Classification Model Research Articles

A novel automatic locating method for pylorus and ileocecal valve in wireless capsule endoscopy

Wireless capsule endoscopy (WCE) plays an important role in diagnosing gastrointestinal tract diseases. Efficiently pinpointing the pylorus and ileocecal valve, which mark the start and end of the small intestine, is vital for clinicians determining the location of abnormalities in gastrointestinal tract as soon as possible during burdensome reviewing work. In this study, a novel and fault-tolerant automatic locating system for pylorus and ileocecal valve was proposed, combining a ResNet-50-based gastrointestinal classification model and the sliding pin algorithm. The classification model classified the WCE images into three categories, separated by the small intestine. And then the sliding pin algorithm was created to pinpoint the location of the pylorus and ileocecal valve even if there were misclassifications. A total of 108,542 WCE images were used to train and test the gastrointestinal classification model. The average accuracy of classification model was 92.21% with recognition rate varying from 88% to 96%. Furthermore, thirty-one independent WCE cases were selected to verify the proposed locating method. The mean locating error was 26 frames for the pylorus and 65 frames for the ileocecal valve, which was notably smaller than those of other state-of-the-art methods. In conclusion, this study presented an innovative approach for automatically locating the boundary of small intestine in WCE. The promising results indicated that this method was able to accurately identify the pylorus and ileocecal valve, which was expected to improve the endoscopic diagnostic efficiency of gastrointestinal diseases.

A comparative analysis of trauma-related mortality in South Korea using classification models

A comparative analysis of trauma-related mortality in South Korea using classification models

A novel MSN-II feature extracted from T1-weighted MRI for discriminating between BD patients and MDD patients.

Differentiating between patients with bipolar disorder (BD) and major depressive disorder (MDD) is clinically challenging. This study aimed to explore the potential of radiomic textural features for discriminating BD and MDD. A total 253 subjects (114 patients with BD, 139 patients with MDD) with T1-weighted MRI data were recruited. Radiomics features and gray matter volume (GMV) features were extracted from each brain region. A novel high-level MSN_II feature method based on radiomic features was proposed. And a total of 21 MSN features (5 MSN_I and 16 MSN_II) based on different combinations of the 5 types of radiomic textural feature were calculated. Classification models were constructed using various combinations of MSNs or GMV, and their performance and stability was evaluated through 2000 repeated experiments. The model built with combined features (GMV and GMV+MSN_II_GLCM_GLSZM_NGTDM) showed the best classification performance (AUC=0.896±0.058, ACC=0.831±0.064) in the validation cohort. After MANOVA analysis and FDR correlation, the MSN_II_GLCM_GLSZM_NGTDM values in 4 regions (right rectus gyrus, right temporal pole: middle temporal gyrus, Vermis3 and Vermis10) showed significant difference between BD and MDD. The main limitation of this study is that the data is derived from a single center without an external independent test set. Incorporating the high-level MSN_II based on radiomics features can improve the classification performance compared to models solely relying on GMV features alone. This result implied the potential application of the proposed high level MSN method and radiomics textural features on the MDD and BD clinical studies.

High-risk patient profiles for ovarian cancer: a new approach using cluster analysis of tumor markers

ObjectiveOvarian cancer remains a leading cause of cancer-related deaths in women. Early detection improves prognosis, but current diagnostic tools still need improvement. We aimed to identify high-risk patient profiles for ovarian cancer using cluster analysis of age and tumor marker data. Material and MethodsA secondary dataset analysis was conducted using unsupervised learning techniques. Data were from a University Hospital, originally collected between July 2011 and July 2018 in Taiwan. In total, 349 women diagnosed with ovarian masses, including both benign and malignant tumors, were included in this analysis. The median age was 45 years, and 49% were diagnosed with ovarian cancer in pathology. We used a hierarchical clustering algorithm to find groups of patients with similar features. ResultsTwo clusters were identified (N = 204 and 145), with a high-risk cluster (66.2% malignancy) characterized by significantly older age, higher CA125, HE4, CEA, and AFP levels, and a lower CA19-9 level than the low-risk cluster (24.8% malignancy). The assessment of clustering stability and internal validity yielded a figure of merit score of 0.970 and a silhouette coefficient of 0.524. A classification model using age, CA125, HE4, and CA19-9 demonstrated high accuracy (89.4%), sensitivity (94.5%), specificity (83.7%), and a large area under the curve (89.1%) for the risk stratification. ConclusionIntegrating tumor markers with patient demographics improved the differentiation between benign and malignant ovarian masses. This approach can help clinicians prioritize high-risk patients for further diagnostic evaluation and reduce unnecessary invasive procedures for low-risk patients.

