Machine learning models for classifying non-specific neck pain using craniocervical posture and movement.

OBJECTIVE: Physical therapists and clinicians commonly confirm craniocervical posture (CCP), cervical retraction, and craniocervical flexion as screening tests because they contribute to non-specific neck pain (NSNP). We compared the predictive performance of statistical machine learning (ML) models for classifying individuals with and without NSNP using datasets containing CCP and cervical kinematics during pro- and retraction (CKdPR). DESIGN: Exploratory, cross-sectional design. SETTING AND PARTICIPANTS: In total, 773 public service office workers (PSOWs) were screened for eligibility (NSNP, 441; without NSNP, 332). METHODS: We set up five datasets (CCP, cervical kinematics during the protraction, cervical kinematics during the retraction, CKdPR and combination of the CCP and CKdPR). Four ML algorithms-random forest, logistic regression, Extreme Gradient boosting, and support vector machine-were trained. MAIN OUTCOME MEASURES: Model performance were assessed using area under the curve (AUC), accuracy, precision, recall and F1-score. To interpret the predictions, we used Feature permutation importance and SHapley Additive explanation values. RESULTS: The random forest model in the CKdPR dataset classified PSOWs with and without NSNP and achieved the best AUC among the five datasets using the test data (AUC, 0.892 [good]; F1, 0.832). The random forest model in the CCP dataset had the worst AUC among the five datasets using the test data [AUC, 0.738 (fair); F1, 0.715]. CONCLUSION: ML performance was higher for the CKdPR dataset than for the CCP dataset, suggesting that ML algorithms are more suitable than classical statistical methods for developing robust models for classifying PSOWs with and without NSNP.

Atomistic Probing of Defect-Engineered 2H-MoTe2 Monolayers.

Point defects dictate various physical, chemical, and optoelectronic properties of two-dimensional (2D) materials, and therefore, a rudimentary understanding of the formation and spatial distribution of point defects is a key to advancement in 2D material-based nanotechnology. In this work, we performed the demonstration to directly probe the point defects in 2H-MoTe2 monolayers that are tactically exposed to (i) 200 °C-vacuum-annealing and (ii) 532 nm-laser-illumination; and accordingly, we utilize a deep learning algorithm to classify and quantify the generated point defects. We discovered that tellurium-related defects are mainly generated in both 2H-MoTe2 samples; but interestingly, 200 °C-vacuum-annealing and 532 nm-laser-illumination modulate a strong n-type and strong p-type 2H-MoTe2, respectively. While 200 °C-vacuum-annealing generates tellurium vacancies or tellurium adatoms, 532 nm-laser-illumination prompts oxygen atoms to be adsorbed/chemisorbed at tellurium vacancies, giving rise to the p-type characteristic. This work significantly advances the current understanding of point defect engineering in 2H-MoTe2 monolayers and other 2D materials, which is critical for developing nanoscale devices with desired functionality.

Cellular subpopulations identified using an ensemble average of multiple dielectrophoresis measurements.

While the average value measurement approach can successfully analyze and predict the general behavior and biophysical properties of an isogenic cell population, it fails when significant differences among individual cells are generated in the population by intracellular changes such as the cell cycle, or different cellular responses to certain stimuli. Detecting such single-cell differences in a cell population has remained elusive. Here, we describe an easy-to-implement and generalizable platform that measures the dielectrophoretic cross-over frequency of individual cells by decreasing measurement noise with a stochastic method and computing ensemble average statistics. This platform enables multiple, real-time, label-free detection of individual cells with significant dielectric variations over time within an isogenic cell population. Using a stochastic method in combination with the platform, we distinguished cell subpopulations from a mixture of drug-untreated and -treated isogenic cells. Furthermore, we demonstrate that our platform can identify drug-treated isogenic cells with different recovery rates.

Full automation of point defect detection in transition metal dichalcogenides through a dual mode deep learning algorithm.

