RESUMEN

OBJECTIVE: This study provides an adaptive threshold algorithm for burst detection in electroencephalograms (EEG) of preterm infantes and evaluates its performance using clinical real-world EEG data. Approach: We developed an adaptive threshold algorithm for burst detection in EEG recordings from preterm infants. To assess its applicability in the real-world, we tested the algorithm on a dataset of 30 clinical EEG recordings which were not preselected for good quality, to ensure a real-world scenario. Main results: Interrater agreement was substantial at a kappa of 0.73 (0.68 - 0.79 inter-quantile range). The performance of the algorithm showed a similar agreement with one clinical expert of 0.73 (0.67 - 0.76) and a sensitivity and specificity of 0.90 (0.82 - 0.94) and 0.95 (0.93 - 0.97), respectively. Significance: The adaptive threshold algorithm demonstrated robust performance in detecting burst patterns in clinical EEG data from preterm infants, highlighting its practical utility. The fine-tuned algorithm achieved similar performance to human raters. The algorithm proves to be a valuable tool for automated burst detection in the EEG of preterm infants.

RESUMEN

INTRODUCTION: Recent observations suggest that circadian rhythms are implicated in the timing of stroke onset and the speed of infarct progression. We aimed to replicate these observations in a large, multi-center, automated imaging database. METHODS: The RAPID Insights database was queried from 02/01/2016 to 01/31/2022 for patients with perfusion imaging and automated detection of an ischemic stroke due to a presumed large vessel occlusion. Exclusion criteria included: patient age ≤25, mismatch volume of <0 cc, and failure to register a positive value on either relative cerebral blood flow (rCBF) reduction of 38% less than normal or total mismatch volume. Imaging time was subdivided into three epochs: Night: 23:00h-06:59h and Day: 07:00h-14:59h, and Evening: 15:00h-22:59h. Perfusion parameters were defined using standard conventions for core volume, penumbra, and collateral circulation (measured via the Hypoperfusion Intensity Ratio, HIR). Statistical significance was tested using a sinusoidal regression analysis. RESULTS: A total of 18,137 cases were analyzed. The peak incidence of stroke imaging of patients with LVOs occurred around noon. A sinusoidal pattern was present, with larger ischemic core volumes and higher HIR during the night compared to the day: peak ischemic core volume of 23.4 cc occurred with imaging performed at 3:56 AM (p<0.001) and peak HIR of 0.35 at 3:40 AM (p<0.001). CONCLUSION: We found that ischemic core volumes were larger and collateral status worse at nighttime compared to daytime in this large national database. These findings support prior data suggesting that poor collateral recruitment with subsequent larger ischemic stroke volumes may occur at night.

RESUMEN

BACKGROUND: Neonatal seizures are diagnostically challenging and predominantly electrographic-only. Multichannel video continuous electroencephalography (cEEG) is the gold standard investigation, however, out-of-hours access to neurophysiology support can be limited. Automated seizure detection algorithms (SDAs) are designed to detect changes in EEG data, translated into user-friendly seizure probability trends. The aim of this study was to evaluate the diagnostic accuracy of the Persyst neonatal SDA in an intensive care setting. METHODS: Single-centre retrospective service evaluation study in neonates undergoing cEEG during intensive care admission to Great Ormond Street Hospital (GOSH) between May 2019 and December 2022. Neonates with <44 weeks corrected gestational age, who had a cEEG recording duration >60 minutes, whilst inpatient in intensive care, were included in the study. One-hour cEEG clips were created for all cases (seizures detected) and controls (seizure-free) and analysed by the Persyst neonatal SDA. Expert neurophysiology reports of the cEEG recordings were used as the gold standard for diagnostic comparison. A receiver operating characteristic (ROC) curve was created using the highest seizure probability in each recording. Optimal seizure probability thresholds for sensitivity and specificity were identified. RESULTS: Eligibility screening produced 49 cases, and 49 seizure-free controls. Seizure prevalence within those patients eligible for the study, was approximately 19% with 35% mortality. The most common case seizure aetiology was hypoxic ischaemic injury (35%) followed by inborn errors of metabolism (18%). The ROC area under the curve was 0.94 with optimal probability thresholds 0.4 and 0.6. Applying a threshold of 0.6, produced 80% sensitivity and 98% specificity. CONCLUSIONS: The Persyst neonatal SDA demonstrates high diagnostic accuracy in identifying neonatal seizures; comparable to the accuracy of the standard Persyst SDA in adult populations, other neonatal SDAs, and amplitude integrated EEG (aEEG). Overdiagnosis of seizures is a risk, particularly from cEEG recording artefact. To fully examine its clinical utility, further investigation of the Persyst neonatal SDA's accuracy is required, as well as confirming the optimal seizure probability thresholds in a larger patient cohort.