Detecting mild cognitive impairment by applying integrated random forest to finger tapping.

Early detection of dementia is essential to reduce the decline in quality of life (QoL) and the increase in medical and nursing care costs associated with dementia in an aging society. In this study, we aimed to develop a simple screening test for mild cognitive impairment (MCI), a preliminary stage of dementia, by creating an analytical method to accurately detect MCI through finger-tapping measurement. We extracted 248 characteristics from the finger-tapping waveforms of 182 MCI patients and 352 normal controls, applying five conventional classification methods along with an improved Random Forest (RF) method proposed in this study (Integrated RF). In the proposed method, the RF classification model for the MCI and normal control groups is supplementally integrated with the RF classification model for the Alzheimer's disease and normal control groups to generate a new classification model. When comparing the discrimination accuracy of each method, the proposed method achieved the highest accuracy, with an F1-score of 0.795 (recall = 0.778 and precision = 0.814). These results demonstrate the potential of finger-tapping measurement as a highly accurate screening test for MCI.

Artificial intelligence aided serum protein electrophoresis analysis of Finnish patient samples: Retrospective validation.

Serum protein electrophoresis interpretation requires a substantial amount of manual work. In 2020, Chabrun et al. created a machine learning method called SPECTR for the task. We aimed to validate and test the SPECTR method against our results of more precise immunofixation electrophoresis. We gathered 34 625 patients and their first serum protein electrophoresis sample in Helsinki University Hospital. We trained three neural network models: (1) a fractionation model to fractionate electropherograms; (2) a classification model to classify samples to normal, ambiguous, and abnormal (i.e. containing paraprotein); (3) an integration model to predict concentration and location of paraproteins. The fractionation model demonstrated an error rate of ≤0.33g/L in 95% samples. The classification model achieved an area under the curve of 97% in receiver operating characteristic analysis. The integration model demonstrated a coefficient of determination (R2) of 0.991 and a root-mean-square error of 1.37g/L in linear regression. The neural network models proved to be suitable for partial automation in serum protein electrophoresis reporting, i.e. classification of normal electropherograms. Furthermore, the models can accurately suggest the location and concentration of paraproteins.

Tracking land use land cover changes in the twin cities of Odisha, India using a machine learning based Google Earth Engine approach

ABSTRACT The current study is based on analyzing the land use land cover (LULC) changes and its corresponding effects on water and land surface temperature (LST) on the twin cities of Odisha, i.e. Bhubaneswar and Cuttack using a machine learning based Google Earth Engine (GEE) platform. A random forest (RF) classification model was adopted due to its simplicity and high popularity for providing accurate results. For the study, Landsat 8 (OLI/TRIS) and Sentinel 2 were accessed via GEE. With an overall accuracy of about 99% using an RF algorithm, the results indicate an alarming situation for the cities, especially Cuttack where there has been a reduction in water by about 59% in response to increments in the built-up area by 90% and LST by 1.5%. With an expanding city radius, Bhubaneswar faced a reduction in water by 28% in response to the built-up area and LST increase by about 17% and 3.4%. respectively.

Recognition of autism in subcortical brain volumetric images using autoencoding-based region selection method and Siamese Convolutional Neural Network