Point defects often appear in two-dimensional (2D) materials and are mostly correlated with physical phenomena. The direct visualisation of point defects, followed by statistical inspection, is the most promising way to harness structure-modulated 2D materials. Here, we introduce a deep learning-based platform to identify the point defects in 2H-MoTe2: synergy of unit cell detection and defect classification. These processes demonstrate that segmenting the detected hexagonal cell into two unit cells elaborately cropped the unit cells: further separating a unit cell input into the Te2/Mo column part remarkably increased the defect classification accuracies. The concentrations of identified point defects were 7.16 × 1020 cm2 of Te monovacancies, 4.38 × 1019 cm2 of Te divacancies and 1.46 × 1019 cm2 of Mo monovacancies generated during an exfoliation process for TEM sample-preparation. These revealed defects correspond to the n-type character mainly originating from Te monovacancies, statistically. Our deep learning-oriented platform combined with atomic structural imaging provides the most intuitive and precise way to analyse point defects and, consequently, insight into the defect-property correlation based on deep learning in 2D materials.

Deep learning framework for automated goblet cell density analysis in in-vivo rabbit conjunctiva.

Goblet cells (GCs) in the conjunctiva are specialized epithelial cells secreting mucins for the mucus layer of protective tear film and playing immune tolerance functions for ocular surface health. Because GC loss is observed in various ocular surface diseases, GC examination is important for precision diagnosis. Moxifloxacin-based fluorescence microscopy (MBFM) was recently developed for non-invasive high-contrast GC visualization. MBFM showed promise for GC examination by high-speed large-area imaging and a robust analysis method is needed to provide GC information. In this study, we developed a deep learning framework for GC image analysis, named dual-channel attention U-Net (DCAU-Net). Dual-channel convolution was used both to extract the overall image texture and to acquire the GC morphological characteristics. A global channel attention module was adopted by combining attention algorithms and channel-wise pooling. DCAU-Net showed 93.1% GC segmentation accuracy and 94.3% GC density estimation accuracy. Further application to both normal and ocular surface damage rabbit models revealed the spatial variations of both GC density and size in normal rabbits and the decreases of both GC density and size in damage rabbit models during recovery after acute damage. The GC analysis results were consistent with histology. Together with the non-invasive high-contrast imaging method, DCAU-Net would provide GC information for the diagnosis of ocular surface diseases.

Ventricle tracking in transesophageal echocardiography (TEE) images during cardiopulmonary resuscitation (CPR) using deep learning and monogenic filtering.

High-quality cardiopulmonary resuscitation (CPR) is the most important factor in promoting resuscitation outcomes; therefore, monitoring the quality of CPR is strongly recommended in current CPR guidelines. Recently, transesophageal echocardiography (TEE) has been proposed as a potential real-time feedback modality because physicians can obtain clear echocardiographic images without interfering with CPR. The quality of CPR would be optimized if the myocardial ejection fraction (EF) could be calculated in real-time during CPR. We conducted a study to derive a protocol to detect systole and diastole automatically and calculate EF using TEE images acquired from patients with cardiac arrest. The data were supplemented using thin-plate spline transformation to solve the problem of insufficient data. The deep learning model was constructed based on ResUNet + + , and a monogenic filtering method was applied to clarify the ventricular boundary. The performance of the model to which the monogenic filter was added and the existing model was compared. The left ventricle was segmented in the ME LAX view, and the left and right ventricles were segmented in the ME four-chamber view. In most of the results, the performance of the model to which the monogenic filter was added was high, and the difference was very small in some cases; but the performance of the existing model was high. Through this learned model, the effect of CPR can be quantitatively analyzed by segmenting the ventricle and quantitatively analyzing the degree of contraction of the ventricle during systole and diastole. Supplementary Information: The online version contains supplementary material available at 10.1007/s13534-023-00293-9.

Patterns of Familial and Racial Trauma and Their Associations with Substance Use Disorders among Racial/Ethnic Minority Adults.

