Search

Objective Despite extensive exploration of potential biomarkers of cardiovascular diseases (CVDs) derived from retinal images, it remains unclear how retinal images contribute to CVD risk profiling and how the results can inform lifestyle modifications. Therefore, we aimed to determine the performance of cardiovascular risk prediction model from retinal images via explicitly estimating 10 traditional CVD risk factors and compared with the model based on actual risk measurements.Design A prospective cohort study design.Setting The UK Biobank (UKBB), a prospective cohort study, following the health conditions including CVD outcomes of adults recruited between 2006 and 2010.Participants A subset of data from the UKBB which contains 52 297 entries with retinal images and 5-year cumulative incidence of major adverse cardiovascular events (MACE) was used. Our dataset is split into 3:1:1 as training set (n=31 403), validation set (n=10 420) and testing set (n=10 474). We developed a deep learning (DL) model to predict 5-year MACE using a two-stage DL neural network.Primary and secondary outcome measures We computed accuracy, area under the receiver operating characteristic curve (AUC) and compared variations in the risk prediction models combining CVD risk factors and retinal images.Results The first-stage DL model demonstrated that the 10 CVD risk factors can be estimated from a given retinal image with an accuracy ranging between 65.2% and 89.8% (overall AUC of 0.738 with 95% CI: 0.710 to 0.766). In MACE prediction, our model outperformed the traditional score-based models, with 8.2% higher AUC than Systematic COronary Risk Evaluation (SCORE), 3.5% for SCORE 2 and 7.1% for the Framingham Risk Score (with p value

Studies employing neural networks to classify emotions from brain waves and other biological signals provide a quantitative perspective on understanding human physiological phenomena. Typically, multimodal networks process combined data without considering the relationships between electrodes, such as in electroencephalograms (EEG) where data are gathered from multiple inputs. However, incorporating electrode relationships when combining data may improve the model accuracy. This study explores EEG data, often treated as a single modality, input into networks of varied structures as a multimodal data stream to evaluate accuracy. Additionally, it investigates the effect of input electrode combination patterns on accuracy. The results underscore the importance of designing appropriate electrode models when integrating EEG data into networks with various structures. Under the conditions of this study, the highest accuracy of 89.08% was obtained by the most appropriate model, significantly surpassing others. Therefore, when incorporating multi-channel EEG data into neural networks, the structure of the model’s specific section receiving the EEG signal must be considered.

Abstract Rapid and precise intraoperative diagnosing systems are required for improving surgical outcomes and patient prognosis. Because of the poor quality and time-intensive process of the prevalent frozen section procedure, various intraoperative diagnostic imaging systems have been explored. Microscopy with ultraviolet surface excitation (MUSE) is an inexpensive, maintenance-free, and rapid imaging technique that yields images like thin-sectioned samples without sectioning. However, pathologists find it nearly impossible to assign diagnostic labels to MUSE images of unfixed specimens; thus, AI for intraoperative diagnosis cannot be trained in a supervised learning manner. In this study, we propose a deep-learning pipeline model for lymph node metastasis detection, in which CycleGAN translate MUSE images of unfixed lymph nodes to formalin-fixed paraffin-embedded (FFPE) sample, and diagnostic prediction is performed using deep convolutional neural network trained on FFPE sample images. Our pipeline yielded an average accuracy of 84.6% when using each of the three deep convolutional neural networks, which is a 18.3% increase over the classification-only model without CycleGAN. The modality translation to FFPE sample images using CycleGAN can be applied to various intraoperative diagnostic imaging systems and eliminate the difficulty for pathologists in labeling new modality images in clinical sites. We anticipate our pipeline to be a starting point for accurate rapid intraoperative diagnostic systems for new imaging modalities, leading to healthcare quality improvement.

