Introduction & Objective: Early detection of diabetic peripheral neuropathy (DPN) in patients with diabetes mellitus can prevent amputations. However, diagnosing DPN with nerve conduction studies (NCS) is costly and requires highly trained specialists. We present a novel thermal imaging device to detect DPN via foot temperature recovery after cold provocation, which may provide enhanced DPN screening by lowering costs, facilitating nonspecialist testing, and conserving time and resources. Methods: NCS and thermal imaging recovery data were collected from 176 patients. Plantar feet were exposed to a 2°C cooling pad for 2 min. Recovery temperature was recorded for a point on the 1st metatarsal on the right foot with a thermal video camera for 90 s. Subjects were grouped into 21 controls, 110 diabetic subjects with negative NCS testing, and 45 diabetic subjects with positive NCS testing. Results: The figure shows the mean temperature warming following cooling with 95% confidence intervals for the 3 groups. Each curve is statistically different from the others. Conclusion: Thermal imaging is a useful test to diagnose early DPN. It may also detect small fiber diabetic neuropathy before NCS show positive. Because NCS tests are expensive and only available in advanced clinical facilities, thermal imaging will provide greater access to DPN screening and early intervention for many diabetic patients. Disclosure S. Bhatia: None. P. Soliz: None. S. Saint-Lot: None. J. Wigdahl: None. E. Duran-Valdez: None. D.S. Schade: None. Funding National Institutes of Health (R44DK104578)

Diabetic peripheral neuropathy (DPN) is a complication of diabetes that causes severe foot pain and frequently leads to amputation. Early detection is critical to saving patients from foot ulcers and amputation. This paper presents an automatic system to identify thermal biomarkers associated with DPN. Research Design and Methods: 141 subjects diagnosed with diabetes mellitus (DM) were enrolled in the study. Subjects were categorized as those with DM but without DPN, those with DPN based on a positive nerve conduction study, and those without DPN. A support vector machine (SVM) was used to classify the subjects having DPN based on thermal parameters related to temperature recovery after a cold provocation. Results: The classifier produced a sensitivity/specificity of 0.78/0.89 in identifying DPN. Conclusions: The SVM classifier can identify patients with the large fiber form of DPN. A different reference standard had to be used to detect small fiber neuropathy.

Cerebral malaria (CM) is a fatal syndrome found commonly in children less than 5 years old in Sub-saharan Africa and Asia. The retinal signs associated with CM are known as malarial retinopathy (MR), and they include highly specific retinal lesions such as whitening and hemorrhages. Detecting these lesions allows the detection of CM with high specificity. Up to 23% of CM, patients are over-diagnosed due to the presence of clinical symptoms also related to pneumonia, meningitis, or others. Therefore, patients go untreated for these pathologies, resulting in death or neurological disability. It is essential to have a low-cost and high-specificity diagnostic technique for CM detection, for which We developed a method based on transfer learning (TL). Models pre-trained with TL select the good quality retinal images, which are fed into another TL model to detect CM. This approach shows a 96% specificity with low-cost retinal cameras.

Purpose: To demonstrate thermal features of the foot in diabetic subjects without known peripheral neuropathy. Background: Assessment of thermoregulation of the foot in patients diagnosed with diabetes identifies those predisposed to foot ulcer and amputation. Methods: The clinical study of the thermoregulation of the microvascular processes of the plantar foot included 100 subjects with diabetes ages 35 to 76. Subjects underwent a complete clinical exam and nerve conduction study. Thermal imaging with a portable device followed a standard procedure for cooling of the plantar foot, followed by video imaging of the plantar surface during re-warming. The reference standard for neuropathy was a nerve conduction study. Features included: Initial temperature; temperature after cold provocation; and recovery temperatures. Results: The figure shows the 6 standard regions studied. The temperature after cooling for the normal subject was 16 deg C with a recovery temperature of 6 to 8 deg C after 400 s. For the subject with neuropathy the temperature after cooling dropped only to 18 deg C and recovered 2 to 4 deg C. Using these features, the sensitivity was 85% sensitivity and 70% specificity. Conclusion: Quantification of thermoregulation of the foot provides a new and non-invasive assessment of diabetic peripheral neuropathy in patients with diabetes. Identification of these at-risk patients may prevent amputation. Disclosure P.Soliz: None. E.Duran-valdez: None. A.P.Bancroft: None. G.Tang: None. D.S.Schade: Consultant; AbbVie Inc. J.Vijayamohanan: None. Funding NIH 2R44DK104578-04A1

Primary care clinics lack an accurate, easy-to-use examination tool to screen patients with diabetes for diabetic peripheral neuropathy (DPN). This study demonstrates a new approach for the early detection of DPN in patients diagnosed with diabetes. Machine learning techniques were used to analyze thermal videos of the plantar foot after cold provocation and to detect biomarkers associated with abnormal function of the foot's microvascular network. A machine learning technique, support vector machines (SVM), is compared to a direct method of analyzing specific biomarkers present in the thermal recovery of the plantar foot after a cold provocation. These analytical models demonstrated a clinically relevant level of sensitivity and specificity.