BACKGROUND: Racial/ethnic minorities in the United States often experience many different types of traumatic events. We examine the patterns of familial and racial trauma and their associations with substance use disorders (SUDs) among racial/ethnic minority adults. METHODS: We used data from the National Epidemiologic Survey of Alcohol and Related Conditions-III. The study sample included 17,115 individuals who were Hispanic (43.6%), Black (34.9%), Asian American and Pacific Islander (17.0%), and American Indian or Alaska Native (AI/AN, 4.6%). Latent class analysis models with covariates and distal outcomes were analyzed to investigate patterns of trauma exposure and estimate binary outcomes of SUDs. Familial and racial trauma was measured by ten areas of adverse childhood experiences (ACEs) and six items of racial discrimination. RESULTS: We found four distinctive groups: low trauma (Class 1, 62.1%), high discrimination (Class 2, 17.2%), high ACEs (Class 2, 14.9%), and high trauma (Class 4, 5.9%). Compared to Class 1, other groups were more likely to include Black and AI/AN adults. Participants in Class 2 reported greater risks for alcohol and other drug use disorders. Those in Class 3 and 4 reported greater risks for alcohol, opioid, stimulant, and other drug use disorders. CONCLUSION: Given a higher risk of trauma exposure in Black and AI/AN adults, racially and ethnically sensitive trauma-focused interventions may help prevent and reduce SUDs in those populations.

Deep Learning-Based Evaluation of Ultrasound Images for Benign Skin Tumors.

In this study, a combined convolutional neural network for the diagnosis of three benign skin tumors was designed, and its effectiveness was verified through quantitative and statistical analysis. To this end, 698 sonographic images were taken and diagnosed at the Department of Dermatology at Severance Hospital in Seoul, Korea, between 10 November 2017 and 17 January 2020. Through an empirical process, a convolutional neural network combining two structures, which consist of a residual structure and an attention-gated structure, was designed. Five-fold cross-validation was applied, and the train set for each fold was augmented by the Fast AutoAugment technique. As a result of training, for three benign skin tumors, an average accuracy of 95.87%, an average sensitivity of 90.10%, and an average specificity of 96.23% were derived. Also, through statistical analysis using a class activation map and physicians' findings, it was found that the judgment criteria of physicians and the trained combined convolutional neural network were similar. This study suggests that the model designed and trained in this study can be a diagnostic aid to assist physicians and enable more efficient and accurate diagnoses.

A nystagmus extraction system using artificial intelligence for video-nystagmography.

Benign paroxysmal positional vertigo (BPPV), the most common vestibular disorder, is diagnosed by an examiner changing the posture of the examinee and inducing nystagmus. Among the diagnostic methods used to observe nystagmus, video-nystagmography has been widely used recently because it is non-invasive. A specialist with professional knowledge and training in vertigo diagnosis is needed to diagnose BPPV accurately, but the ratio of vertigo patients to specialists is too high, thus necessitating the need for automated diagnosis of BPPV. In this paper, a convolutional neural network-based nystagmus extraction system, ANyEye, optimized for video-nystagmography data is proposed. A pupil was segmented to track the exact pupil trajectory from real-world data obtained during field inspection. A deep convolutional neural network model was trained with the new video-nystagmography dataset for the pupil segmentation task, and a compensation algorithm was designed to correct pupil position. In addition, a slippage detection algorithm based on moving averages was designed to eliminate the motion artifacts induced by goggle slippage. ANyEye outperformed other eye-tracking methods including learning and non-learning-based algorithms with five-pixel error detection rate of 91.26%.

CoAt-Mixer: Self-attention deep learning framework for left ventricular hypertrophy using electrocardiography.