Video summarization aims to select the most informative subset of frames in a video to facilitate efficient video browsing. Past efforts have invariantly involved training summarization models with annotated summaries or heuristic objectives. In this work, we reveal that features pre-trained on image-level tasks contain rich semantic information that can be readily leveraged to quantify frame-level importance for zero-shot video summarization. Leveraging pre-trained features and contrastive learning, we propose three metrics featuring a desirable keyframe: local dissimilarity, global consistency, and uniqueness. We show that the metrics can well-capture the diversity and representativeness of frames commonly used for the unsupervised generation of video summaries, demonstrating competitive or better performance compared to past methods when no training is needed. We further propose a contrastive learning-based pre-training strategy on unlabeled videos to enhance the quality of the proposed metrics and, thus, improve the evaluated performance on the public benchmarks TVSum and SumMe.

Abstract Background This work aims to develop a deep learning model, assessing atlantoaxial subluxation (AAS) in rheumatoid arthritis (RA), which can often be ambiguous in clinical practice. Methods We collected 4691 X-ray images of the cervical spine of the 906 patients with RA. Among these images, 3480 were used for training the deep learning model, 803 were used for validating the model during the training process, and the remaining 408 were used for testing the performance of the trained model. The two-dimensional key points’ detection model of Deep High-Resolution Representation Learning for Human Pose Estimation was adopted as the base convolutional neural network model. The model inferred four coordinates to calculate the atlantodental interval (ADI) and space available for the spinal cord (SAC). Finally, these values were compared with those by clinicians to evaluate the performance of the model. Results Among the 408 cervical images for testing the performance, the trained model correctly identified the four coordinates in 99.5% of the dataset. The values of ADI and SAC were positively correlated among the model and two clinicians. The sensitivity of AAS diagnosis with ADI or SAC by the model was 0.86 and 0.97 respectively. The specificity of that was 0.57 and 0.5 respectively. Conclusions We present the development of a deep learning model for the evaluation of cervical lesions of patients with RA. The model was demonstrably shown to be useful for quantitative evaluation.

Abstract Purpose Estimating the surgery length has the potential to be utilized as skill assessment, surgical training, or efficient surgical facility utilization especially if it is done in real-time as a remaining surgery duration (RSD). Surgical length reflects a certain level of efficiency and mastery of the surgeon in a well-standardized surgery such as cataract surgery. In this paper, we design and develop a real-time RSD estimation method for cataract surgery that does not require manual labeling and is transferable with minimum fine-tuning. Methods A regression method consisting of convolutional neural networks (CNNs) and long short-term memory (LSTM) is designed for RSD estimation. The model is firstly trained and evaluated for the single main surgeon with a large number of surgeries. Then, the fine-tuning strategy is used to transfer the model to the data of the other two surgeons. Mean Absolute Error (MAE in seconds) was used to evaluate the performance of the RSD estimation. The proposed method is compared with the naïve method which is based on the statistic of the historical data. A transferability experiment is also set to demonstrate the generalizability of the method. Result The mean surgical time for the sample videos was 318.7 s (s) (standard deviation 83.4 s) for the main surgeon for the initial training. In our experiments, the lowest MAE of 19.4 s (equal to about 6.4% of the mean surgical time) is achieved by our best-trained model for the independent test data of the main target surgeon. It reduces the MAE by 35.5 s (-10.2%) compared to the naïve method. The fine-tuning strategy transfers the model trained for the main target to the data of other surgeons with only a small number of training data (20% of the pre-training). The MAEs for the other two surgeons are 28.3 s and 30.6 s with the fine-tuning model, which decreased by -8.1 s and -7.5 s than the Per-surgeon model (average declining of -7.8 s and 1.3% of video duration). External validation study with Cataract-101 outperformed 3 reported methods of TimeLSTM, RSDNet, and CataNet. Conclusion An approach to build a pre-trained model for estimating RSD estimation based on a single surgeon and then transfer to other surgeons demonstrated both low prediction error and good transferability with minimum fine-tuning videos.

