Search

Abstract The use of deep learning for OCT image classification could enhance the diagnosis and monitoring of retinal diseases. However, challenges like variability in retinal abnormalities, noise, and artifacts in OCT images limit its clinical use. Our study aimed to evaluate the performance of various deep learning (DL) architectures in classifying retinal pathologies versus healthy cases based on OCT images, under data scarcity and label noise. We examined five DL architectures: ResNet18, ResNet34, ResNet50, VGG16, and InceptionV3. Fine-tuning of the pre-trained models was conducted on 5526 OCT images and reduced subsets down to 21 images to evaluate performance under data scarcity. The performance of models fine-tuned on subsets with label noise levels of 10%, 15%, and 20% was evaluated. All DL architectures achieved high classification accuracy (> 90%) with training sets of 345 or more images. InceptionV3 achieved the highest classification accuracy (99%) when trained on the entire training set. However, classification accuracy decreased and variability increased as sample size decreased. Label noise significantly affected model accuracy. Compensating for labeling errors of 10%, 15%, and 20% requires approximately 4, 9, and 14 times more images in the training set to reach the performance of 345 correctly labeled images. The results showed that DL models fine-tuned on sets of 345 or more OCT images can accurately classify retinal pathologies versus healthy controls. Our findings highlight that while mislabeling errors significantly impact classification performance in OCT analysis, this can be effectively mitigated by increasing the training sample size. By addressing data scarcity and labeling errors, our research aims to improve the real-world application and accuracy of retinal disease management.

Abstract BackgroundOcular following responses (OFRs)—small-amplitude, short-latency reflexive eye movements—have been used to study visual motion processing, with potential diagnostic applications. However, they are difficult to record with commercial, video-based eye trackers, especially in children. ObjectiveWe aimed to design and develop a noninvasive eye tracker specialized for measuring OFRs, trading off lower temporal resolution and a smaller range for higher spatial resolution. MethodsWe developed a high-resolution eye-tracking system based on a high-resolution camera operating in the near-infrared spectral range, coupled with infrared illuminators and a dedicated postprocessing pipeline, optimized to measure OFRs in children. To assess its performance, we: (1) evaluated our algorithm for compensating small head movements in both artificial and real-world settings, (2) compared OFRs measured simultaneously by our system and a reference scleral search coil eye-tracking system, and (3) tested the system’s ability to measure OFRs in a clinical setting with children. ResultsThe simultaneous measurement by our system and a reference system showed that our system achieved an in vivo resolution of approximately 0.06°, which is sufficient for recording OFRs. Head motion compensation was successfully tested, showing a displacement error of less than 5 μm. Finally, robust OFRs were detected in 16 children during recording sessions lasting less than 5 minutes. ConclusionsOur high-resolution, noninvasive eye-tracking system successfully detected OFRs with minimal need for subject cooperation. The system effectively addresses the limits of other OFR measurement methods and offers a versatile solution suitable for clinical applications, particularly in children, where eye tracking is more challenging. The system could potentially be suitable for diagnostic applications, particularly in pediatric populations where early detection of visual disorders like stereodeficiencies is critical.

Abstract Large language models (LLMs) can potentially transform healthcare, particularly in providing the right information to the right provider at the right time in the hospital workflow. This study investigates the integration of LLMs into healthcare, specifically focusing on improving clinical decision support systems (CDSSs) through accurate interpretation of medical guidelines for chronic Hepatitis C Virus infection management. Utilizing OpenAI’s GPT-4 Turbo model, we developed a customized LLM framework that incorporates retrieval augmented generation (RAG) and prompt engineering. Our framework involved guideline conversion into the best-structured format that can be efficiently processed by LLMs to provide the most accurate output. An ablation study was conducted to evaluate the impact of different formatting and learning strategies on the LLM’s answer generation accuracy. The baseline GPT-4 Turbo model’s performance was compared against five experimental setups with increasing levels of complexity: inclusion of in-context guidelines, guideline reformatting, and implementation of few-shot learning. Our primary outcome was the qualitative assessment of accuracy based on expert review, while secondary outcomes included the quantitative measurement of similarity of LLM-generated responses to expert-provided answers using text-similarity scores. The results showed a significant improvement in accuracy from 43 to 99% (p

