A preliminary study of commercially available general-purpose chest radiography artificial intelligence-based software for detecting airspace opacity lesions in COVID-19 patients

Purpose To validate commercially available general-purpose artificial intelligence (AI)-based software for detecting airspace opacity in chest radiographs (CXRs) of COVID-19 patients. Materials and Methods We used the ieee8023-covid-chestxray-dataset to validate commercial AI software capable of detecting "Nodule/Mass" and "Airspace opacity" as regions of interest with probability scores. From this dataset, we excluded computed tomography images and CXR images taken using an anteroposterior spine view and analyzed CXR images tagged with "Pneumonia/Viral/COVID-19" and "no findings". A radiologist then reviewed the images and rated them on a 3-point opacity score for the presence of airspace opacity. The maximum probability score of airspace opacity for each image was calculated using this software. The difference in each maximum probability for each opacity score was evaluated using Wilcoxon's rank sum test. The threshold of the probability score was determined by receiver operator characteristic curve analysis for the presence or absence of COVID-19, and the true positive rate (TPR) and false positive rate (FPR) were determined for the individual and overall opacity scores. Results Images from 342 patients with COVID-19 and 15 normal images were included. Opacity scores of 1, 2, and 3 were observed in 44, 70, and 243 images, respectively, of which 33 (75%), 66 (94.2%), and 243 (100%), respectively, were from COVID-19 patients. The overall TPR and FPR were 0.82 and 0.13, respectively, at an area under the curve of 0.88 and a threshold of 0.06, while the FPR for opacity score 1 was 0.18 and the TPR for score 3 was 0.97. Conclusion Using a public database containing CXR images of COVID-19 patients, commercial AI software was shown to be able to detect airspace opacity in severe pneumonia. Summary Commercially available AI software was capable of detecting airspace opacity in CXR images of COVID-19 patients in a public database.

Validation of the Southend Giant Cell Arteritis Probability Score in a Scottish single centre fast-track pathway

Objective To externally validate the Southend GCA Probability Score (GCAPS) in patients attending a GCA Fast-Track Pathway (GCA FTP) in NHS Lanarkshire. Methods Consecutive GCA FTP patients between November 2018 and December 2020 underwent GCAPS assessment as part of routine care. GCA diagnoses were supported by USS +/- TAB and confirmed at 6 months. Percentages of patients with GCA according to GCAPS risk group, performance of total GCAPS in distinguishing GCA/non-GCA final diagnoses, and test characteristics using different GCAPS binary cut-offs, were assessed. Associations between individual GCAPS components and GCA, and the value of USS and TAB in the diagnostic process, were also explored. Results 44/129 patients were diagnosed with GCA, including 0/41 GCAPS low risk patients (GCAPS <9), 3/40 medium risk (GCAPS 9–12), and 41/48 high risk (GCAPS >12). Overall performance of GCAPS in distinguishing GCA/non-GCA was excellent [ROC AUC 0.976 (95% CI 0.954–0.999)]. GCAPS cut-off ≥10 had 100.0% sensitivity and 67.1% specificity for GCA. GCAPS cut-off ≥13 had highest accuracy (91.5%), with 93.2% sensitivity and 90.6% specificity. Several individual GCAPS components were associated with GCA. Sensitivity of USS increased by ascending GCAPS risk group (nil, 33.3%, 90.2% respectively). TAB was diagnostically useful in cases where USS was inconclusive. Conclusion This is the first published study describing application of GCAPS outside the specialist centre where it was developed. Performance of GCAPS as a risk stratification tool was excellent. GCAPS may have additional value for screening GCA FTP referrals and guiding empirical glucocorticoid treatment.