Lensless imaging protects visual privacy by capturing heavily blurred images that are imperceptible for humans to recognize the subject but contain enough information for machines to infer information. Unfortunately, protecting visual privacy comes with a reduction in recognition accuracy and vice versa. We propose a learnable lensless imaging framework that protects visual privacy while maintaining recognition accuracy. To make captured images imperceptible to humans, we designed several loss functions based on total variation, invertibility, and the restricted isometry property. We studied the effect of privacy protection with blurriness on the identification of personal identity via a quantitative method based on a subjective evaluation. Moreover, we validate our simulation by implementing a hardware realization of lensless imaging with photo-lithographically printed masks

IntroductionDeveloping approaches for early detection of possible risk clusters for mental health problems among undergraduate university students is warranted to reduce the duration of untreated illness (DUI). However, little is known about indicators of need for care by others. Herein, we aimed to clarify the specific value of study engagement and lifestyle habit variables in predicting potentially high-risk cluster of mental health problems among undergraduate university students.MethodsThis cross-sectional study used a web-based demographic questionnaire [the Utrecht Work Engagement Scale for Students (UWES-S-J)] as study engagement scale. Moreover, information regarding life habits such as sleep duration and meal frequency, along with mental health problems such as depression and fatigue were also collected. Students with both mental health problems were classified as high risk. Characteristics of students in the two groups were compared. Univariate logistic regression was performed to identify predictors of membership. Receiver Operating Characteristic (ROC) curve was used to clarify the specific values that differentiated the groups in terms of significant predictors in univariate logistic analysis. Cut-off point was calculated using Youden index. Statistical significance was set at p

Abstract This study sought to develop a deep learning-based diagnostic algorithm for plaque vulnerability by analyzing intravascular optical coherence tomography (OCT) images and to investigate the relation between AI-plaque vulnerability and clinical outcomes in patients with coronary artery disease (CAD). A total of 1791 study patients who underwent OCT examinations were recruited from a multicenter clinical database, and the OCT images were first labeled as either normal, a stable plaque, or a vulnerable plaque by expert cardiologists. A DenseNet-121-based deep learning algorithm for plaque characterization was developed by training with 44,947 prelabeled OCT images, and demonstrated excellent differentiation among normal, stable plaques, and vulnerable plaques. Patients who were diagnosed with vulnerable plaques by the algorithm had a significantly higher rate of both events from the OCT-observed segments and clinical events than the patients with normal and stable plaque (log-rank p

In this paper, we present a prototype pseudo-direct time-of-flight (ToF) CMOS image sensor, achieving high distance accuracy, precision, and robustness to multipath interference. An indirect ToF (iToF)-based image sensor, which enables high spatial resolution, is used to acquire temporal compressed signals in the charge domain. Whole received light waveforms, like those acquired with conventional direct ToF (dToF) image sensors, can be obtained after image reconstruction based on compressive sensing. Therefore, this method has the advantages of both dToF and iToF depth image sensors, such as high resolution, high accuracy, immunity to multipath interference, and the absence of motion artifacts. Additionally, two approaches to refine the depth resolution are explained: (1) the introduction of a sub-time window; and (2) oversampling in image reconstruction and quadratic fitting in the depth calculation. Experimental results show the separation of two reflections 40 cm apart under multipath interference conditions and a significant improvement in distance precision down to around 1 cm. Point cloud map videos demonstrate the improvements in depth resolution and accuracy. These results suggest that the proposed method could be a promising approach for virtually implementing dToF imaging suitable for challenging environments with multipath interference.

The morphological feature of retinal arterio-venous crossing patterns is a valuable source of cardiovascular risk stratification as it directly captures vascular health. Although Scheie's classification, which was proposed in 1953, has been used to grade the severity of arteriolosclerosis as diagnostic criteria, it is not widely used in clinical settings as mastering this grading is challenging as it requires vast experience. In this paper, we propose a deep learning approach to replicate a diagnostic process of ophthalmologists while providing a checkpoint to secure explainability to understand the grading process. The proposed pipeline is three-fold to replicate a diagnostic process of ophthalmologists. First, we adopt segmentation and classification models to automatically obtain vessels in a retinal image with the corresponding artery/vein labels and find candidate arterio-venous crossing points. Second, we use a classification model to validate the true crossing point. At last, the grade of severity for the vessel crossings is classified. To better address the problem of label ambiguity and imbalanced label distribution, we propose a new model, named multi-diagnosis team network (MDTNet), in which the sub-models with different structures or different loss functions provide different decisions. MDTNet unifies these diverse theories to give the final decision with high accuracy. Our automated grading pipeline was able to validate crossing points with precision and recall of 96.3% and 96.3%, respectively. Among correctly detected crossing points, the kappa value for the agreement between the grading by a retina specialist and the estimated score was 0.85, with an accuracy of 0.92. The numerical results demonstrate that our method can achieve a good performance in both arterio-venous crossing validation and severity grading tasks following the diagnostic process of ophthalmologists. By the proposed models, we could build a pipeline reproducing ophthalmologists' diagnostic process without requiring subjective feature extractions. The code is available (https://github.com/conscienceli/MDTNet).

A camera captures multidimensional information of the real world by convolving it into two dimensions using a sensing matrix. The original multidimensional information is then reconstructed from captured images. Traditionally, multidimensional information has been captured by uniform sampling, but by optimizing the sensing matrix, we can capture images more efficiently and reconstruct multidimensional information with high quality. Although compressive video sensing requires random sampling as a theoretical optimum, when designing the sensing matrix in practice, there are many hardware limitations (such as exposure and color filter patterns). Existing studies have found random sampling is not always the best solution for compressive sensing because the optimal sampling pattern is related to the scene context, and it is hard to manually design a sampling pattern and reconstruction algorithm. In this paper, we propose an end-to-end learning approach that jointly optimizes the sampling pattern as well as the reconstruction decoder. We applied this deep sensing approach to the video compressive sensing problem. We modeled the spatio–temporal sampling and color filter pattern using a convolutional neural network constrained by hardware limitations during network training. We demonstrated that the proposed method performs better than the manually designed method in gray-scale video and color video acquisitions.

Semantic video segmentation is a key challenge for various applications. This paper presents a new model named Noisy-LSTM, which is trainable in an end-to-end manner, with convolutional LSTMs (ConvLSTMs) to leverage the temporal coherence in video frames, together with a simple yet effective training strategy that replaces a frame in a given video sequence with noises. Our training strategy spoils the temporal coherence in video frames and thus makes the temporal links in ConvLSTMs unreliable; this may consequently improve the ability of the model to extract features from video frames and serve as a regularizer to avoid overfitting, without requiring extra data annotations or computational costs. Experimental results demonstrate that the proposed model can achieve state-of-the-art performances on both the CityScapes and EndoVis2018 datasets. The code for the proposed method is available at https://github.com/wbw520/NoisyLSTM.

Learners are expected to stay wakeful and focused while interacting with e-learning platforms. Although wakefulness of learners strongly relates to educational outcomes, detecting drowsy learning behaviors only from log data is not an easy task. In this study, we describe the results of our research to model learners’ wakefulness based on multimodal data generated from heart rate, seat pressure, and face recognition. We collected multimodal data from learners in a blended course of informatics and conducted two types of analysis on them. First, we clustered features based on learners’ wakefulness labels as generated by human raters and ran a statistical analysis. This analysis helped us generate insights from multimodal data that can be used to inform learner and teacher feedback in multimodal learning analytics. Second, we trained machine learning models with multiclass-Support Vector Machine (SVM), Random Forest (RF) and CatBoost Classifier (CatBoost) algorithms to recognize learners’ wakefulness states automatically. We achieved an average macro-F1 score of 0.82 in automated user-dependent models with CatBoost. We also showed that compared to unimodal data from each sensor, the multimodal sensor data can improve the accuracy of models predicting the wakefulness states of learners while they are interacting with e-learning platforms.