Swimmer’s shoulder is a common condition among elite swimmers, often leading to pain and reduced performance. Fatigue can exacerbate this condition by affecting shoulder strength, proprioception, and range of motion, potentially increasing the risk of overuse injuries. This preliminary study aimed to evaluate the impact of physiotherapy treatment and the effects of fatigue on shoulder kinematics using inertial and magnetic measurement units (IMUs). Five male swimmers (aged 21–27) with at least 3 years of training and suffering from swimmer’s shoulder pain participated in the study. The protocol included three sessions: dry front crawl exercises using one arm in the first and third sessions, and a fatiguing swimming exercise in the second. IMUs were used to capture 3D rotation angles, focusing on flexion/extension, abduction/adduction, and internal/external rotations during the first and third sessions. Stroke amplitude was analyzed before and after the physiotherapy treatment and fatiguing exercise. The results showed a significant increase in internal/external rotation amplitude post-fatigue before physiotherapy (p = 0.03), with a non-significant decrease in flexion/extension after treatment, suggesting improved shoulder stabilization. Despite these preliminary findings being based on a reduced number of participants, they indicate that physiotherapy may enhance shoulder motion control in swimmers with shoulder pain. Nevertheless, further studies with larger cohorts are needed to confirm these results.

Neonatal hyperbilirubinemia (NH) is a common condition in newborns, with elevated bilirubin levels potentially causing neurological damage or death. Accurate and timely measurements of total serum bilirubin are essential to prevent these outcomes. Direct spectrophotometry, a reliable method for measuring bilirubin, is particularly useful in constrained settings due to its potential for portable low-cost instrumentation. However, this method is susceptible to interference from hemoglobin, often present due to hemolysis. Typically, this interference is reduced using complex optical filters, reagents, multiple wavelengths, or combinations thereof, which increase costs and complexity while reducing usability. This study presents a hemoglobin compensation algorithm applied to a simple, portable, two-wavelength (465 and 590 nm) reflectance photometer designed to receive 30 µL of plasma or whole blood samples and perform the measurement without any reagents. Testing across five bilirubin and hemoglobin levels (4.96 to 28 mg/dL and 0.06 to 0.99 g/dL, respectively) demonstrated that the algorithm effectively reduces hemoglobin interference and overestimation errors. The overall root mean square error was reduced from 4.86 to 1.45 mg/dL, while the measurement bias decreased from −4.46 to −0.10 mg/dL. This substantial reduction in overestimation errors supports future clinical trials with neonatal blood samples.

Brain–computer interfaces (BCIs) are promising tools for motor neurorehabilitation. Achieving a balance between classification accuracy and system responsiveness is crucial for real-time applications. This study aimed to assess how the duration of time windows affects performance, specifically classification accuracy and the false positive rate, to optimize the temporal parameters of MI-BCI systems. We investigated the impact of time window duration on classification accuracy and false positive rate, employing Linear Discriminant Analysis (LDA), Multilayer Perceptron (MLP), and Support Vector Machine (SVM) on data acquired from six post-stroke patients and on the external BCI IVa dataset. EEG signals were recorded and processed using the Common Spatial Patterns (CSP) algorithm for feature extraction. Our results indicate that longer time windows generally enhance classification accuracy and reduce false positives across all classifiers, with LDA performing the best. However, to maintain the real-time responsiveness, crucial for practical applications, a balance must be struck. The results suggest an optimal time window of 1–2 s, offering a trade-off between classification performance and excessive delay to guarantee the system responsiveness. These findings underscore the importance of temporal optimization in MI-BCI systems to improve usability in real rehabilitation scenarios.

Abstract Cognitive impairment is one of the most prevalent symptoms of post Severe Acute Respiratory Syndrome COronaVirus 2 (SARS-CoV-2) state, which is known as Long COVID. Advanced neuroimaging techniques may contribute to a better understanding of the pathophysiological brain changes and the underlying mechanisms in post-COVID-19 subjects. We aimed at investigating regional cerebral perfusion alterations in post-COVID-19 subjects who reported a subjective cognitive impairment after a mild SARS-CoV-2 infection, using a non-invasive Arterial Spin Labeling (ASL) MRI technique and analysis. Using MRI-ASL image processing, we investigated the brain perfusion alterations in 24 patients (53.0 ± 14.5 years, 15F/9M) with persistent cognitive complaints in the post COVID-19 period. Voxelwise and region-of-interest analyses were performed to identify statistically significant differences in cerebral blood flow (CBF) maps between post-COVID-19 patients, and age and sex matched healthy controls (54.8 ± 9.1 years, 13F/9M). The results showed a significant hypoperfusion in a widespread cerebral network in the post-COVID-19 group, predominantly affecting the frontal cortex, as well as the parietal and temporal cortex, as identified by a non-parametric permutation testing (p