Left ventricular hypertrophy is a significant independent risk factor for all-cause mortality and morbidity, and an accurate diagnosis at an early stage of heart change is clinically significant. Electrocardiography is the most convenient, economical, and non-invasive method for screening in primary care. However, the coincidence rate of the actual left ventricular hypertrophy and diagnostic findings was low, consequently increasing the interest in algorithms using big data and deep learning. We attempted to diagnose left ventricular hypertrophy using big data and deep learning algorithms, and aimed to confirm its diagnostic power according to the differences between males and females. This retrospective study used electrocardiographs obtained at Yonsei University Wonju Severance Christian Hospital, Wonju, Korea, from October 2010 to February 2020. Binary classification was performed for primary screening for left ventricular hypertrophy. Three datasets were used for the experiment: the male, female, and entire dataset. A cutoff for binary classification was defined as the meaningful as a screening test (<132 g/m2 vs. ≥132 g/m2, <109 g/m2 vs. ≥109 g/m2). Six types of input were used for the classification tasks. We attempted to determine whether electrocardiography had predictive power for left ventricular hypertrophy diagnosis. For the entire dataset, the model achieved an area under the receiver operating characteristic (AUROC) curve of 0.836 (95% CI, 0.833-838) with a sensitivity of 78.37% (95% CI, 76.79-79.95). For the male dataset, the AUROC was 0.826 (95% CI, 0.822-830) with a sensitivity of 76.73% (95% CI, 75.14-78.33). For the female dataset, the AUROC was 0.772 (95% CI, 0.769-775) with a sensitivity of 72.90% (95% CI, 70.33-75.46). Our model confirmed that left ventricular hypertrophy can be classified to some extent using electrocardiography, demographics, and electrocardiography features. In particular, a learning environment that considered gender differences was constructed. Consequently, the difference in diagnostic power between men and women was confirmed. Our model will help patients with suspected left ventricular hypertrophy to undergo screening tests at a low cost. In addition, our research and attempts will show the expected effect that gender-consideration approaches can help with various currently proposed diagnostic methods.

Deep Learning Algorithms for Estimation of Demographic and Anthropometric Features from Electrocardiograms.

The electrocardiogram (ECG) has been known to be affected by demographic and anthropometric factors. This study aimed to develop deep learning models to predict the subject's age, sex, ABO blood type, and body mass index (BMI) based on ECGs. This retrospective study included individuals aged 18 years or older who visited a tertiary referral center with ECGs acquired from October 2010 to February 2020. Using convolutional neural networks (CNNs) with three convolutional layers, five kernel sizes, and two pooling sizes, we developed both classification and regression models. We verified a classification model to be applicable for age (<40 years vs. ≥40 years), sex (male vs. female), BMI (<25 kg/m2 vs. ≥25 kg/m2), and ABO blood type. A regression model was also developed and validated for age and BMI estimation. A total of 124,415 ECGs (1 ECG per subject) were included. The dataset was constructed by dividing the entire set of ECGs at a ratio of 4:3:3. In the classification task, the area under the receiver operating characteristic (AUROC), which represents a quantitative indicator of the judgment threshold, was used as the primary outcome. The mean absolute error (MAE), which represents the difference between the observed and estimated values, was used in the regression task. For age estimation, the CNN achieved an AUROC of 0.923 with an accuracy of 82.97%, and a MAE of 8.410. For sex estimation, the AUROC was 0.947 with an accuracy of 86.82%. For BMI estimation, the AUROC was 0.765 with an accuracy of 69.89%, and a MAE of 2.332. For ABO blood type estimation, the CNN showed an inferior performance, with a top-1 accuracy of 31.98%. For the ABO blood type estimation, the CNN showed an inferior performance, with a top-1 accuracy of 31.98% (95% CI, 31.98-31.98%). Our model could be adapted to estimate individuals' demographic and anthropometric features from their ECGs; this would enable the development of physiologic biomarkers that can better reflect their health status than chronological age.

Left ventricle segmentation in transesophageal echocardiography images using a deep neural network.

