Improved Training Efficiency for Retinopathy of Prematurity Deep Learning Models Using Comparison versus Class Labels.

Purpose: To compare the efficacy and efficiency of training neural networks for medical image classification using comparison labels indicating relative disease severity versus diagnostic class labels from a retinopathy of prematurity (ROP) image dataset. Design: Evaluation of diagnostic test or technology. Participants: Deep learning neural networks trained on expert-labeled wide-angle retinal images obtained from patients undergoing diagnostic ROP examinations obtained as part of the Imaging and Informatics in ROP (i-ROP) cohort study. Methods: Neural networks were trained with either class or comparison labels indicating plus disease severity in ROP retinal fundus images from 2 datasets. After training and validation, all networks underwent evaluation using a separate test dataset in 1 of 2 binary classification tasks: normal versus abnormal or plus versus nonplus. Main Outcome Measures: Area under the receiver operating characteristic curve (AUC) values were measured to assess network performance. Results: Given the same number of labels, neural networks learned more efficiently by comparison, generating significantly higher AUCs in both classification tasks across both datasets. Similarly, given the same number of images, comparison learning developed networks with significantly higher AUCs across both classification tasks in 1 of 2 datasets. The difference in efficiency and accuracy between models trained on either label type decreased as the size of the training set increased. Conclusions: Comparison labels individually are more informative and more abundant per sample than class labels. These findings indicate a potential means of overcoming the common obstacle of data variability and scarcity when training neural networks for medical image classification tasks.

Plus Disease in Retinopathy of Prematurity: Convolutional Neural Network Performance Using a Combined Neural Network and Feature Extraction Approach.

Purpose: Retinopathy of prematurity (ROP), a leading cause of childhood blindness, is diagnosed by clinical ophthalmoscopic examinations or reading retinal images. Plus disease, defined as abnormal tortuosity and dilation of the posterior retinal blood vessels, is the most important feature to determine treatment-requiring ROP. We aimed to create a complete, publicly available and feature-extraction-based pipeline, I-ROP ASSIST, that achieves convolutional neural network (CNN)-like performance when diagnosing plus disease from retinal images. Methods: We developed two datasets containing 100 and 5512 posterior retinal images, respectively. After segmenting retinal vessels, we detected the vessel centerlines. Then, we extracted features relevant to ROP, including tortuosity and dilation measures, and used these features in the classifiers including logistic regression, support vector machine and neural networks to assess a severity score for the input. We tested our system with fivefold cross-validation and calculated the area under the curve (AUC) metric for each classifier and dataset. Results: For predicting plus versus not-plus categories, we achieved 99% and 94% AUC on the first and second datasets, respectively. For predicting pre-plus or worse versus normal categories, we achieved 99% and 88% AUC on the first and second datasets, respectively. The CNN method achieved 98% and 94% for predicting two categories on the second dataset. Conclusions: Our system combining automatic retinal vessel segmentation, tracing, feature extraction and classification is able to diagnose plus disease in ROP with CNN-like performance. Translational Relevance: The high performance of I-ROP ASSIST suggests potential applications in automated and objective diagnosis of plus disease.

Siamese neural networks for continuous disease severity evaluation and change detection in medical imaging.

Using medical images to evaluate disease severity and change over time is a routine and important task in clinical decision making. Grading systems are often used, but are unreliable as domain experts disagree on disease severity category thresholds. These discrete categories also do not reflect the underlying continuous spectrum of disease severity. To address these issues, we developed a convolutional Siamese neural network approach to evaluate disease severity at single time points and change between longitudinal patient visits on a continuous spectrum. We demonstrate this in two medical imaging domains: retinopathy of prematurity (ROP) in retinal photographs and osteoarthritis in knee radiographs. Our patient cohorts consist of 4861 images from 870 patients in the Imaging and Informatics in Retinopathy of Prematurity (i-ROP) cohort study and 10,012 images from 3021 patients in the Multicenter Osteoarthritis Study (MOST), both of which feature longitudinal imaging data. Multiple expert clinician raters ranked 100 retinal images and 100 knee radiographs from excluded test sets for severity of ROP and osteoarthritis, respectively. The Siamese neural network output for each image in comparison to a pool of normal reference images correlates with disease severity rank (ρ = 0.87 for ROP and ρ = 0.89 for osteoarthritis), both within and between the clinical grading categories. Thus, this output can represent the continuous spectrum of disease severity at any single time point. The difference in these outputs can be used to show change over time. Alternatively, paired images from the same patient at two time points can be directly compared using the Siamese neural network, resulting in an additional continuous measure of change between images. Importantly, our approach does not require manual localization of the pathology of interest and requires only a binary label for training (same versus different). The location of disease and site of change detected by the algorithm can be visualized using an occlusion sensitivity map-based approach. For a longitudinal binary change detection task, our Siamese neural networks achieve test set receiving operator characteristic area under the curves (AUCs) of up to 0.90 in evaluating ROP or knee osteoarthritis change, depending on the change detection strategy. The overall performance on this binary task is similar compared to a conventional convolutional deep-neural network trained for multi-class classification. Our results demonstrate that convolutional Siamese neural networks can be a powerful tool for evaluating the continuous spectrum of disease severity and change in medical imaging.

