Deep convolutional neural network analysis may be useful to clinical microbiology laboratories, which increasingly use automation and technical advancements to optimize testing efficiency and improve performance, according to recent research (Clin Chem 2022; doi: 10.1093/clinchem/hvab270).
Deep convolutional neural networks are a type of machine-learning algorithm used for feature extraction, object identification, image classification, and tracking. The networks combine convolution and processing components. The convolution component uses feature maps to learn different image representation, while the processing component consists of a fully connected multilayer perceptron—a type of neural network unit—that processes learned features and makes a prediction.
Labs’ decisions to classify urine cultures as insignificant growth, contamination, or growth consistent with a urinary tract infection is subjective. The researchers identified and quantitated subjective labeling of urine cultures using a deep-learning approach, BacterioSight, to improve harmonization and quality of urine culture interpretation in a clinical microbiology laboratory.
Researchers used BacterioSight on routine clinical urine cultures from two large institutions. BacterioSight displayed performance on par with standard-of-care-trained technologist interpretations. BacterioSight’s accuracy ranged from 97% when compared wtih standard-of-care involving a single technologist to 100% when compared with the gold standard: consensus by a group of technologists.
Training and testing performed within the same institutions performed well, giving area under the curve (AUC) of 0.98 or greater for negative and positive plates. In contrast, cross-testing on images trained on one institution’s images and tested on images from the other institution showed decreased performance, with AUC of 0.90 or greater for negative and positive plates.
The researchers said that their study can be a roadmap for how BacterioSight or other deep-learning prototypes can screen for microbial growth, flag difficult cases for review by multiple people, or verify a subset of cultures with high confidence. The results also highlight variability of image interpretation by technologists within and across institutions.
New eGFR Equation Drives CKD Disease Classification Changes
New equations used to calculate estimated glomerular filtration rate (eGFR) without consideration of race would move millions of U.S. adults into a new kidney function classification, mostly in the moderate stages of chronic kidney disease (CKD) (JAMA Netw Open 2022; doi: 10.1001/jamanetworkopen.2022. 0460).
In the cross-sectional study, researchers used data from the U.S. National Health and Nutrition Examination Survey (NHANES 2011–2018) to compare estimates of CKD severity using the old and new eGFR formulas, with a focus on disease stage and the CKD-related complications anemia, acidosis, hyperphosphatemia, and hypertension.
Using the new formula, about 5.5 million adults were reclassified, researchers found. About 1 million Black people with CKD moved into a more severe category, and about 4.5 million people who are not Black moved into a less severe category. The new formula did not substantially change CKD-related complication prevalence estimates.
The researchers noted that their research was limited by use of single laboratory measurements. Limited numbers in some subgroups prevented precise estimates, especially of complications. Limited subgroup numbers also prevented separate analysis of smaller racial and ethnic groups. Findings could drive changes to CKD diagnosis and treatmen for many patients, the researchers said.
Study Suggests Lower Troponin Cutoff for Early COVID-19 Cardiac Damage
A high-sensitivity troponin 1 (hs‐cTnI) level of 5 ng/L or more may be a manifestation of early cardiac damage in patients with nonsevere COVID-19, according to a recent paper (Sci Rep 2022; doi: 10.1038/s41598-022-06378-2).
To explore the manifestations of cardiac damage at presentation in nonsevere patients with COVID‐19, the researchers grouped 113 patients with nonsevere COVID‐19 according to the length of time between symptom onset to hospital admission. That time span was a week or less for group 1, one to two weeks for group 2, more than two to three weeks for group 3, and more than three weeks for group 4. The researchers compared clinical, cardiovascular, and radiological data on hospital admission across
the four groups.
Group 2 patients in the second week after symptom onset had the highest levels of cardiac biomarkers. The proportion of patients who had a hs‐cTnI of 5 ng/L or more in group 2 was 85.71%, compared with 37.04% for group 1, 51.85% for group 3, and 25% for group 4. Group 2 patients also had the highest levels of C-reactive protein (CRP) and lactate dehydrogenase.
Compared with patients with hs‐cTnI under 5 ng/L, those with hs‐cTnI of 5 or mor ng/L had lower lymphocyte count and higher CRP. Patients with hs‐cTnI ