely related to practical applications.

Impaired or not impaired: The accuracy of the Montreal Cognitive Assessment in detecting cognitive impairment among patients with alcohol use disorder.

Cognitive impairments are common in alcohol use disorder (AUD), but only a few studies have investigated the accuracy of the Montreal Cognitive Assessment (MoCA) in this population. We examined the accuracy and precision of the MoCA in detecting cognitive impairment in a sample of patients with AUD. In addition, we investigated whether the MoCA predicts premature discontinuation from treatment. A sample of 126 persons with AUD undergoing treatment in specialist health services were administered the MoCA and a battery of 12 neuropsychological tests. Five cognitive domains were derived from the reference tests. A composite total score from these tests was used as a reference criterion for determining correct and incorrect classifications for the MoCA. We analyzed the optimal cut-off score for the MoCA and the accuracy and agreement of classification between the MoCA and the reference tests. Receiver operating characteristic (ROC) curve analyzes yielded an area under the curve (AUC) of 0.77 (95% CI [0.67, 0.87]). Applying 25 as the cut-off, MoCA sensitivity was 0.77 and specificity 0.62. The PPV was 0.53. The NPV was 0.84. Using a cut-off score of 24 yielded a lower sensitivity 0.60, but specificity was significantly better i.e., 0.79. PPV was 0.68. The NPV was 0.82. Kappa agreement between MoCA and the reference tests was fair to moderate, 0.38 for the cut-off of 25, and 0.44 for the cut-off of 24. MoCA did not predict discontinuation from treatment. Our findings indicate limitations in the classification accuracy of the MoCA in predicting cognitive impairment in AUD. Achieving the right balance between accurately identifying impaired cases without including too many false positives can be challenging. Further, MoCA does not predict discontinuation from treatment. Overall, the results do not support MoCA as a time-efficient screening instrument.

Infection Inspection: using the power of citizen science for image-based prediction of antibiotic resistance in Escherichia coli treated with ciprofloxacin

Antibiotic resistance is an urgent global health challenge, necessitating rapid diagnostic tools to combat its threat. This study uses citizen science and image feature analysis to profile the cellular features associated with antibiotic resistance in Escherichia coli. Between February and April 2023, we conducted the Infection Inspection project, in which 5273 volunteers made 1,045,199 classifications of single-cell images from five E. coli strains, labelling them as antibiotic-sensitive or antibiotic-resistant based on their response to the antibiotic ciprofloxacin. User accuracy in image classification reached 66.8 ± 0.1%, lower than our deep learning model's performance at 75.3 ± 0.4%, but both users and the model were more accurate when classifying cells treated at a concentration greater than the strain’s own minimum inhibitory concentration. We used the users’ classifications to elucidate which visual features influence classification decisions, most importantly the degree of DNA compaction and heterogeneity. We paired our classification data with an image feature analysis which showed that most of the incorrect classifications happened when cellular features varied from the expected response. This understanding informs ongoing efforts to enhance the robustness of our diagnostic methodology. Infection Inspection is another demonstration of the potential for public participation in research, specifically increasing public awareness of antibiotic resistance.

Optimization of activity-driven event detection for long-term ambulatory urodynamics.

Lower urinary tract dysfunction (LUTD) is a debilitating condition that affects millions of individuals worldwide, greatly diminishing their quality of life. The use of wireless, catheter-free implantable devices for long-term ambulatory bladder monitoring, combined with a single-sensor system capable of detecting various bladder events, has the potential to significantly enhance the diagnosis and treatment of LUTD. However, these systems produce large amounts of bladder data that may contain physiological noise in the pressure signals caused by motion artifacts and sudden movements, such as coughing or laughing, potentially leading to false positives during bladder event classification and inaccurate diagnosis/treatment. Integration of activity recognition (AR) can improve classification accuracy, provide context regarding patient activity, and detect motion artifacts by identifying contractions that may result from patient movement. This work investigates the utility of including data from inertial measurement units (IMUs) in the classification pipeline, and considers various digital signal processing (DSP) and machine learning (ML) techniques for optimization and activity classification. In a case study, we analyze simultaneous bladder pressure and IMU data collected from an ambulating female Yucatan minipig. We identified 10 important, yet relatively inexpensive to compute signal features, with which we achieve an average 91.5% activity classification accuracy. Moreover, when classified activities are included in the bladder event analysis pipeline, we observe an improvement in classification accuracy, from 81% to 89.0%. These results suggest that certain IMU features can improve bladder event classification accuracy with low computational overhead.Clinical Relevance: This work establishes that activity recognition may be used in conjunction with single-channel bladder event detection systems to distinguish between contractions and motion artifacts for reducing the incorrect classification of bladder events. This is relevant for emerging sensors that measure intravesical pressure alone or for data analysis of bladder pressure in ambulatory subjects that contain significant abdominal pressure artifacts.

Brain Tumor Detection using Hybrid CNN in fMRI images

One of the deadliest brain tumors in the world. Both adults and children are at risk for the disease. Depending on their position, appearance and shape, it has a low survival rate and is available in different varieties. Incorrect classification of the brain tumor will have negative consequences. Early detection of the tumor type and stage is necessary for selecting a specific treatment. Identifying brain tumors can be done effectively by examining fMRI, or functional magnetic resonance imaging. Due to the various information and various types of tumors in brain, the manual process takes lot of time to identify, and there may be chances of human error. Therefore, it is necessary to have an intelligent diagnostic system. The recent evolution of image categorization systems has been influenced by the use of deep convolution neural networks (CNN). In this research work, a new hybrid paradigm with images combines a convolution neural network (CNN) and a masked autoencoder for distribution estimation (MADE). Three different forms of brain tumors were tested using T1-weighted contrast enhanced images. The CNN-MADE hybrid's excellent classification performance is demonstrated by the results, despite the limited availability of medical images. A new layered model was proposed to categorize brain tumors through MRI. The proposed approach is more successful than the alternatives based on the experiments' results.

Use of whistles for acoustic classification of delphinids (odontoceti: Delphinidae) in the Western South Atlantic Ocean.

This study focuses on the acoustic classification of delphinid species at the southern continental slope of Brazil. Recordings were collected between 2013 and 2015 using towed arrays and were processed using a classifier to identify the species in the recordings. Using Raven Pro 1.6 software (Cornell Laboratory of Ornithology, Ithaca, NY), we analyzed whistles for species identification. The random forest algorithm in R facilitates classification analysis based on acoustic parameters, including low, high, delta, center, beginning, and ending frequencies, and duration. Evaluation metrics, such as correct and incorrect classification percentages, global accuracy, balanced accuracy, and p-values, were employed. Receiver operating characteristic curves and area-under-the-curve (AUC) values demonstrated well-fitting models (AUC ≥ 0.7) for species definition. Duration and delta frequency emerged as crucial parameters for classification, as indicated by the decrease in mean accuracy. Multivariate dispersion plots visualized the proximity between acoustic and visual match data and exclusively acoustic encounter (EAE) data. The EAE results classified as Delphinus delphis (n = 6), Stenella frontalis (n = 3), and Stenella longirostris (n = 2) provide valuable insights into the presence of these species between approximately 23° and 34° S in Brazil. This study demonstrates the effectiveness of acousting classification in discriminating delphinids through whistle parameters.

Estimating synthetic hematocrit and extracellular volume from native blood pool T1 times at 3 Tesla CMR: Derivation of a conversion equation, accuracy and comparison with published formulas

PurposeCalculation of extracellular volume fraction (ECV), a marker of myocardial fibrosis in cardiac magnetic resonance imaging (CMR), requires hematocrit (Hct). We aimed to correlate Hct levels with native blood T1 times, to derive a formula for estimating synthetic Hct (Hctsyn) and synthetic ECV (ECVsyn), to assess accuracy of ECVsyn and to compare our model with published formulas. MethodIn this retrospective study, a cohort of 250 CMR scans with T1 mapping (3T, MOLLI 5(3)3, endsystolic aquisition), was divided into a derivation and validation cohort. Native T1 times of the left ventricular blood pool (T1native,midLV) were correlated with Hct levels from blood sampling within 24 h (Hct24h) and a formula for calculation of Hctsyn was derived by linear regression. ResultsIn the derivation cohort (n = 167), Hct24h showed a good association with T1native,midLV (r = −0.711, p < 0.001). The resulting regression equation was Hctsyn = 1/T1native,midLV * 1355.52–0.310. In the validation cohort (n = 83), Hctsyn and Hct24h showed good correlation (r = 0.726, p < 0.001), while ECVsyn, and ECV24h demonstrated excellent correlation (r = 0.940, p < 0.001). ECVsyn had a minimal bias of 0.28 % and the misclassification rate (8.8 %) was comparable to the variability introduced by repeated Hct measurements (misclassification in 7.5 %). Applying published formulas in our cohort resulted in incorrect classification in up to 60 %. ConclusionWe provide a formula for estimating Hctsyn from native blood T1 on a 3T scanner. The measurement error of ECVsyn is low and comparable to the error due to retest variability of conventional Hct. Scanner- and sequence-specific formulas should be used.