Abstract Topic modeling that can automatically assign topics to legal documents is very important in the domain of computational law. The relevance of the modeled topics strongly depends on the legal context they are used in. On the other hand, references to laws and prior cases are key elements for judges to rule on a case. Taken together, these references form a network, whose structure can be analysed with network analysis. However, the content of the referenced documents may not be always accessed. Even in that case, the reference structure itself shows that documents share latent similar characteristics. We propose to use this latent structure to improve topic modeling of law cases using document homophily. In this paper, we explore the use of homophily networks extracted from two types of references: prior cases and statute laws, to enhance topic modeling on legal case documents. We conduct in detail, an analysis on a dataset consisting of rich legal cases, i.e., the COLIEE dataset, to create these networks. The homophily networks consist of nodes for legal cases, and edges with weights for the two families of references between the case nodes. We further propose models to use the edge weights for topic modeling. In particular, we propose a cutting model and a weighting model to improve the relational topic model (RTM). The cutting model uses edges with weights higher than a threshold as document links in RTM; the weighting model uses the edge weights to weight the link probability function in RTM. The weights can be obtained either from the co-citations or from the cosine similarity based on an embedding of the homophily networks. Experiments show that the use of the homophily networks for topic modeling significantly outperforms previous studies, and the weighting model is more effective than the cutting model.

This study aimed to clarify the adaptation features of University students exposed to fully online education during the novel coronavirus disease 2019 (COVID-19) pandemic and to identify accompanying mental health problems and predictors of school adaptation. The pandemic has forced many universities to transition rapidly to delivering online education. However, little is known about the impact of this drastic change on students' school adaptation. This cross-sectional study used an online questionnaire, including assessments of impressions of online education, study engagement, mental health, and lifestyle habits. In total, 1,259 students were assessed. The characteristics of school adaptation were analyzed by a two-step cluster analysis. The proportion of mental health problems was compared among different groups based on a cluster analysis. A logistic regression analysis was used to identify predictors of cluster membership. P-values < 0.05 were considered statistically significant. The two-step cluster analysis determined three clusters: school adaptation group, school maladaptation group, and school over-adaptation group. The last group significantly exhibited the most mental health problems. Membership of this group was significantly associated with being female (OR = 1.42; 95% CI 1.06–1.91), being older (OR = 1.21; 95% CI 1.01–1.44), those who considered online education to be less beneficial (OR = 2.17; 95% CI 1.64–2.88), shorter sleep time on weekdays (OR = 0.826; 95% CI 0.683–.998), longer sleep time on holidays (OR = 1.21; 95% CI 1.03–1.43), and worse restorative sleep (OR = 2.27; 95% CI 1.81–2.86). The results suggest that academic staff should understand distinctive features of school adaptation owing to the rapid transition of the educational system and should develop support systems to improve students' mental health. They should consider ways to incorporate online classes with their lectures to improve students' perceived benefits of online education. Additionally, educational guidance on lifestyle, such as sleep hygiene, may be necessary.

An ultra-high-speed computational CMOS image sensor with a burst frame rate of 303 megaframes per second, which is the fastest among the solid-state image sensors, to our knowledge, is demonstrated. This image sensor is compatible with ordinary single-aperture lenses and can operate in dual modes, such as single-event filming mode or multi-exposure imaging mode, by reconfiguring the number of exposure cycles. To realize this frame rate, the charge modulator drivers were adequately designed to suppress the peak driving current taking advantage of the operational constraint of the multi-tap charge modulator. The pixel array is composed of macropixels with 2 × 2 4-tap subpixels. Because temporal compressive sensing is performed in the charge domain without any analog circuit, ultrafast frame rates, small pixel size, low noise, and low power consumption are achieved. In the experiments, single-event imaging of plasma emission in laser processing and multi-exposure transient imaging of light reflections to extend the depth range and to decompose multiple reflections for time-of-flight (TOF) depth imaging with a compression ratio of 8× were demonstrated. Time-resolved images similar to those obtained by the direct-type TOF were reproduced in a single shot, while the charge modulator for the indirect TOF was utilized.

Multi-path interference causes depth errors in indirect time-of-flight (ToF) cameras. In this paper, resolving multi-path interference caused by surface reflections using a multi-tap macro-pixel computational CMOS image sensor is demonstrated. The imaging area is implemented by an array of macro-pixels composed of four subpixels embodied by a four-tap lateral electric field charge modulator (LEFM). This sensor can simultaneously acquire 16 images for different temporal shutters. This method can reproduce more than 16 images based on compressive sensing with multi-frequency shutters and sub-clock shifting. In simulations, an object was placed 16 m away from the sensor, and the depth of an interference object was varied from 1 to 32 m in 1 m steps. The two reflections were separated in two stages: coarse estimation based on a compressive sensing solver and refinement by a nonlinear search to investigate the potential of our sensor. Relative standard deviation (precision) and relative mean error (accuracy) were evaluated under the influence of photon shot noise. The proposed method was verified using a prototype multi-tap macro-pixel computational CMOS image sensor in single-path and dual-path situations. In the experiment, an acrylic plate was placed 1 m or 2 m and a mirror 9.3 m from the sensor.