Allowme to make these observations on the article by Lee et al. (1) in this issue of Diabetes Care. The topic of artificial intelligence (AI) techniques, deep learning in particular, for interpretation of medical images for use in clinical environments is timely. We congratulate the authors for their contribution to a highly relevant topic. The strengths of the article are many. For example, the results show that some implementations of AI for diabetic retinopathy (DR) screening are highly effective and can have a significant impact on increasing access to DR screening in a cost-effective manner. There are five notable observations that I wish to make. First, the referral criteria were somewhat academic and not consistentwith the goal of evaluating the algorithms in a realistic environment. The case for setting the referral criterion at level 1 (any DR) or 2 (mild nonproliferative DR [NPDR]) should be justified in light of realistic environments. We can visualize environments where these criteria would have clinical value. However, as given in Table 1, 72% of referrals present with only mild NPDR, which by referring the 3,321 mild NPDR cases of the 3,861 total would overwhelm the ophthalmology clinic. The U.S. Food and Drug Administration has permitted marketing of twoAI systems for DR screening based on the detection threshold of more than mild DR, i.e., level 2 (moderate or greater NPDR). The referral would be to an eye care professional. Those patients with less than moderate NPDR would be evaluated in 12 months. Second, most studies report performance by “case,” onwhether a patient is to be referred or not. Reporting performance by the individual image, sometimes as many as 13 for one patient (based on 311,604 images and 23,727 cases), makes comparison with other studies difficult. This manner of reporting results clouds the referral process and the results of the algorithms’ effectiveness. For example, were patients’ referrals counted multiple times if multiple scans for the same person were positive? Third, access to both mydriatic and nonmydriatic data presented an opportunity to shed light on the effect of dilation on the performance of the algorithms. With the data available from the two Veterans Affairs hospitals, one primarily using nonmydriatic imaging and the other dilating all patients, an opportunity was missed by not analyzing these two categories of mydriasis. The impact of mydriasis would be an important consideration when implementing a DR screening program. Fourth, the study had a binary grade for image quality, i.e., the image was gradable or was not. Was it possible that one patient had multiple acceptable quality images that would have been sufficient to determine a referral/no referral result, even though the same patient had one or many more ungradable images? Fifth, no results were given for diabetic macular edema or clinically significant macular edema. Granted, macular edema is best detected with OCT, yet for screening purposes, many direct eye exams and algorithms use surrogate markers, e.g., hard exudates on or near the fovea. This is a critical omission.

Cerebral Malaria (CM) is a severe neurological syndrome of malaria mainly found in children and is associated with highly specific retinal lesions. The manifestation of these indications of CM in the retina is called malarial retinopathy (MR). All patients showing clinical signs of CM are commonly diagnosed and treated accordingly; however, 23% of them are misdiagnosed as they suffer from another infection with identical clinical symptoms. Due to these underlying symptoms, the false positive cases may go untreated and could result in death of the patients. A diagnostic test is needed that is highly specific in order to reduce false positives. The purpose of this study to demonstrate a technique based on a transfer learning technique using images from three different retinal cameras to identify the hemorrhages and whitening lesions in the retina which can accurately identify the patients with MR. The MR detection model gives a specificity of 100% and a sensitivity of 90% with an AUC of 0.98. The algorithm demonstrates the potential of accurate MR detection with a low-cost retinal camera.

In this work, we demonstrate a novel approach to assessing the risk of Diabetic Peripheral Neuropathy (DPN) using only the retinal images of the patients. Our methodology consists of convolutional neural network feature extraction, dimensionality reduction and feature selection with random projections, combination of image features to case-level representations, and the training and testing of a support vector machine classifier. Using clinical diagnosis as ground truth for DPN, we achieve an overall accuracy of 89% on a held-out test set, with sensitivity reaching 78% and specificity reaching 95%.

Purpose: Studies have shown that individuals with retinal blood vessel abnormalities often present with other biomarkers associated with diabetic peripheral neuropathy (DPN). Existing software requires extensive manual intervention to measure structures in the retinal vasculature (e.g., arterial caliber, artery/vein ratios, fractals, etc.) for the purpose of studying the relationship of these structures to DPN. The goal of this study was to demonstrate an artificial intelligence (AI) software system for determining DPN risk based on retinal vascular abnormalities. Our approach circumvented the need to explicitly measure vessel characteristics by using AI as applied to a state-of-the-art methodology known as “convolutional neural networks” (CNN) to characterize the retinal vascular structures. CNNs have been demonstrated to be effective in a number of medical applications. This study is the first time CNNs have been used to extract retinal vascular features for DPN risk assessment. Methods: From a database of 25,000 cases previously labeled as healthy (controls), unaffected diabetes (DM), or DPN, we have identified 331 cases that fit the requirements for this study. Controls (N=103) had monofilament and vibration tests to confirm normal peripheral nerve function. DM patients (N=163) were classified based on their medical records and monofilament and vibration tests. DPN patients (N=65) were confirmed by clinical examination and their medical record. With these cases, an AI classifier was trained to identify individuals with DPN biomarkers based on spatial features extracted from the retinal abnormalities. Results: 80% of cohort data were used for training and 20% was used for testing. The resulting AI tool demonstrated 95% specificity and 78% sensitivity when using the clinical exam as the reference standard. Conclusions: AI-based CNN analysis of retinal vascular features show promise for identifying subjects with DPN and reduced human manipulation and measurement of tedious vessel geometries. Disclosure M.R. Burge: None. P. Soliz: None. J. Benson: None. V. Joshi: None. Funding National Institutes of Health (DK104578)