PURPOSE: There has been little progress in research on the best anatomical position for effective chest compressions and cardiac function during cardiopulmonary resuscitation (CPR). This study aimed to divide the left ventricle (LV) into segments to determine the best position for effective chest compressions using the LV systolic function seen during CPR. METHODS: We used transesophageal echocardiography images acquired during CPR. A deep neural network with an attention mechanism and a residual feature aggregation module were applied to the images to segment the LV. The results were compared between the proposed model and U-Net. RESULTS: The results of the proposed model showed higher performance in most metrics when compared to U-Net: dice coefficient (0.899±0.017 vs. 0.792±0.027, p<0.05); intersection of union (0.822±0.026 vs. 0.668±0.034, p<0.05); recall (0.904±0.023 vs. 0.757±0.037, p<0.05); precision (0.901±0.021 vs. 0.859±0.034, p>0.05). There was a significant difference between the proposed model and U-Net. CONCLUSION: Compared to U-Net, the proposed model showed better performance for all metrics. This model would allow us to evaluate the systolic function of the heart during CPR in greater detail by segmenting the LV more accurately.

Routes Out of Child Abuse to School Adjustments: A Comparison Between North Korean Adolescent Refugees and Native South Korean Adolescents.

This study aimed to examine pathways from child abuse to school adjustment and the roles of self-control and academic stress on the link among North Korean adolescent refugees living in South Korea and native South Korean adolescents. A total of 610 students (adolescents from South Korea = 325 and adolescents from North Korea = 285) living in South Korea, from juniors in middle schools to seniors in high schools, were interviewed in 2017. Multigroup structural equation modeling was used to examine differences in the country of origin on the pathways from abuse to school adjustment via self-control and academic stress. North Korean adolescent refugees were less likely to adjust to their school life than South Korean adolescents. Academic stress was found as a significant mediator between self-control and school adjustment in both South Korean and North Korean adolescents. Child abuse was associated with self-control of South Korean adolescents. Childhood abuse from parents can have an overall influence on individual characteristics and school life for adolescents. By paying attention to this process, comprehensive solutions are urgently required not only to intervene in the problem of abusive parenting behaviors but also to block the path of the expanding negative consequences among both groups of adolescents.

DEEP LEARNING-BASED PREDICTION OF OUTCOMES FOLLOWING NONCOMPLICATED EPIRETINAL MEMBRANE SURGERY.

PURPOSE: We used deep learning to predict the final central foveal thickness (CFT), changes in CFT, final best corrected visual acuity, and best corrected visual acuity changes following noncomplicated idiopathic epiretinal membrane surgery. METHODS: Data of patients who underwent noncomplicated epiretinal membrane surgery at Severance Hospital from January 1, 2010, to December 31, 2018, were reviewed. Patient age, sex, hypertension and diabetes statuses, and preoperative optical coherence tomography scans were noted. For image analysis and model development, a pre-trained VGG16 was adopted. The mean absolute error and coefficient of determination (R 2 ) were used to evaluate the model performances. The study involved 688 eyes of 657 patients. RESULTS: For final CFT, the mean absolute error was the lowest in the model that considered only clinical and demographic characteristics; the highest accuracy was achieved by the model that considered all clinical and surgical information. For CFT changes, models utilizing clinical and surgical information showed the best performance. However, our best model failed to predict the final best corrected visual acuity and best corrected visual acuity changes. CONCLUSION: A deep learning model predicted the final CFT and CFT changes in patients 1 year after epiretinal membrane surgery. Central foveal thickness prediction showed the best results when demographic factors, comorbid diseases, and surgical techniques were considered.

Deep-Learning-Based Automatic Detection and Segmentation of Brain Metastases with Small Volume for Stereotactic Ablative Radiotherapy.