A Quantitative Severity Scale for Retinopathy of Prematurity Using Deep Learning to Monitor Disease Regression After Treatment.

Importance: Retinopathy of prematurity (ROP) is a leading cause of childhood blindness worldwide, but treatment failure and disease recurrence are important causes of adverse outcomes in patients with treatment-requiring ROP (TR-ROP). Objectives: To apply an automated ROP vascular severity score obtained using a deep learning algorithm and to assess its utility for objectively monitoring ROP regression after treatment. Design, Setting, and Participants: This retrospective cohort study used data from the Imaging and Informatics in ROP consortium, which comprises 9 tertiary referral centers in North America that screen high volumes of at-risk infants for ROP. Images of 5255 clinical eye examinations from 871 infants performed between July 2011 and December 2016 were assessed for eligibility in the present study. The disease course was assessed with time across the numerous examinations for patients with TR-ROP. Infants born prematurely meeting screening criteria for ROP who developed TR-ROP and who had images captured within 4 weeks before and after treatment as well as at the time of treatment were included. Main Outcomes and Measures: The primary outcome was mean (SD) ROP vascular severity score before, at time of, and after treatment. A deep learning classifier was used to assign a continuous ROP vascular severity score, which ranged from 1 (normal) to 9 (most severe), at each examination. A secondary outcome was the difference in ROP vascular severity score among eyes treated with laser or the vascular endothelial growth factor antagonist bevacizumab. Differences between groups for both outcomes were assessed using unpaired 2-tailed t tests with Bonferroni correction. Results: Of 5255 examined eyes, 91 developed TR-ROP, of which 46 eyes met the inclusion criteria based on the available images. The mean (SD) birth weight of those patients was 653 (185) g, with a mean (SD) gestational age of 24.9 (1.3) weeks. The mean (SD) ROP vascular severity scores significantly increased 2 weeks prior to treatment (4.19 [1.75]), peaked at treatment (7.43 [1.89]), and decreased for at least 2 weeks after treatment (4.00 [1.88]) (all P < .001). Eyes requiring retreatment with laser had higher ROP vascular severity scores at the time of initial treatment compared with eyes receiving a single treatment (P < .001). Conclusions and Relevance: This quantitative ROP vascular severity score appears to consistently reflect clinical disease progression and posttreatment regression in eyes with TR-ROP. These study results may have implications for the monitoring of patients with ROP for treatment failure and disease recurrence and for determining the appropriate level of disease severity for primary treatment in eyes with aggressive disease.

Monitoring Disease Progression With a Quantitative Severity Scale for Retinopathy of Prematurity Using Deep Learning.

Importance: Retinopathy of prematurity (ROP) is a leading cause of childhood blindness worldwide, but clinical diagnosis is subjective and qualitative. Objective: To describe a quantitative ROP severity score derived using a deep learning algorithm designed to evaluate plus disease and to assess its utility for objectively monitoring ROP progression. Design, Setting, and Participants: This retrospective cohort study included images from 5255 clinical examinations of 871 premature infants who met the ROP screening criteria of the Imaging and Informatics in ROP (i-ROP) Consortium, which comprises 9 tertiary care centers in North America, from July 1, 2011, to December 31, 2016. Data analysis was performed from July 2017 to May 2018. Exposure: A deep learning algorithm was used to assign a continuous ROP vascular severity score from 1 (most normal) to 9 (most severe) at each examination based on a single posterior photograph compared with a reference standard diagnosis (RSD) simplified into 4 categories: no ROP, mild ROP, type 2 ROP or pre-plus disease, or type 1 ROP. Disease course was assessed longitudinally across multiple examinations for all patients. Main Outcomes and Measures: Mean ROP vascular severity score progression over time compared with the RSD. Results: A total of 5255 clinical examinations from 871 infants (mean [SD] gestational age, 27.0 [2.0] weeks; 493 [56.6%] male; mean [SD] birth weight, 949 [271] g) were analyzed. The median severity scores for each category were as follows: 1.1 (interquartile range [IQR], 1.0-1.5) (no ROP), 1.5 (IQR, 1.1-3.4) (mild ROP), 4.6 (IQR, 2.4-5.3) (type 2 and pre-plus), and 7.5 (IQR, 5.0-8.7) (treatment-requiring ROP) (P < .001). When the long-term differences in the median severity scores across time between the eyes progressing to treatment and those who did not eventually require treatment were compared, the median score was higher in the treatment group by 0.06 at 30 to 32 weeks, 0.75 at 32 to 34 weeks, 3.56 at 34 to 36 weeks, 3.71 at 36 to 38 weeks, and 3.24 at 38 to 40 weeks postmenstrual age (P < .001 for all comparisons). Conclusions and Relevance: The findings suggest that the proposed ROP vascular severity score is associated with category of disease at a given point in time and clinical progression of ROP in premature infants. Automated image analysis may be used to quantify clinical disease progression and identify infants at high risk for eventually developing treatment-requiring ROP. This finding has implications for quality and delivery of ROP care and for future approaches to disease classification.