Worldwide, glaucoma and age-related macular degeneration (AMD) cause 12.3% and 8.7% of the cases of blindness and/or vision loss, respectively. According to a 5-year study of Medicare beneficiaries, patients who undergo a regular eye screening, experience less decline of vision than those who had less-frequent examinations. A computer-based screening of retinopathies can be highly cost-effective and efficient; however, most auto-screening software address only one eye disease, limiting their clinical utility and cost-effectiveness. Therefore, we propose a computer-based retinopathy screening system for detection of AMD and glaucoma by integrating information from retinal fundus images and clinical data. First, the retinal image analysis algorithms were developed using Transfer Learning approach to determine presence or absence of the eye disease. The clinical data was then utilized to improve disease detection performance where the image-analysis based algorithms provided sub-optimal classification. The results for binary detection (present/absent) of AMD and Glaucoma were compared with the ground truth provided by a certified retinal reader. We applied the proposed method to a dataset of 304 retinal images with AMD, 299 retinal images with Glaucoma, and 2,341 control retinal images. The algorithms demonstrated sensitivity/specificity of 100%/99.5% for detection of any AMD, 82%/70% for detection of referable AMD, and 75%/81% for detection of referable Glaucoma. The automated detection results agree well with the ground truth suggesting its potential in screening for AMD and Glaucoma.

Adequate image quality is a necessary component to any retinal screening program whether the cases are to be read by a human reader or processed by an artificial intelligence system (AI). The need for expanded screening for retinal diseases has led to the adoption of low-cost, portable cameras that are ideal for reaching large underserved populations. However, the low-cost cameras generally require a higher level of operator skill to produce high quality images in comparison with more expensive table-top retinal cameras. This study, conducted at thirteen clinics in Monterrey, Mexico, compares unreadable rates between a table-top retinal camera (Canon CR2-AF) and a low-cost portable camera (Volk Pictor Plus) before and after implementation of automatic image quality assessment software, Image Quality Analyzer (IQA). The software determines if an image is of adequate quality to be read by a human or AI system; what the image quality issues are; and tips for fixing the issues. The process is performed in real time. Results show a significant decrease in unreadable cases (9% to 0%) for the Pictor Plus after IQA implementation bringing the percent of rejected cases in line with the table-top camera (3% to 5%).

The purpose of this study was to identify biomarkers that are indicative of diabetic peripheral neuropathy (DPN). From a cohort of 21 DPN and 18 controls, optic disc- and macula-centered images were taken. Sixty vascular parameters were measured using computer-assisted software. Statistically significant differences in vascular parameters between subjects with DPN and controls showed that DPN subjects could be differentiated from controls using retinal biomarkers. Statistical significance (p < 0.05) was obtained for 13 retinal features, including fractal dimension, vein width, number of 1st branching vessels, and vessel tortuosity. The reduced fractal dimensions in DPN patients indicate a deterioration of the retinal vascular architecture. The reduced total number of vessels, as well as the total number of venules, contribute to the lower fractals in the DPN patients and supports the findings of a sparser retinal vascular network for DPN subjects. A difference in standard deviation for vessel width may suggest a greater loss of vascular tone and regulation in DPN subjects. A difference in the number of 1st branching vessels in venules was observed, which may indicate that the DPN subjects had fewer numbers of these first bifurcations. This feature would also contribute to the lower fractal dimension as well as a sparser retinal vascular network. Similarly, the statistical difference in the angle of the first daughter vessel suggests that the retinal vascular architecture has deviated from its optimal flow and/or function due to the disease process. The increased difference in curvature tortuosity in the arterioles in DPN correlates with chronic hypertensive retinopathy. We found statistically significant differences in 13 out of 60 retinal vascular features (22%) between control and DPN subjects. As such, the retina reflects the widespread vascular structural abnormalities occurring in diabetic patients with DPN. Disclosure S.C. Nemeth: None. M.R. Burge: None. J. Wigdahl: None. J.C. Carmichael: None. X. Guo: None. V. Joshi: None. P. Soliz: Stock/Shareholder; Self; VisionQuest Biomedical LLC. Funding National Institutes of Health (R43DK104578-01)