Inertial sensors (IMUs) have been recently widely used in exercise and rehabilitation science as they can provide reliable quantitative measures of range of motion (RoM). Moreover, the pressure pain threshold (PPT) evaluation provides an objective measure of pain sensation in different body areas. The aim of this study was to evaluate the efficacy of physiotherapy treatment in people with adhesive capsulitis in terms of RoM and pain improvement measured by IMUs and the PPT. A combined prospective cohort/cross-sectional study was conducted. Nineteen individuals with adhesive capsulitis (10/19 females, 54 ± 8 years) and nineteen healthy controls (10/19 females, 51 ± 6 years) were evaluated for active glenohumeral joint RoM and PPT on shoulder body areas. Then, individuals with adhesive capsulitis were invited to 20 sessions of a physiotherapy protocol, and the assessments were repeated within 1 week from the last session. The range of motion in the flexion (p = 0.001) and abduction (p < 0.001) of the shoulder increased significantly after the physiotherapy protocol. Similarly, the PPT was found to increase significantly in all the assessed shoulder body areas, leading to no significant differences compared to the healthy controls. IMU and PPT assessments could be used to evaluate the efficacy of physical therapy in people with adhesive capsulitis.

Attention-deficit/hyperactivity disorder (ADHD) is a neurobehavioral disorder with known brain abnormalities but no biomarkers to support clinical diagnosis. Recently, EEG analysis methods such as functional connectivity have rekindled interest in using EEG for ADHD diagnosis. Most studies have focused on resting-state EEG, while connectivity during sleep and spindle activity has been underexplored. Here we present the results of a preliminary study exploring spindle-related connectivity as a possible biomarker for ADHD. We compared sensor-space connectivity parameters in eight children with ADHD and nine age/sex-matched healthy controls during sleep, before, during, and after spindle activity in various frequency bands. All connectivity parameters were significantly different between the two groups in the delta and gamma bands, and Principal Component Analysis (PCA) in the gamma band distinguished ADHD from healthy subjects. Cluster coefficient and path length values in the sigma band were also significantly different between epochs, indicating different spindle-related brain activity in ADHD.

Abstract Background Sublingual varices (SV) are dilatations of tortuous veins that increased with age. Previous studies showed that this pathology could be correlated to some risk factors such as hypertension, age, gender and diabetes mellitus. In this study we evaluated, on a large number of subjects, the relationship between SV and different grades of hypertension as well as some risk factors extending the analysis to new risk factors such as dyslipidemia, obesity and antihypertensive therapy, modelling a possible dependence of SV on all these factors. Methods In the study 1008 subjects, 284 with and 724 without SV, were examined. The blood pressure was measured in office condition and, to exclude subjects with white coat syndrome or masked hypertension, also using a 24 h Holter pressure monitor. Hypertensive subjects were divided in resistant, drugs controlled (compensated) and patients with prior unknown hypertension (new diagnosed) groups. The presence or absence of SV as well as of the risk factors was assessed clinically. We tested the influence of age on the presence of SV by using the chi-square test and the relation between each risk factor and SV by the Cochran–Mantel–Haenszel test. Finally, we carried out a multivariate regression tree analysis in order to predict the presence of SV. Results We confirmed the influence of age on SV and found a significant relationship between SV and both the compensated and resistant hypertension grades. We highlighted a relationship between SV and dyslipidemia in subjects with new diagnosed hypertension, and between SV and smoking in subjects with compensated hypertension grade. The regression tree showed a classification accuracy of about 75% using as variables hypertension grades, age and antihypertensive treatment. Conclusions We confirmed the SV dependence on age, resistant hypertension and smoking, highlighting a new association with dyslipidemia in new diagnosed hypertensive subjects and new relations depending on the hypertension grades. Thus, the SV inspection could be used to suggest a lipidologist as well as a hypertension specialist visit for a pharmacological and pressure check particularly in subjects presenting SV and dyslipidemia. However, further parameters are to be considered to improve the sensitivity of the prognostic tree model.