Asunto(s)

RESUMEN

Pretreatment steps of current rapid detection methods for mycotoxins in edible oils not only restrict detection efficiency, but also produce organic waste liquid to pollute environment. In this work, a pretreatment-free and eco-friendly rapid detection method for edible oil is established. This proposed method does not require pretreatment operation, and automated quantitative detection could be achieved by directly adding oil samples. According to polarity of target molecules, the content of surfactant in reaction solutions could be adjusted to achieve the quantitative detection of AFB1 in peanut oil and ZEN in corn oil. The recoveries are between 96.5%-110.7% with standard deviation <10.4%, and the limit of detection is 0.17 µg/kg for AFB1 and 4.91 µg/kg for ZEN. This method realizes full automation of the whole chain detection, i.e. sample in-result out, and is suitable for the on-site detection of batches of edible oils samples.

Asunto(s)

RESUMEN

BACKGROUND: Learning to perform strabismus surgery is an essential aspect of ophthalmologists' surgical training. Automated classification strategy for surgical steps can improve the effectiveness of training curricula and the efficient evaluation of residents' performance. To this end, we aimed to develop and validate a deep learning (DL) model for automated detecting strabismus surgery steps in the videos. METHODS: In this study, we gathered 479 strabismus surgery videos from Shanghai Children's Hospital, affiliated to Shanghai Jiao Tong University School of Medicine, spanning July 2017 to October 2021. The videos were manually cut into 3345 clips of the eight strabismus surgical steps based on the International Council of Ophthalmology's Ophthalmology Surgical Competency Assessment Rubrics (ICO-OSCAR: strabismus). The videos dataset was randomly split by eye-level into a training (60%), validation (20%) and testing dataset (20%). We evaluated two hybrid DL algorithms: a Recurrent Neural Network (RNN) based and a Transformer-based model. The evaluation metrics included: accuracy, area under the receiver operating characteristic curve, precision, recall and F1-score. RESULTS: DL models identified the steps in video clips of strabismus surgery achieved macro-average AUC of 1.00 (95% CI 1.00-1.00) with Transformer-based model and 0.98 (95% CI 0.97-1.00) with RNN-based model, respectively. The Transformer-based model yielded a higher accuracy compared with RNN-based models (0.96 vs. 0.83, p < 0.001). In detecting different steps of strabismus surgery, the predictive ability of the Transformer-based model was better than that of the RNN. Precision ranged between 0.90 and 1 for the Transformer-based model and 0.75 to 0.94 for the RNN-based model. The f1-score ranged between 0.93 and 1 for the Transformer-based model and 0.78 to 0.92 for the RNN-based model. CONCLUSION: The DL models can automate identify video steps of strabismus surgery with high accuracy and Transformer-based algorithms show excellent performance when modeling spatiotemporal features of video frames.