Color doppler ultrasound and the giant cell arteritis probability score for the diagnosis of giant cell arteritis: a Canadian single center experience

Objectives To compare accuracy of colour doppler ultrasonography (CDUS) and temporal artery biopsy (TAB) to establish the final diagnosis of GCA and to determine how the giant cell arteritis probability score (GCAPS) performs as a risk stratification tool. Methods Descriptive statistics were performed on a retrospective cohort of patients referred to our vasculitis referral center between July 1st, 2017 and October 1st, 2020 for suspected GCA. CDUS, TAB, center-specific TAB (vasculitis center vs.s referring hospitals) and GCAPS were compared against the final diagnosis of GCA as determined by a GCA expert; CDUS was also compared with TAB results. Results Data from 198 patients were included: 60 patients with GCA and 138 patients without GCA. Sixty-two patients had a TAB. Using the final diagnosis by a GCA expert as a reference, sensitivity (Se), specificity (Sp), positive predictive value (PPV) and negative predictive value (NPV) were 93.3%, 98.5%, 96.6%, 97.1% for CDUS; and 69.2%, 100%, 100%, 81.8% for TAB. The false negative rate was 6.7% for CDUS and 30.8% for TAB. False negative TAB mostly occurred when performed in referring hospitals (57.1%) as opposed to our vasculitis center (21.1%). With a cut-off at 9.5 points, Se for GCAPS was 98.3% while Sp was 74.3%. Conclusion CDUS of the temporal and axillary arteries showed a high sensitivity and specificity and helped to diagnose GCA in patients with negative TAB. We validated that GCAPS is a useful clinical tool with a score < 9.5 making the diagnosis of GCA improbable.

A Road Paved with Good Intentions: A Platelet Count-Based Alert to Facilitate Diagnosis of Heparin-Induced Thrombocytopenia

Background: Heparin-induced thrombocytopenia (HIT) is a rare disorder with potential to cause significant morbidity and mortality. Early identification and initiation of non-heparin anticoagulation can prevent devastating thrombotic events. However, over-testing is common and can lead to result misinterpretation, unnecessary heparin avoidance, and increased cost. When there is concern for HIT, guidelines from the American Society of Hematology recommend calculation of the 4Ts score to determine the need for laboratory testing. The Choosing Wisely® initiative recommends against laboratory testing in patients with a low probability score of ≤3. In patients with an intermediate or high probability score (≥4), screening with enzyme-linked immunosorbent assay (ELISA) is performed first. If positive, the diagnosis of HIT is confirmed with a functional assay, commonly the serotonin release assay (SRA). Methods: In an effort to increase recognition of HIT, providers at a large academic medical center received a non-interruptive alert in the electronic medical record (EMR) on all patients in whom the platelet count declined by ≥50% starting in Aug 2017. We performed a retrospective evaluation of 1) the number of alerts and 2) all ELISA results obtained with or without an alert, over a 90-day period (Dec 2019 to March 2020). A 4Ts score was calculated by chart review by the first author in real-time as the alert was sent (blinded to ELISA and SRA results). Among those patients with multiple alerts or test orders, the first instance was used for analysis. Demographic and clinical characteristics were reported using frequencies and percentages, means (standard deviation, SD), and medians (interquartile range, IQR). Patients with alerts and ELISA testing ordered were compared with 2 groups: 1) patients with alerts but no ELISA ordered; 2) patients with no alerts but ELISA ordered. Comparisons were performed using chi squared tests, Fisher's exact tests, t-tests and Wilcoxon rank-sum tests as appropriate. Results: In the 90-day observation period, 302 alerts were fired in 270 patients (Figure 1). Thirty alerts (28 patients, 10%) were generated for patients admitted for organ donation or post-stem cell transplantation, for whom platelet count decline was expected. Excluding these patients, there were 272 alerts in 242 patients (approximately 3 alerts per day in a 1,157-bed hospital). Of patients with alerts, 22 (8%) had a platelet count inaccuracy (i.e. platelets clump or another reason) and 18 (7%) did not receive heparin prior to platelet decline, for a cumulative total of 40 (15%) inappropriate alerts. In patients with an alert, the ELISA was ordered more frequently for those with a lower platelet nadir (77x10 9/L vs. 115x10 9/L, p<0.0001) or in those with a thrombotic event (11 patients (17%) vs. 6 patients (4%), p=0.0021) (Table 1). Those without an ELISA ordered were more likely to have a low 4Ts score (23 patients (36%) vs. 81 patients (58%), p<0.0001). In addition to 71 patients with an alert, an ELISA was also ordered for 67 patients without an alert (n=138) (Figure 1). Close to half of ELISA-tested patients had a low 4Ts score (n=51, 46%) (Figure 2). In patients with an alert and ELISA not ordered, 18 (27%) had an intermediate or high 4Ts score. Seven patients were diagnosed with HIT based on a positive SRA, 6 with an alert and 1 without. The alert demonstrated a sensitivity of 86% (95% CI, 59.8-100%) and specificity of 50% (95% CI, 41.8-58.9%) with a positive predictive value of 0.0845 (95% CI, 0.0198-0.1492) and negative predictive value of 0.9851 (95% CI, 0.9560-1.0000). Conclusion: An EMR alert based on platelet count decline had multiple shortcomings including frequent inappropriate firings and a lack of guidance on appropriate indications for testing. This evaluation of institutional testing practices indicates frequent use and misinterpretation of ELISA discordant with evidence-based guidelines. Although prompt diagnosis of HIT is important, alternative strategies for identification of at-risk patients and communication of recommended actions to providers should be considered. Because the 4Ts score includes variables difficult to automate in the EMR, our institution is exploring electronic consultation and real-time expert provider access to overcome these limitations. Figure 1 Figure 1. Disclosures No relevant conflicts of interest to declare.