Hematocrit (HCT) is a crucial parameter for both adult and pediatric patients, indicating potentially severe pathological conditions. Most common methods for HCT assessment are microhematocrit and automated analyzers; however, developing countries present specific needs often not addressed by these technologies. Paper-based devices can be suitable for those environments being inexpensive, rapid, easy to use, and portable. The aim of this study is to describe and validate against a reference method, a novel HCT estimation method based on penetration velocity in lateral flow test strips complying with the requirements in low- or middle-income country (LMIC) scenarios. To calibrate and test the proposed method, 145 blood samples of 105 healthy neonates with gestational age greater than 37 weeks were collected (29 calibration set, 116 test set) in the range of HCT values (31.6–72.5%). The time difference (Δt) from the whole blood sample loading into the test strip instant till the nitrocellulose membrane saturation instant was measured by a reflectance meter. A nonlinear relation was observed between HCT and Δt and was estimated by a third-degree polynomial equation (R2 = 0.91) valid in 30% to 70% HCT interval. The proposed model was subsequently used to estimate HCT values on the test set showing a good agreement between the estimated HCT and the HCT measured by the reference method (r = 0.87, p < 0.001), with a low mean difference of 0.53 ± 5.04% and a slight trend of overestimation for higher hematocrit values. The mean absolute error was 4.29%, while the maximum absolute error was 10.69%. Although the proposed method did not present a sufficient accuracy to be used for diagnostic purposes, it could be suitable as a fast, low-cost, easy-to-use screening tool especially in LMIC scenarios.

There is a growing research interest in wireless non-invasive solutions for core temperature estimation and their application in clinical settings. This study aimed to investigate the use of a novel wireless non-invasive heat flux-based thermometer in acute stroke patients admitted to a stroke unit and compare the measurements with the currently used infrared (IR) tympanic temperature readings. The study encompassed 30 acute ischemic stroke patients who underwent continuous measurement (Tcore) with the novel wearable non-invasive CORE device. Paired measurements of Tcore and tympanic temperature (Ttym) by using a standard IR-device were performed 3–5 times/day, yielding a total of 305 measurements. The predicted core temperatures (Tcore) were significantly correlated with Ttym (r = 0.89, p < 0.001). The comparison of the Tcore and Ttym measurements by Bland–Altman analysis showed a good agreement between them, with a low mean difference of 0.11 ± 0.34 °C, and no proportional bias was observed (B = −0.003, p = 0.923). The Tcore measurements correctly predicted the presence or absence of Ttym hyperthermia or fever in 94.1% and 97.4% of cases, respectively. Temperature monitoring with a novel wireless non-invasive heat flux-based thermometer could be a reliable alternative to the Ttym method for assessing core temperature in acute ischemic stroke patients.

Circadian heart rate (HR) is influenced by hypertension and other cardiovascular risk factors particularly smoking, obesity and dyslipidemia. Until now, to evaluate the HR changes due to presence of these risk factors, a single HR office measure or a mean evaluated on day time or night time or 24h was used. However, since HR shows a circadian behavior, a single value represents only a rough approximation of this behavior. In this study, we analyzed the influence of smoking, obesity and dyslipidemia on the circadian rhythm in normotensive and hypertensive subject groups presenting only one of these risk factors. The 24h HR recordings of 170 normotensive (83 without risk factors, 20 smokers, 44 with dyslipidemia, 23 obese) and 353 hypertensive (169 without risk factors, 32 smokers, 99 with dyslipidemia, 53 obese) subjects were acquired using a Holter Blood Pressure Monitor. Results highlighted a specific circadian behavior with three characteristic periods presenting different HR means and rates of HR change in the eight subject groups. The slopes could be used both to estimate the morning HR surge associated with acute cardiovascular effects in the awakening and to evaluate the decline during the night. Moreover, we suggest to use three HR mean values (one for each identified period of the day) rather than two HR values to better describe the circadian HR behavior. Furthermore, smoking increased and dyslipidemia decreased mean HR values from 10:00 to 04:00, both in normotensive and hypertensive subjects in comparison with subjects without risk factors. In this time interval, hypertensive obese subjects showed higher values while normotensive ones presented quite similar values than subjects without risk factors. During the awakening (05:00-10:00) the slopes were similar among all groups with no significant difference among the mean HR values.