Asunto(s)

RESUMEN

Cervical auscultation is a simple, noninvasive method for diagnosing dysphagia, although the reliability of the method largely depends on the subjectivity and experience of the evaluator. Recently developed methods for the automatic detection of swallowing sounds facilitate a rough automatic diagnosis of dysphagia, although a reliable method of detection specialized in the peculiar feature patterns of swallowing sounds in actual clinical conditions has not been established. We investigated a novel approach for automatically detecting swallowing sounds by a method wherein basic statistics and dynamic features were extracted based on acoustic features: Mel Frequency Cepstral Coefficients and Mel Frequency Magnitude Coefficients, and an ensemble learning model combining Support Vector Machine and Multi-Layer Perceptron were applied. The evaluation of the effectiveness of the proposed method, based on a swallowing-sounds database synchronized to a video fluorographic swallowing study compiled from 74 advanced-age patients with dysphagia, demonstrated an outstanding performance. It achieved an F1-micro average of approximately 0.92 and an accuracy of 95.20%. The method, proven effective in the current clinical recording database, suggests a significant advancement in the objectivity of cervical auscultation. However, validating its efficacy in other databases is crucial for confirming its broad applicability and potential impact.

Asunto(s)

RESUMEN

The objective of this experiment was to evaluate the effect on reproductive performance of a targeted reproductive management (TRM) program for first postpartum insemination (AI) that prioritized AI at detected estrus (AIE) by providing different intervals for estrus detection based on records of automated estrus alerts (AEA) during the voluntary waiting period (VWP). A secondary objective was to evaluate the association between occurrence of AEA during the VWP and reproductive performance. Lactating Holstein cows (n = 1,260) fitted with neck behavior monitoring sensors for detection of estrus were randomly assigned to a program that used all-timed AI (TAI) for first service (ALL-TAI; n = 632) or a TRM program that prioritized AIE and used TAI only for cows not detected in estrus (TP-AIE; n = 628). Cows in the ALL-TAI treatment received TAI at 76 ± 3 days in milk (DIM) after a Double-Ovsynch protocol. Cows in the TP-AIE treatment were eligible for AIE for 30 ± 3 or 16 ± 3 d after a 49 d VWP if at least one (n = 346) or no (n = 233) AEA were recorded from 15 to 49 DIM. Cows not AIE received TAI after an Ovsynch protocol with progesterone supplementation at 90 ± 3 or 76 ± 3 DIM if the cow had or did not have AEA during the VWP, respectively. Data were analyzed by logistic and Cox's proportional hazard regression. In the TP-AIE treatment, 69.3 % of cows received AIE and more cows with (83.3 %) than without (45.0 %) AEA during the VWP received AIE. Cows in the TP-AIE (69.0 ± 0.7 d) treatment had fewer days from calving to first AI than cows in the ALL-TAI (75.7 ± 0.8 d) treatment. The proportion of cows pregnant by 150 DIM (ALL-TAI = 59.1 % and TP-AIE = 56.0 %) and the hazard ratio (HR) for time to pregnancy (1.0 [95 % confidence interval: 0.9, 1.2]) did not differ between treatments and median days to pregnancy were 102 and 107 for the ALL-TAI and TP-AIE treatments, respectively. Overall, the ALL-TAI (42.3 %) treatment had more first service pregnancies per AI (P/AI) than the TP-AIE (29.0 %) treatment. Cows with AEA during the VWP had greater P/AI (42.5 % vs. 28.9 %), proportion of cows pregnant by 150 DIM (67.4 % vs. 47.0 %), and HR for time to pregnancy (1.6 [1.4, 1.9]) than cows without AEA during the VWP. We conclude that a TRM program that prioritized AIE based on AEA during the VWP led to a similar pregnancy rate and proportion of cows pregnant by mid-lactation than a program that used all-TAI with extended VWP despite fewer P/AI to first service. Also, expression of estrus during the VWP was associated with improved reproductive performance. Thus, AEA during the VWP could be used as a predictor of reproductive potential for TRM of lactating dairy cows.