Offline Handwritten Character Recognition using Convolutional Neural Network

This paper focuses on the task of recognizing handwritten Hindi characters using a Convolutional Neural Network (CNN) based. The recognized characters can then be stored digitally in the computer or used for other purposes. The dataset used is obtained from the UC Irvine Machine Learning Repository which contains 92,000 images divided into training (80%) and test set (20%). It contains different forms of handwritten Devanagari characters written by different individuals which can be used to train and test handwritten text recognizers. It contains four CNN layers followed by three fully connected layers for recognition. Grayscale handwritten character images are used as input. Filters are applied on the images to extract different features at each layer. This is done by the Convolution operation. The two other main operations involved are Pooling and Flattening. The output of the CNN layers is fed to the fully connected layers. Finally, the chance or probability score of each character is determined and the character with the highest probability score is shown as the output. A recognition accuracy of 98.94% is obtained. Similar models exist for the purpose, but the proposed model achieved a better performance and accuracy than some of the earlier models. Keywords: Devanagari characters, Convolutional Neural Networks, Image Processing

Ultrasonography based imaging criterion to ascertain pancreatic enlargement in pediatric acute pancreatitis

Imaging studies have shown enlargement of pancreatic parts in children diagnosed with acute pancreatitis. The purpose here is to develop imaging based diagnostic evaluation criterion for acute pancreatitis in children. This study included 62 children of acute pancreatitis in the age range of 0.33 to 13 years, as reported in a single hospital center (1994-2019). A study was planned including 1116 normal healthy children in the age range of 0.16 to 18 years for pancreatic evaluation during 2016-17. Ultrasonography based measurement of three pancreatic parts were obtained for each individual in disease and normal groups. Age-adjusted receiver operating characteristics curve analysis was performed on each pancreatic part independently to derive respective cut-offs using a training set. These cut-offs were further referred to dichotomize the measurement data for each individual and was subjected to multiple logistic regression with presence/absence of acute pancreatitis as dependent variable. A probability score and accordingly the cut-off were obtained indicating a collective impression of enlargement of pancreas in disease condition independently for males and females. On test data, the accuracy of age-adjusted cut-offs for three parts was near 80% for males, while it ranged between 81-85% for females. ROC analysis of probability score resulted into threshold value of 0.024 for males and 0.044 for females, with sensitivity of 94.11% and 90.91% respectively. The classification accuracy of score derived for males and females was nearly same (83%). The extent of enlargement of pancreas in acute pancreatitis in children can be determined using the MLR method along with hypoechogenicity.