BackgroundAutism Spectrum Disorder (ASD) is a neurodevelopmental condition that affects social interactions and behavior. Accurate and early diagnosis of ASD is still challenging even with the improvements in neuroimaging technology and machine learning algorithms. It’s challenging because of the wide range of symptoms, delayed appearance of symptoms, and the subjective nature of diagnosis. In this study, the aim is to enhance ASD recognition by focusing on brain subcortical regions, which are critical for understanding ASD pathology. MethodologyFirst, subcortical structures were extracted from a collection of brain MRI datasets using sophisticated processing steps. Next, a 3D autoencoder was trained on these 3D images to help identify brain regions related to ASD. Two distinct feature selection methods were then applied to the features extracted from the encoder. The highest-ranked features were iteratively selected and increased to reconstruct a specific percentage of the brain that represents the most relevant parts for ASD. Finally, a Siamese Convolutional Neural Network (SCNN) was employed as the classifier model. ResultsThe 3D autoencoder stage helped in identifying and reconstructing the significant subcortical regions related to ASD. Based on the studied dataset, high agreement in regions like the Putamen and Pallidum indicated the critical nature of these structures in distinguishing Autism from controls cases. Subsequently, applying SCNN on these selected subcortical regions yielded promising results. For example, using the classifier on the output regions identified by the Mutual Information (MI) features selection method achieved the highest accuracy of 0.66. ConclusionsThis study shows that using a two-stage model involving autoencoder and SCNN can notably improve the classification of ASD from brain MRI volumetric images. Applying an iterative feature extraction approach allowed to achieve a more accurate identification of ASD-related brain areas. This two-stage approach not only improved classification performance but also enhanced the interpretability of the neuroimaging data.

ECCDN-Net: A deep learning-based technique for efficient organic and recyclable waste classification.

Efficient waste management is essential to minimizing environmental harm as well as encouraging sustainable progress. The escalating volume and sophistication of waste present significant challenges, prompting innovative methods for effective waste categorization and management. Deep learning models have become highly intriguing tools for automating trash categorization activities, providing effective ways to optimize processes for handling waste. Ourwork presents a novel deep learning method for trash classification, with the goal to improve the accuracy, also efficiency of garbage image categorization. We examined the effectiveness of several pre-trained models, such as InceptionV2, Densenet201, MobileNet v2, and Resnet18, using objective evaluation and cross-validation. We proposed an Eco Cycle Classifier Deep Neural Network (ECCDN-Net) model that is particularly built for the categorization of waste images. ECCDN-Net utilizes the advantageous qualities of Densenet201 and Resnet18 by merging their capacities to extract features, enhanced with auxiliary outputs to optimize the classification procedure. The set of imagesused in this study comprises 24,705 images that are divided into two distinct classes: Organic and Recyclable. The set allows extensive evaluation and training of deep learning models for waste classification of images tasks. Our research demonstrates that the ECCDN-Net model classifies waste images with 96.10% accuracy, outperforming other pre-trained models. Resnet18 had 92.68% accuracy, MobileNet v2 93.27%, Inception v3 94.77%, and Densenet201, a significant improvement, 95.98%. ECCDN-Net outperformed these models in waste image categorization with 96.10% accuracy. We ensure the reliability and generalizability of our methods throughout the dataset by integrating and cross-validating deep learning models. The current work introduces an innovative deep learning-based approach that has promising potential for waste categorization and management strategies.

Revolutionizing air quality forecasting: Fusion of state-of-the-art deep learning models for precise classification

Revolutionizing air quality forecasting: Fusion of state-of-the-art deep learning models for precise classification

A novel ViT-BILSTM model for physical activity intensity classification in adults using gravity-based acceleration

AimThe aim of this study is to apply a novel hybrid framework incorporating a Vision Transformer (ViT) and bidirectional long short-term memory (Bi-LSTM) model for classifying physical activity intensity (PAI) in adults using gravity-based acceleration. Additionally, it further investigates how PAI and temporal window (TW) impacts the model’ s accuracy.MethodThis research used the Capture-24 dataset, consisting of raw accelerometer data from 151 participants aged 18 to 91. Gravity-based acceleration was utilised to generate images encoding various PAIs. These images were subsequently analysed using the ViT-BiLSTM model, with results presented in confusion matrices and compared with baseline models. The model’s robustness was evaluated through temporal stability testing and examination of accuracy and loss curves.ResultThe ViT-BiLSTM model excelled in PAI classification task, achieving an overall accuracy of 98.5% ± 1.48% across five TWs-98.7% for 1s, 98.1% for 5s, 98.2% for 10s, 99% for 15s, and 98.65% for 30s of TW. The model consistently exhibited superior accuracy in predicting sedentary (98.9% ± 1%) compared to light physical activity (98.2% ± 2%) and moderate-to-vigorous physical activity (98.2% ± 3%). ANOVA showed no significant accuracy variation across PAIs (F = 2.18, p = 0.13) and TW (F = 0.52, p = 0.72). Accuracy and loss curves show the model consistently improves its performance across epochs, demonstrating its excellent robustness.ConclusionThis study demonstrates the ViT-BiLSTM model’s efficacy in classifying PAI using gravity-based acceleration, with performance remaining consistent across diverse TWs and intensities. However, PAI and TW could result in slight variations in the model’s performance. Future research should concern and investigate the impact of gravity-based acceleration on PAI thresholds, which may influence model’s robustness and reliability.