Asunto(s)

RESUMEN

Objective: To create a deep learning (DL) model that can accurately detect and classify three distinct types of rat dorsal root ganglion neurons: normal, segmental chromatolysis, and central chromatolysis. The DL model has the potential to improve the efficiency and precision of neuron classification in research related to spinal injuries and diseases. Methods: H&E slide images were divided into an internal training set (80%) and a test set (20%). The training dataset was labeled by two pathologists using pre-defined grades. Using this dataset, a two-component DL model was developed with the first component being a convolutional neural network (CNN) that was trained to detect the region of interest (ROI) and the second component being another CNN used for classification. Results: A total of 240 lumbar dorsal root ganglion (DRG) pathology slide images from rats were analyzed. The internal testing results showed an accuracy of 93.13%, and the external dataset testing demonstrated an accuracy of 93.44%. Conclusion: The DL model demonstrated a level of agreement comparable to that of pathologists in detecting and classifying normal and segmental chromatolysis neurons, although its agreement was slightly lower for central chromatolysis neurons. Significance: DL in improving the accuracy and efficiency of pathological analysis suggests that it may have a role in enhancing medical decision-making.

RESUMEN

AbstractAutonomous sensors provide opportunities to observe organisms across spatial and temporal scales that humans cannot directly observe. By processing large data streams from autonomous sensors with deep learning methods, researchers can make novel and important natural history discoveries. In this study, we combine automated acoustic monitoring with deep learning models to observe breeding-associated activity in the endangered Sierra Nevada yellow-legged frog (Rana sierrae), a behavior that current surveys do not measure. By deploying inexpensive hydrophones and developing a deep learning model to recognize breeding-associated vocalizations, we discover three undocumented R. sierrae vocalization types and find an unexpected temporal pattern of nocturnal breeding-associated vocal activity. This study exemplifies how the combination of autonomous sensor data and deep learning can shed new light on species' natural history, especially during times or in locations where human observation is limited or impossible.

Asunto(s)

RESUMEN

PURPOSE: Focal cortical dysplasia (FCD) is a common etiology of drug-resistant focal epilepsy. Visual identification of FCD is usually time-consuming and depends on personal experience. Herein, we propose an automated type II FCD detection approach utilizing multi-modal data and 3D convolutional neural network (CNN). METHODS: MRI and positron emission tomography (PET) data of 82 patients with FCD were collected, including 55 (67.1%) histopathologically, and 27 (32.9%) radiologically diagnosed patients. Three types of morphometric feature maps and three types of tissue maps were extracted from the T1-weighted images. These maps, T1, and PET images formed the inputs for CNN. Five-fold cross-validations were carried out on the training set containing 62 patients, and the model behaving best was chosen to detect FCD on the test set of 20 patients. Furthermore, ablation experiments were performed to estimate the value of PET data and CNN. RESULTS: On the validation set, FCD was detected in 90.3% of the cases, with an average of 1.7 possible lesions per patient. The sensitivity on the test set was 90.0%, with 1.85 possible lesions per patient. Without the PET data, the sensitivity decreased to 80.0%, and the average lesion number increased to 2.05 on the test set. If an artificial neural network replaced the CNN, the sensitivity decreased to 85.0%, and the average lesion number increased to 4.65. SIGNIFICANCE: Automated detection of FCD with high sensitivity and few false-positive findings is feasible based on multi-modal data. PET data and CNN could improve the performance of automated detection.

Asunto(s)

RESUMEN

Diabetic retinopathy (DR), retinal vein occlusion (RVO), and age-related macular degeneration (AMD) pose significant global health challenges, often resulting in vision impairment and blindness. Automatic detection of these conditions is crucial, particularly in underserved rural areas with limited access to ophthalmic services. Despite remarkable advancements in artificial intelligence, especially convolutional neural networks (CNNs), their complexity can make interpretation difficult. In this study, we curated a dataset consisting of 15,089 color fundus photographs (CFPs) obtained from 8110 patients who underwent fundus fluorescein angiography (FFA) examination. The primary objective was to construct integrated models that merge CNNs with an attention mechanism. These models were designed for a hierarchical multilabel classification task, focusing on the detection of DR, RVO, AMD, and other fundus conditions. Furthermore, our approach extended to the detailed classification of DR, RVO, and AMD according to their respective subclasses. We employed a methodology that entails the translation of diagnostic information obtained from FFA results into CFPs. Our investigation focused on evaluating the models' ability to achieve precise diagnoses solely based on CFPs. Remarkably, our models showcased improvements across diverse fundus conditions, with the ConvNeXt-base + attention model standing out for its exceptional performance. The ConvNeXt-base + attention model achieved remarkable metrics, including an area under the receiver operating characteristic curve (AUC) of 0.943, a referable F1 score of 0.870, and a Cohen's kappa of 0.778 for DR detection. For RVO, it attained an AUC of 0.960, a referable F1 score of 0.854, and a Cohen's kappa of 0.819. Furthermore, in AMD detection, the model achieved an AUC of 0.959, an F1 score of 0.727, and a Cohen's kappa of 0.686. Impressively, the model demonstrated proficiency in subclassifying RVO and AMD, showcasing commendable sensitivity and specificity. Moreover, our models enhanced interpretability by visualizing attention weights on fundus images, aiding in the identification of disease findings. These outcomes underscore the substantial impact of our models in advancing the detection of DR, RVO, and AMD, offering the potential for improved patient outcomes and positively influencing the healthcare landscape.

RESUMEN

The objective of this diagnostic accuracy study was to develop and validate an alert to identify calves at risk for a diarrhea bout using milk feeding behavior data (behavior) from automated milk feeders (AMF). We enrolled Holstein calves (n = 259) as a convenience sample size from 2 facilities that were health scored daily preweaning and offered either 10 or 15 L/d of milk replacer. For alert development, 132 calves were enrolled and the ability of milk intake, drinking speed, and rewarded visits collected from AMF to identify calves at risk for diarrhea was tested. Alerts that had high diagnostic accuracy in the alert development phase were validated using a holdout validation strategy of 127 different calves from the same facilities (all offered 15 L/d) for -3 to 1 d relative to diarrhea diagnosis. We enrolled calves that were either healthy or had a first diarrheal bout (loose feces ≥2 d or watery feces ≥1 d). Relative change and rolling dividends for each milk feeding behavior were calculated for each calf from the previous 2 d. Logistic regression models and receiver operator curves (ROC) were used to assess the diagnostic ability for relative change and rolling dividends behavior relative to alert d) to classify calves at risk for a diarrhea bout from -2 to 0 d relative to diagnosis. To maximize sensitivity (Se), alert thresholds were based on ROC optimal classification cutoff. Diagnostic accuracy was met when the alert had a moderate area under the ROC curve (≥0.70), high accuracy (Acc; ≥0.80), high Se (≥0.80), and very high precision (Pre; ≥0.85). For alert development, deviations in rolling dividend milk intake with drinking speed had the best performance (10 L/d: ROC area under the curve [AUC] = 0.79, threshold ≤0.70; 15 L/d: ROC AUC = 0.82, threshold ≤0.60). Our diagnostic criteria were only met in calves offered 15 L/d (10 L/d: Se 75%, Acc 72%, Pre 92%, specificity [Sp] 55% vs. 15 L/d: Se 91%, Acc 91%, Pre 89%, Sp 73%). For holdout validation, rolling dividend milk intake with drinking speed met diagnostic criteria for one facility (threshold ≤0.60, Se 86%, Acc 82%, Pre 94%, Sp 50%). However, no milk feeding behavior alerts met diagnostic criteria for the second facility due to poor Se (relative change milk intake -0.36 threshold, Se 71%, Acc 70%, and Pre 97%). We suggest that changes in milk feeding behavior may indicate diarrhea bouts in dairy calves. Future research should validate this alert in commercial settings; furthermore, software updates, support, and new analytics might be required for on-farm application to implement these types of alerts.

RESUMEN

Background: Assessing patients' suicide risk is challenging, especially among those who deny suicidal ideation. Primary care providers have poor agreement in screening suicide risk. Patients' speech may provide more objective, language-based clues about their underlying suicidal ideation. Text analysis to detect suicide risk in depression is lacking in the literature. Objective: This study aimed to determine whether suicidal ideation can be detected via language features in clinical interviews for depression using natural language processing (NLP) and machine learning (ML). Methods: This cross-sectional study recruited 305 participants between October 2020 and May 2022 (mean age 53.0, SD 11.77 years; female: n=176, 57%), of which 197 had lifetime depression and 108 were healthy. This study was part of ongoing research on characterizing depression with a case-control design. In this study, 236 participants were nonsuicidal, while 56 and 13 had low and high suicide risks, respectively. The structured interview guide for the Hamilton Depression Rating Scale (HAMD) was adopted to assess suicide risk and depression severity. Suicide risk was clinician rated based on a suicide-related question (H11). The interviews were transcribed and the words in participants' verbal responses were translated into psychologically meaningful categories using Linguistic Inquiry and Word Count (LIWC). Results: Ordinal logistic regression revealed significant suicide-related language features in participants' responses to the HAMD questions. Increased use of anger words when talking about work and activities posed the highest suicide risk (odds ratio [OR] 2.91, 95% CI 1.22-8.55; P=.02). Random forest models demonstrated that text analysis of the direct responses to H11 was effective in identifying individuals with high suicide risk (AUC 0.76-0.89; P<.001) and detecting suicide risk in general, including both low and high suicide risk (AUC 0.83-0.92; P<.001). More importantly, suicide risk can be detected with satisfactory performance even without patients' disclosure of suicidal ideation. Based on the response to the question on hypochondriasis, ML models were trained to identify individuals with high suicide risk (AUC 0.76; P<.001). Conclusions: This study examined the perspective of using NLP and ML to analyze the texts from clinical interviews for suicidality detection, which has the potential to provide more accurate and specific markers for suicidal ideation detection. The findings may pave the way for developing high-performance assessment of suicide risk for automated detection, including online chatbot-based interviews for universal screening.

RESUMEN

Intracranial hemorrhages require an immediate diagnosis to optimize patient management and outcomes, and CT is the modality of choice in the emergency setting. We aimed to evaluate the performance of the first scanner-integrated artificial intelligence algorithm to detect brain hemorrhages in a routine clinical setting. This retrospective study includes 435 consecutive non-contrast head CT scans. Automatic brain hemorrhage detection was calculated as a separate reconstruction job in all cases. The radiological report (RR) was always conducted by a radiology resident and finalized by a senior radiologist. Additionally, a team of two radiologists reviewed the datasets retrospectively, taking additional information like the clinical record, course, and final diagnosis into account. This consensus reading served as a reference. Statistics were carried out for diagnostic accuracy. Brain hemorrhage detection was executed successfully in 432/435 (99%) of patient cases. The AI algorithm and reference standard were consistent in 392 (90.7%) cases. One false-negative case was identified within the 52 positive cases. However, 39 positive detections turned out to be false positives. The diagnostic performance was calculated as a sensitivity of 98.1%, specificity of 89.7%, positive predictive value of 56.7%, and negative predictive value (NPV) of 99.7%. The execution of scanner-integrated AI detection of brain hemorrhages is feasible and robust. The diagnostic accuracy has a high specificity and a very high negative predictive value and sensitivity. However, many false-positive findings resulted in a relatively moderate positive predictive value.