Classification and comparison via neural networks.

We consider learning from comparison labels generated as follows: given two samples in a dataset, a labeler produces a label indicating their relative order. Such comparison labels scale quadratically with the dataset size; most importantly, in practice, they often exhibit lower variance compared to class labels. We propose a new neural network architecture based on siamese networks to incorporate both class and comparison labels in the same training pipeline, using Bradley-Terry and Thurstone loss functions. Our architecture leads to a significant improvement in predicting both class and comparison labels, increasing classification AUC by as much as 35% and comparison AUC by as much as 6% on several real-life datasets. We further show that, by incorporating comparisons, training from few samples becomes possible: a deep neural network of 5.9 million parameters trained on 80 images attains a 0.92 AUC when incorporating comparisons.

Predicting aggression to others in youth with autism using a wearable biosensor.

Unpredictable and potentially dangerous aggressive behavior by youth with Autism Spectrum Disorder (ASD) can isolate them from foundational educational, social, and familial activities, thereby markedly exacerbating morbidity and costs associated with ASD. This study investigates whether preceding physiological and motion data measured by a wrist-worn biosensor can predict aggression to others by youth with ASD. We recorded peripheral physiological (cardiovascular and electrodermal activity) and motion (accelerometry) signals from a biosensor worn by 20 youth with ASD (ages 6-17 years, 75% male, 85% minimally verbal) during 69 independent naturalistic observation sessions with concurrent behavioral coding in a specialized inpatient psychiatry unit. We developed prediction models based on ridge-regularized logistic regression. Our results suggest that aggression to others can be predicted 1 min before it occurs using 3 min of prior biosensor data with an average area under the curve of 0.71 for a global model and 0.84 for person-dependent models. The biosensor was well tolerated, we obtained useable data in all cases, and no users withdrew from the study. Relatively high predictive accuracy was achieved using antecedent physiological and motion data. Larger trials are needed to further establish an ideal ratio of measurement density to predictive accuracy and reliability. These findings lay the groundwork for the future development of precursor behavior analysis and just-in-time adaptive intervention systems to prevent or mitigate the emergence, occurrence, and impact of aggression in ASD. Autism Res 2019, 12: 1286-1296. © 2019 International Society for Autism Research, Wiley Periodicals, Inc. LAY SUMMARY: Unpredictable aggression can create a barrier to accessing community, therapeutic, medical, and educational services. The present study evaluated whether data from a wearable biosensor can be used to predict aggression to others by youth with autism spectrum disorder (ASD). Results demonstrate that aggression to others can be predicted 1 min before it occurs with high accuracy, laying the groundwork for the future development of preemptive behavioral interventions and just-in-time adaptive intervention systems to prevent or mitigate the emergence, occurrence, and impact of aggression to others in ASD.

Evaluation of a deep learning image assessment system for detecting severe retinopathy of prematurity.

BACKGROUND: Prior work has demonstrated the near-perfect accuracy of a deep learning retinal image analysis system for diagnosing plus disease in retinopathy of prematurity (ROP). Here we assess the screening potential of this scoring system by determining its ability to detect all components of ROP diagnosis. METHODS: Clinical examination and fundus photography were performed at seven participating centres. A deep learning system was trained to detect plus disease, generating a quantitative assessment of retinal vascular abnormality (the i-ROP plus score) on a 1-9 scale. Overall ROP disease category was established using a consensus reference standard diagnosis combining clinical and image-based diagnosis. Experts then ranked ordered a second data set of 100 posterior images according to overall ROP severity. RESULTS: 4861 examinations from 870 infants were analysed. 155 examinations (3%) had a reference standard diagnosis of type 1 ROP. The i-ROP deep learning (DL) vascular severity score had an area under the receiver operating curve of 0.960 for detecting type 1 ROP. Establishing a threshold i-ROP DL score of 3 conferred 94% sensitivity, 79% specificity, 13% positive predictive value and 99.7% negative predictive value for type 1 ROP. There was strong correlation between expert rank ordering of overall ROP severity and the i-ROP DL vascular severity score (Spearman correlation coefficient=0.93; p<0.0001). CONCLUSION: The i-ROP DL system accurately identifies diagnostic categories and overall disease severity in an automated fashion, after being trained only on posterior pole vascular morphology. These data provide proof of concept that a deep learning screening platform could improve objectivity of ROP diagnosis and accessibility of screening.