EEG microstate analysis and machine learning classification in patients with obsessive-compulsive disorder.

Microstate characterization of electroencephalogram (EEG) is a data-driven approach to explore the functional changes and interrelationships of multiple brain networks on a millisecond scale. This study aimed to explore the pathological changes of whole-brain functional networks in patients with obsessive-compulsive disorders (OCD) through microstate analysis and further to explore its potential value as an auxiliary diagnostic index. Forty-eight OCD patients (33 with more than moderate anxiety symptoms, 15 with mild anxiety symptoms) and 52 healthy controls (HCs) were recruited. Brain activities during eyes-closed period were collected using 64-channel electroencephalography. The differences in microstate features between OCD patients and HCs were compared, and the relationship between the microstate features and clinical symptoms were explored. Key microstate features were selected for machine learning modeling to achieve targeted classifications. The probability of transition from microstate B to C was significantly lower in OCD patients compared to HCs, and the obsessive thoughts factor scores were significantly correlated with the duration of microstate A, the occurrence of microstate B, and the transition probability from microstate C to B. The occurrence rate of microstate C was significantly negatively correlated with the Hamilton rating scale for anxiety (HAMA) scores. The AUC (Area Under the Receiver Operating Characteristic Curve) of the machine learning model in the test set classification between the above two groups and between OCD patients with more than moderate/mild anxiety symptoms could achieve 70.43% and 77.13%, respectively. EEG microstate characteristics were altered in OCD patients, and these changes were closely associated with obsessive thoughts and anxiety symptoms. Besides, the machine learning classification model based on microstate features has limited ability to identify OCD, and further optimization on this classification approach is still needed in the future.

Lightweight deep learning models for aerial scene classification: A comprehensive survey

Lightweight deep learning models for aerial scene classification: A comprehensive survey

A comparative study on rapid qualitative and quantitative determination of olive oil adulteration.

Authenticity of olive oil is a significant concern for producers, consumers, and policymakers. To help address this issue, a rapid, efficient, and accurate flow injection mass spectrometric (FIMS) fingerprinting approach, combined with SVM and PLS classification and regression models, was proposed for the identification and quantitative analysis of olive oil adulteration. Based on the comprehensive comparative analysis, SVM outperformed those of PLS-DA, achieving higher values for accuracy, sensitivity, and specificity, as well as positive predictive and negative predictive values, in identifying adulterated olive oil samples. Furthermore, compared with PLSR model, the SVR model demonstrated superior performance in determining the content of adulterated olive oil, with a higher coefficient of determination and lower Root Mean Square Error. In conclusion, FIMS fingerprinting technology in combination with SVM can be effectively implemented for rapid, reliable, and accurate identification and quantification of olive oil adulteration.

Comprehensive evaluation of the toxicological effects of commonly encountered synthetic cathinones using in silico methods.

Synthetic cathinones (SCs), a group of new psychoactive substances (NPS), are designer molecules with hallucinogenic and psychostimulatory effects. Although the structural similarities of SCs to amphetamines suggest that they may have similar toxicity profiles to those of amphetamine congeners, little is known about SCs from a toxicological point of view. In the present study, the toxicity profiles of commonly encountered SCs (n = 65), listed in the 2020 Report of the United Nations Office on Drugs and Crime (UNODC), were evaluated using in silico methods. We aimed to gain a deeper understanding of key toxicological endpoints: acute oral toxicity (LD50), mutagenicity, genotoxicity, and carcinogenicity prediction using EPA TEST (v.5.1.2 and 4.2.1), VEGA (v.1.2.3), and ProTox (v.3.0). Physicochemical and pharmacokinetic (ADME) properties were estimated using SwissADME and pkCSM. 2,3-MDMC (41) was predicted to be the most lethal SC by the VEGA KNN and the EPA TEST v.5.1.2 with an oral rat LD50 value of 105.17 and 117.77mg/kg, respectively. 4-BEC (2) was the only molecule with a consensus score of positive prediction greater than 0.90 in both TEST mutagenicity models. 2,3-MDMC (41) and methylone (52) were predicted as carcinogenic by VEGA carcinogenicity CAESAR, ISS, IRFMN-ISSCAN-CGX, oral classification, and ProTox models. These two SCs were predicted to be active by VEGA chromosomal aberration (CORAL) and in vitro micronuclei-inducing activity (IRFMN-VERMEER) models. Our results concluded that given the prolonged exposure duration and age range, the genotoxic and carcinogenic potential of SCs should be considered, among other known toxic effects.

PREDICTION BASED ON MACHINE LEARNING OF TOOTH SENSITIVITY FOR IN-OFFICE DENTAL BLEACHING

To develop a supervised machine learning model to predict the occurrence and intensity of tooth sensitivity (TS) in patients undergoing in-office dental bleaching testing various algorithm models. Retrospective data from 458 patients were analyzed, including variables such as the occurrence and intensity of TS, basal tooth color, bleaching material characteristics (concentration and pH), intervention details (number and duration of applications), and patient age. Classification and regression models were evaluated using 5-fold cross-validation and assessed based on various performance parameters. For the predictive classification task (occurrence of TS), the developed models achieved a maximum area under the receiver operating characteristic curve (AUC) of 0.76 [0.62-0.88] on the test data, with an F1-score of 0.80 [0.71-0.87]. In cross-validation, the highest AUC reached 0.86 [0.84-0.88], and the highest F1-score was 0.78 [0.75-0.83]. For predicting TS intensity, the regression models demonstrated a minimum mean absolute error (MAE) of 1.76 [1.45-2.06] and a root mean square error (RMSE) of 2.38 [2.06-2.69] on the test set. During cross-validation, the lowest MAE was 1.84 [1.67-2.03], with an RMSE of 2.39 [2.20-2.58]. The supervised machine learning model for estimating the occurrence and intensity of TS in patients undergoing in-office bleaching demonstrated good predictive power. The Gradient Boosting Classifier and Support Vector Machine Regressor algorithms stood out as having the greatest predictive power among those tested. These models can serve as valuable tools for anticipating tooth sensitivity in this patient population, facilitating better post-treatment management and control.

Deep learning for automated hip fracture detection and classification : achieving superior accuracy.

The aim of this study was to develop and evaluate a deep learning-based model for classification of hip fractures to enhance diagnostic accuracy. A retrospective study used 5,168 hip anteroposterior radiographs, with 4,493 radiographs from two institutes (internal dataset) for training and 675 radiographs from another institute for validation. A convolutional neural network (CNN)-based classification model was trained on four types of hip fractures (Displaced, Valgus-impacted, Stable, and Unstable), using DAMO-YOLO for data processing and augmentation. The model's accuracy, sensitivity, specificity, Intersection over Union (IoU), and Dice coefficient were evaluated. Orthopaedic surgeons' diagnoses served as the reference standard, with comparisons made before and after artificial intelligence assistance. The accuracy, sensitivity, specificity, IoU, and Dice coefficients of the model for the four fracture categories in the internal dataset were as follows: Displaced (1.0, 0.79, 1.0, 0.70, 0.82), Valgus-impacted (1.0, 0.80, 1.0, 0.70, 0.82), Stable (0.99, 0.95, 0.99, 0.83, 0.89), and Unstable (1.0, 0.98, 0.99, 0.86, 0.92), respectively. For the external validation dataset, the sensitivity and specificity were as follows: Displaced (0.83, 0.94), Valgus-impacted (0.89, 0.90), Stable (0.88, 0.95), and Unstable (0.85, 0.99), respectively. The overall means (Micro AVG and Macro AVG) for the external dataset were Micro AVG (0.83 (SD 0.05), 0.96 (SD 0.01)) and Macro AVG (0.69 (SD 0.02), 0.95 (SD 0.02)), respectively. Compared to human diagnosis alone, our study demonstrates that the developed model significantly improves the accuracy of detecting and classifying hip fractures. Our model has shown great potential in assisting clinicians with the accurate diagnosis and classification of hip fractures.

Comprehensive two dimensional gas chromatography – Time of flight mass spectrometry (GC×GC-TOFMS) for the investigation of botanical origin of raw spirits

Comprehensive two dimensional gas chromatography – time of flight mass spectrometry (GC × GC-TOFMS) with sample introduction using headspace solid phase microextraction (HS-SPME) was used for the botanical classification of raw spirits obtained from C3 (corn and sorghum) and C4 (rye, wheat and potato) plants. 45 spirit samples representing these raw materials (10 spirits produced from rye, corn, wheat and potato, and 5 from sorghum) were analyzed. Volatile compounds profiles were compared by PCA, and after removal of outliers samples were subjected to the classification model. OPLS-DA model was built (R2Y = 0.924 Q2Y = 0.895) that enabled clear separation of all tested spirits of different botanical origin. The model was validated with training and testing sets and 100 % correct assignment was achieved. GC × GC proved to be a method that not only can be used as a tool for botanic origin of raw spirits, but also provides detailed information of volatile fermentation by-products, characteristic for particular spirit.

Smartphone pupillometry with machine learning differentiates ischemic from hemorrhagic stroke: A pilot study.

Similarities between acute ischemic and hemorrhagic stroke make diagnosis and triage challenging. We studied a smartphone-based quantitative pupillometer for differentiation of acute ischemic and hemorrhagic stroke. Stroke patients were recruited prior to surgical or interventional treatment. Smartphone pupillometry was used to quantify components of the pupillary light reflex (PLR). A synthetic minority oversampling technique (SMOTE) was applied to correct sample size imbalance. Four binary classification model types were trained using all possible combinations of the PLR components with 10-fold cross validation stratified by cohort. Models were evaluated for accuracy, sensitivity, specificity, area under the curve (AUC), and F1 score. The three best-performing models were selected based on AUC. Shapley additive explanation plots were produced to explain PLR parameter impacts on model predictions. Eleven subjects with intraparenchymal hemorrhage and 22 subjects with acute ischemic stroke were enrolled. One way ANOVA demonstrated significant differences between healthy control data, AIS, and IPH in five out of seven PLR parameters. After SMOTE, each class had n=22 PLR recordings for model training. The best-performing model was random forest using a combination of latency, mean and maximum constriction velocity, and mean dilation velocity to discriminate between stroke types with 91.5% (95% confidence interval: 84.1-98.9) accuracy, 90% (82.9-97.1) sensitivity, 93.3% (83-100) specificity, 0.917 (0.847-0.987) AUC, and 90.7% (84.1-97.3) F1 score. Smartphone-based quantitative pupillometry could be useful in differentiating between acute ischemic and hemorrhagic stroke.

Deep learning enabled rapid classification of yeast species in food by imaging of yeast microcolonies.

Diverse species of yeasts are commonly associated with food and food production environments. The contamination of food products by spoilage yeasts poses significant challenges, leading to quality degradation and food loss. Similarly, the introduction of undesirable strains during fermentation can cause considerable challenges with the quality and progress of the fermentation process. Conventional detection methods require the isolation of visible yeast colonies for genetic or biochemical characterization, which takes 5-7days and demands significant labor. This study presents a deep learning-based yeast classification approach that combines conventional cultivation methods, white light optical microscopy of microcolony, and deep learning techniques for rapidly detecting and classifying yeasts. Utilizing deep convolutional neural networks, the model accurately discriminates 7 different yeasts within 6h, achieving a mPrecision of 96.0% and a mRecall of 96.3%. Synthetic image dataset generated by generative adversarial networks (GAN) model further improved the model performance for Debaryomyces hansenii and Wickerhamomyces anomalus, yeast species with lower initial classification performance. With the addition of synthetic images in the training process, Precision for W. anomalus and Recall for D. hansenii increased by 7.7% and 5.6%, respectively. The yeast classification model was validated in the presence of microscopic food debris using tomato and tomato juice as representative examples of fresh produce and processed juice. The model maintained high classification accuracy in the presence of food debris (mPrecision and mRecall >93.9%). Overall, this methodology significantly accelerates the detection and classification of yeast species using conventional cultivation and simple white light microscopy in combination with deep learning. The simplicity, including low cost of the experimental approaches and the robustness of the deep learning model make it a highly applicable approach for routine yeast monitoring and yeast spoilage control in the food industry.