RESUMEN

Coronal Holes (CHs) are regions of open magnetic-field lines, resulting in high-speed solar wind. Accurate detection of CHs is vital for space-weather prediction. This paper presents an intramethod ensemble for coronal-hole detection based on the Active Contours Without Edges (ACWE) segmentation algorithm. The purpose of this ensemble is to develop a confidence map that defines, for all ondisk regions of a solar extreme ultraviolet (EUV) image, the likelihood that each region belongs to a CH based on that region's proximity to, and homogeneity with, the core of identified CH regions. By relying on region homogeneity, and not intensity (which can vary due to various factors, including line-of-sight changes and stray light from nearby bright regions), to define the final confidence of any given region, this ensemble is able to provide robust, consistent delineations of the CH regions. Using the metrics of global consistency error (GCE), local consistency error (LCE), intersection over union (IOU), and the structural similarity index measure (SSIM), the method is shown to be robust to different spatial resolutions maintaining a median IOU >0.75 and minimum SSIM >0.93 even when the segmentation process was performed on an EUV image decimated from 4096×4096 pixels down to 512×512 pixels. Furthermore, using the same metrics, the method is shown to be robust across short timescales, producing segmentation with a mean IOU of 0.826 from EUV images taken at a 1-h cadence, and showing a smooth decay in similarity across all metrics as a function of time, indicating self-consistent segmentations even when corrections for exposure time have not been applied to the data. Finally, the accuracy of the segmentations and confidence maps are validated by considering the skewness (i.e., unipolarity) of the underlying magnetic field.

RESUMEN

Introduction: With increasing use of robotic surgical adjuncts, artificial intelligence and augmented reality in neurosurgery, the automated analysis of digital images and videos acquired over various procedures becomes a subject of increased interest. While several computer vision (CV) methods have been developed and implemented for analyzing surgical scenes, few studies have been dedicated to neurosurgery. Research question: In this work, we present a systematic literature review focusing on CV methodologies specifically applied to the analysis of neurosurgical procedures based on intra-operative images and videos. Additionally, we provide recommendations for the future developments of CV models in neurosurgery. Material and methods: We conducted a systematic literature search in multiple databases until January 17, 2023, including Web of Science, PubMed, IEEE Xplore, Embase, and SpringerLink. Results: We identified 17 studies employing CV algorithms on neurosurgical videos/images. The most common applications of CV were tool and neuroanatomical structure detection or characterization, and to a lesser extent, surgical workflow analysis. Convolutional neural networks (CNN) were the most frequently utilized architecture for CV models (65%), demonstrating superior performances in tool detection and segmentation. In particular, mask recurrent-CNN manifested most robust performance outcomes across different modalities. Discussion and conclusion: Our systematic review demonstrates that CV models have been reported that can effectively detect and differentiate tools, surgical phases, neuroanatomical structures, as well as critical events in complex neurosurgical scenes with accuracies above 95%. Automated tool recognition contributes to objective characterization and assessment of surgical performance, with potential applications in neurosurgical training and intra-operative safety management.

RESUMEN

INTRODUCTION: The prompt detection of intracranial hemorrhage (ICH) on a non-contrast head CT (NCCT) is critical for the appropriate triage of patients, particularly in high volume/high acuity settings. Several automated ICH detection tools have been introduced; however, at present, most suffer from suboptimal specificity leading to false-positive notifications. METHODS: NCCT scans from 4 large databases were evaluated for the presence of an ICH (IPH, IVH, SAH or SDH) of >0.4 ml using fully-automated RAPID ICH 3.0 as compared to consensus detection from at least two neuroradiology experts. Scans were excluded for (1) severe CT artifacts, (2) prior neurosurgical procedures, or (3) recent intravenous contrast. ICH detection accuracy, sensitivity, specificity, positive predictive value, negative predictive value, and positive and negative likelihood ratios by were determined. RESULTS: A total of 881 studies were included. The automated software correctly identified 453/463 ICH-positive cases and 416/418 ICH-negative cases, resulting in a sensitivity of 97.84% and specificity 99.52%, positive predictive value 99.56%, and negative predictive value 97.65% for ICH detection. The positive and negative likelihood ratios for ICH detection were similarly favorable at 204.49 and 0.02 respectively. Mean processing time was <40 seconds. CONCLUSIONS: In this large data set of nearly 900 patients, the automated software demonstrated high sensitivity and specificity for ICH detection, with rare false-positives.

Asunto(s)

RESUMEN

Lumbar spine magnetic resonance imaging (MRI) is a critical diagnostic tool for the assessment of various spinal pathologies, including degenerative disc disease, spinal stenosis, and spondylolisthesis. The accurate identification and quantification of the dural sack cross-sectional area are essential for the evaluation of these conditions. Current manual measurement methods are time-consuming and prone to inter-observer variability. Our study developed and validated deep learning models, specifically U-Net, Attention U-Net, and MultiResUNet, for the automated detection and measurement of the dural sack area in lumbar spine MRI, using a dataset of 515 patients with symptomatic back pain and externally validating the results based on 50 patient scans. The U-Net model achieved an accuracy of 0.9990 and 0.9987 on the initial and external validation datasets, respectively. The Attention U-Net model reported an accuracy of 0.9992 and 0.9989, while the MultiResUNet model displayed a remarkable accuracy of 0.9996 and 0.9995, respectively. All models showed promising precision, recall, and F1-score metrics, along with reduced mean absolute errors compared to the ground truth manual method. In conclusion, our study demonstrates the potential of these deep learning models for the automated detection and measurement of the dural sack cross-sectional area in lumbar spine MRI. The proposed models achieve high-performance metrics in both the initial and external validation datasets, indicating their potential utility as valuable clinical tools for the evaluation of lumbar spine pathologies. Future studies with larger sample sizes and multicenter data are warranted to validate the generalizability of the model further and to explore the potential integration of this approach into routine clinical practice.

RESUMEN

In hereditary spherocytosis (HS), genetic mutations in the cell membrane and cytoskeleton proteins cause structural defects in red blood cells (RBCs). As a result, cells are rigid and misshapen, usually with a characteristic spherical form (spherocytes), too stiff to circulate through microcirculation regions, so they are prone to undergo hemolysis and phagocytosis by splenic macrophages. Mild to severe anemia arises in HS, and other derived symptoms like splenomegaly, jaundice, and cholelithiasis. Although abnormally shaped RBCs can be identified under conventional light microscopy, HS diagnosis relies on several clinical factors and sometimes on the results of complex molecular testing. It is specially challenging when other causes of anemia coexist or after recent blood transfusions. We propose two different approaches to characterize RBCs in HS: (i) an immunofluorescence assay targeting protein band 3, which is affected in most HS cases and (ii) a three-dimensional morphology assay, with living cells, staining the membrane with fluorescent dyes. Confocal laser scanning microscopy (CLSM) was used to carry out both assays, and in order to complement the latter, a software was developed for the automated detection of spherocytes in blood samples. CLSM allowed the precise and unambiguous assessment of cell shape and protein expression.

Asunto(s)

RESUMEN

With a view of the post-COVID-19 world and probable future pandemics, this paper presents an Internet of Things (IoT)-based automated healthcare diagnosis model that employs a mixed approach using data augmentation, transfer learning, and deep learning techniques and does not require physical interaction between the patient and physician. Through a user-friendly graphic user interface and availability of suitable computing power on smart devices, the embedded artificial intelligence allows the proposed model to be effectively used by a layperson without the need for a dental expert by indicating any issues with the teeth and subsequent treatment options. The proposed method involves multiple processes, including data acquisition using IoT devices, data preprocessing, deep learning-based feature extraction, and classification through an unsupervised neural network. The dataset contains multiple periapical X-rays of five different types of lesions obtained through an IoT device mounted within the mouth guard. A pretrained AlexNet, a fast GPU implementation of a convolutional neural network (CNN), is fine-tuned using data augmentation and transfer learning and employed to extract the suitable feature set. The data augmentation avoids overtraining, whereas accuracy is improved by transfer learning. Later, support vector machine (SVM) and the K-nearest neighbors (KNN) classifiers are trained for lesion classification. It was found that the proposed automated model based on the AlexNet extraction mechanism followed by the SVM classifier achieved an accuracy of 98%, showing the effectiveness of the presented approach.

Asunto(s)