Predicting Imminent Aggression Onset in Minimally-Verbal Youth with Autism Spectrum Disorder Using Preceding Physiological Signals.

We test the hypothesis that changes in preceding physiological arousal can be used to predict imminent aggression proximally before it occurs in youth with autism spectrum disorder (ASD) who are minimally verbal (MV-ASD). We evaluate this hypothesis through statistical analyses performed on physiological biosensor data wirelessly recorded from 20 MV-ASD youth over 69 independent naturalistic observations in a hospital inpatient unit. Using ridge-regularized logistic regression, results demonstrate that, on average, our models are able to predict the onset of aggression 1 minute before it occurs using 3 minutes of prior data with a 0.71 AUC for global, and a 0.84 AUC for person-dependent models.

Time-Series Prediction of Proximal Aggression Onset in Minimally-Verbal Youth with Autism Spectrum Disorder Using Physiological Biosignals.

It has been suggested that changes in physiological arousal precede potentially dangerous aggressive behavior in youth with autism spectrum disorder (ASD) who are minimally verbal (MV-ASD). The current work tests this hypothesis through time-series analyses on biosignals acquired prior to proximal aggression onset. We implement ridge-regularized logistic regression models on physiological biosensor data wirelessly recorded from 15 MV-ASD youth over 64 independent naturalistic observations in a hospital inpatient unit. Our results demonstrate proof-of-concept, feasibility, and incipient validity predicting aggression onset 1 minute before it occurs using global, person-dependent, and hybrid classifier models.

Automated Diagnosis of Plus Disease in Retinopathy of Prematurity Using Deep Convolutional Neural Networks.

Importance: Retinopathy of prematurity (ROP) is a leading cause of childhood blindness worldwide. The decision to treat is primarily based on the presence of plus disease, defined as dilation and tortuosity of retinal vessels. However, clinical diagnosis of plus disease is highly subjective and variable. Objective: To implement and validate an algorithm based on deep learning to automatically diagnose plus disease from retinal photographs. Design, Setting, and Participants: A deep convolutional neural network was trained using a data set of 5511 retinal photographs. Each image was previously assigned a reference standard diagnosis (RSD) based on consensus of image grading by 3 experts and clinical diagnosis by 1 expert (ie, normal, pre-plus disease, or plus disease). The algorithm was evaluated by 5-fold cross-validation and tested on an independent set of 100 images. Images were collected from 8 academic institutions participating in the Imaging and Informatics in ROP (i-ROP) cohort study. The deep learning algorithm was tested against 8 ROP experts, each of whom had more than 10 years of clinical experience and more than 5 peer-reviewed publications about ROP. Data were collected from July 2011 to December 2016. Data were analyzed from December 2016 to September 2017. Exposures: A deep learning algorithm trained on retinal photographs. Main Outcomes and Measures: Receiver operating characteristic analysis was performed to evaluate performance of the algorithm against the RSD. Quadratic-weighted κ coefficients were calculated for ternary classification (ie, normal, pre-plus disease, and plus disease) to measure agreement with the RSD and 8 independent experts. Results: Of the 5511 included retinal photographs, 4535 (82.3%) were graded as normal, 805 (14.6%) as pre-plus disease, and 172 (3.1%) as plus disease, based on the RSD. Mean (SD) area under the receiver operating characteristic curve statistics were 0.94 (0.01) for the diagnosis of normal (vs pre-plus disease or plus disease) and 0.98 (0.01) for the diagnosis of plus disease (vs normal or pre-plus disease). For diagnosis of plus disease in an independent test set of 100 retinal images, the algorithm achieved a sensitivity of 93% with 94% specificity. For detection of pre-plus disease or worse, the sensitivity and specificity were 100% and 94%, respectively. On the same test set, the algorithm achieved a quadratic-weighted κ coefficient of 0.92 compared with the RSD, outperforming 6 of 8 ROP experts. Conclusions and Relevance: This fully automated algorithm diagnosed plus disease in ROP with comparable or better accuracy than human experts. This has potential applications in disease detection, monitoring, and prognosis in infants at risk of ROP.