Spirometry findings of chronic lung disease in high-altitude residents of Ladakh (&gt;11000 feet above sea level)

Ladakh is a hilly Himalayan dry desert, situated at an altitude of &gt;11000 feet. Studies have demonstrated that the spirometric values of high-altitude residents are significantly higher than those of low-landers. This is a retrospective observational study that analyzes the spirometry pattern in chronic lung diseases among people from Ladakh. Enrolled subjects were clinic-radiologically diagnosed and had at least one spirometry report. The spirometric parameters were analyzed for normal and abnormal patterns of lung function. The abnormal patterns were further classified into types of ventilator defects and their severity. A total of 122 cases were included, with 67 (55%) men. The mean age was 52.2±15.4 years. The most common diseases were chronic obstructive pulmonary disease (COPD) in 51 cases (41%), and asthma in 41 (33%). The median predicted percentage of forced vital capacity (FVC) was 116% (63-179%) with &gt;100% in 105 (85%) patients. The median predicted percentage of the forced expiratory volume in the 1st second (FEV1) was 113% (99-175%) with &gt;100% in 90 (74%) patients. FVC was reduced in 9 (7%) cases, normal in 62 (51%), and more than normal in 49 (42%), with 11 (9%) cases having &gt;150% of the predicted percentage. FEV1 was reduced in 9 (8%) cases, normal in 67 (55%), and more than normal in 46 (37%) cases, with &gt;150% predicted seen in 10 (8%) cases. Similarly, overall, the predicted percentages of both FVC and FEV1 were &gt;100% in all obstructive airway diseases as well as in the separate COPD and asthma subgroups. FVC and FEV1 amongst chronic lung disease patients from Ladakh were more than normal in the majority. These higher values of spirometry led to incorrect disease severity classifications and disease patterns. We propose that studies should be done to devise local reference equations for spirometry for Himalayan high-altitude residents of India.

An expert opinion on rescuing atypically pronucleated human zygotes by molecular genetic fertilization checks in IVF.

IVF laboratories routinely adopt morphological pronuclear assessment at the zygote stage to identify abnormally fertilized embryos deemed unsuitable for clinical use. In essence, this is a pseudo-genetic test for ploidy motivated by the notion that biparental diploidy is required for normal human life and abnormal ploidy will lead to either failed implantation, miscarriage, or significant pregnancy complications, including molar pregnancy and chorionic carcinoma. Here, we review the literature associated with ploidy assessment of human embryos derived from zygotes displaying a pronuclear configuration other than the canonical two, and the related pregnancy outcome following transfer. We highlight that pronuclear assessment, although associated with aberrant ploidy outcomes, has a low specificity in the prediction of abnormal ploidy status in the developing embryo, while embryos deemed abnormally fertilized can yield healthy pregnancies. Therefore, this universal strategy of pronuclear assessment invariably leads to incorrect classification of over 50% of blastocysts derived from atypically pronucleated zygotes, and the systematic disposal of potentially viable embryos in IVF. To overcome this limitation of current practice, we discuss the new preimplantation genetic testing technologies that enable accurate identification of the ploidy status of preimplantation embryos and suggest a progress from morphology-based checks to molecular fertilization check as the new gold standard. This alternative molecular fertilization checking represents a possible non-incremental and controversy-free improvement to live birth rates in IVF as it adds to the pool of viable embryos available for transfer. This is especially important for the purposes of 'family building' or for poor-prognosis IVF patients where embryo numbers are often limited.

Impact the Classes’ Number on the Convolutional Neural Networks Performance for Image Classification

Deep learning was developed as a realistic artificial intelligence technique that takes in numerous layers of information and produces the best results in various classes. Deep learning has demonstrated excellent performance in several areas, particularly picture grouping, division, and recognition. The convolutional neural network (CNN) is one of the algorithms that relies on deep learning in its work. It has proven its effectiveness in classifying images with high efficiency in medical images and their diagnoses, face recognition, and other different fields. In this paper, the focus was on images to alert new researchers to their effects on the performance of CNN in terms of the number of classes that existed within the database, in addition to the impact of incorrect classification of images by the source on the classification result and the necessity of adopting reliable and correct sources of data to avoid inaccurate results. A group of face images has been used, and three experiments on them were conducted using all existing classes with reduction. The results showed a significant improvement in the performance of the algorithm whenever the number of classes was reduced. The best result was when only two classes were chosen for classification, reaching a validation accuracy of 85%.