Climate change is one of the most important issues for humanity. To defuse this problem, it is considered necessary to improve energy efficiency, make energy sources cleaner, and reduce energy consumption in urban areas. The Japanese government has recommended an air conditioner setting of 28°C in summer and 20°C in winter since 2005. The aim of this setting is to save energy by keeping room temperatures constant. However, it is unclear whether this is an appropriate temperature for workers and students. This study examined whether thermal environments influence task performance over time. To examine whether the relationship between task performance and thermal environments influences the psychological states of participants, we recorded their subjective rating of mental workload along with their working memory score, electroencephalogram (EEG), heart rate variability, skin conductance level (SCL), and tympanum temperature during the task and compared the results among different conditions. In this experiment, participants were asked to read some texts and answer questions related to those texts. Room temperature (18, 22, 25, or 29°C) and humidity (50%) were manipulated during the task and participants performed the task at these temperatures. The results of this study showed that the temporal cost of task and theta power of EEG, which is an index for concentration, decreased over time. However, subjective mental workload increased with time. Moreover, the low frequency to high frequency ratio and SCL increased with time and heat (25 and 29°C). These results suggest that mental workload, especially implicit mental workload, increases in warmer environments, even if learning efficiency is facilitated. This study indicates integrated evidence for relationships among task performance, psychological state, and thermal environment by analyzing behavioral, subjective, and physiological indexes multidirectionally.

The photometric stereo method enables estimation of surface normals from images that have been captured using different but known lighting directions. The classical photometric stereo method requires at least three images to determine the normals in a given scene. However, this method cannot be applied to dynamic scenes because it is assumed that the scene remains static while the required images are captured. In this work, we present a dynamic photometric stereo method for estimation of the surface normals in a dynamic scene. We use a multi-tap complementary metal-oxide-semiconductor (CMOS) image sensor to capture the input images required for the proposed photometric stereo method. This image sensor can divide the electrons from the photodiode from a single pixel into the different taps of the exposures and can thus capture multiple images under different lighting conditions with almost identical timing. We implemented a camera lighting system and created a software application to enable estimation of the normal map in real time. We also evaluated the accuracy of the estimated surface normals and demonstrated that our proposed method can estimate the surface normals of dynamic scenes.

A thermal camera captures the temperature distribution of a scene as a thermal image. In thermal images, facial appearances of different people under different lighting conditions are similar. This is because facial temperature distribution is generally constant and not affected by lighting condition. This similarity in face appearances is advantageous for face detection. To detect faces in thermal images, cascade classifiers with Haar-like features are generally used. However, there are few studies exploring the local features for face detection in thermal images. In this paper, we introduce two approaches relying on local features for face detection in thermal images. First, we create new feature types by extending Multi-Block LBP. We consider a margin around the reference and the generally constant distribution of facial temperature. In this way, we make the features more robust to image noise and more effective for face detection in thermal images. Second, we propose an AdaBoost-based training method to get cascade classifiers with multiple types of local features. These feature types have different advantages. In this way we enhance the description power of local features. We did a hold-out validation experiment and a field experiment. In the hold-out validation experiment, we captured a dataset from 20 participants, comprising 14 males and 6 females. For each participant, we captured 420 images with 10 variations in camera distance, 21 poses, and 2 appearances (participant with/without glasses). We compared the performance of cascade classifiers trained by different sets of the features. The experiment results showed that the proposed approaches effectively improve the performance of face detection in thermal images. In the field experiment, we compared the face detection performance in realistic scenes using thermal and RGB images, and gave discussion based on the results.