Recently, several efforts have been made to develop the deep learning (DL) algorithms for automatic detection and segmentation of brain metastases (BM). In this study, we developed an advanced DL model to BM detection and segmentation, especially for small-volume BM. From the institutional cancer registry, contrast-enhanced magnetic resonance images of 65 patients and 603 BM were collected to train and evaluate our DL model. Of the 65 patients, 12 patients with 58 BM were assigned to test-set for performance evaluation. Ground-truth for BM was assigned to one radiation oncologist to manually delineate BM and another one to cross-check. Unlike other previous studies, our study dealt with relatively small BM, so the area occupied by the BM in the high-resolution images were small. Our study applied training techniques such as the overlapping patch technique and 2.5-dimensional (2.5D) training to the well-known U-Net architecture to learn better in smaller BM. As a DL architecture, 2D U-Net was utilized by 2.5D training. For better efficacy and accuracy of a two-dimensional U-Net, we applied effective preprocessing include 2.5D overlapping patch technique. The sensitivity and average false positive rate were measured as detection performance, and their values were 97% and 1.25 per patient, respectively. The dice coefficient with dilation and 95% Hausdorff distance were measured as segmentation performance, and their values were 75% and 2.057 mm, respectively. Our DL model can detect and segment BM with small volume with good performance. Our model provides considerable benefit for clinicians with automatic detection and segmentation of BM for stereotactic ablative radiotherapy.

Evaluation of Atopic Dermatitis Improvement Caused by Low-Level, Low-Frequency Pulsed Electromagnetic Fields.

This study aimed to evaluate the effectiveness of using low-level, low-frequency pulsed electromagnetic field (LLLF_PEMF) stimulation to improve atopic dermatitis induced by 2,4-dinitrochlorobenzene (DNCB). Twenty 6-week-old hairless mice were randomly divided into Normal (n = 5), PEMF 15 Hz (n = 5), PEMF 75 Hz (n = 5), and Sham (n = 5) groups. Following the onset of atopic dermatitis symptoms, PEMF groups (15 and 75 Hz) were stimulated with LLLF_PEMF (15 mT) for 8 h per day for 1 week. Sensory evaluation analysis revealed a significant difference between the PEMF 15 Hz group and Sham group (P < 0.05), but these differences were not visually obvious. While both the PEMF and Sham groups had atopic dermatitis lesions, lesion size was significantly smaller in the two PEMF groups than in the Sham group (P < 0.001). Additionally, changes in epithelial thickness because of skin inflammation significantly decreased for both PEMF groups, compared with the Sham group (P < 0.001). In conclusion, these results suggest that PEMF stimulation in vivo triggers electro-chemical reactions that affect immune response. © 2022 Bioelectromagnetics Society.

Comparative effects of passive and active mode robot-assisted gait training on brain and muscular activities in sub-acute and chronic stroke.

BACKGROUND: Robot-assisted gait training (RAGT) was initially developed based on the passive controlled (PC) mode, where the target or ideal locomotor kinematic trajectory is predefined and a patient basically 'rides' the robot instead of actively participating in the actual locomotor relearning process. A new insightful contemporary neuroscience and mechatronic evidence suggest that robotic-based locomotor relearning can be best achieved through active interactive (AI) mode rather than PC mode. OBJECTIVE: The purpose of this study was to compare the pattern of gait-related cortical activity, specifically gait event-related spectral perturbations (ERSPs), and muscle activity from the tibialis anterior (TA) and clinical functional tests in subacute and chronic stroke patients during robot-assisted gait training (RAGT) in passive controlled (PC) and active interactive (AI) modes. METHODS: The present study involves a two-group pretest-posttest design in which two groups (i.e., PC-RAGT group and AI-RAGT group) of 14 stroke subjects were measured to assess changes in ERSPs, the muscle activation of TA, and the clinical functional tests, following 15- 18 sessions of intervention according to the protocol of each group. RESULTS: Our preliminary results demonstrated that the power in the µ band (8- 12 Hz) was increased in the leg area of sensorimotor cortex (SMC) and supplementary motor area (SMA) at post-intervention as compared to pre-intervention in both groups. Such cortical neuroplasticity change was associated with TA muscle activity during gait and functional independence in functional ambulation category (FAC) and motor coordination in Fugl- Meyer Assessment for lower extremity (FMA-LE) test as well as spasticity in the modified Ashworth scale (MAS) measures. CONCLUSIONS: We have first developed a novel neuroimaging experimental paradigm which distinguished gait event related cortical involvement between pre- and post-intervention with PC-RAGT and AI-RAGT in individuals with subacute and chronic hemiparetic stroke.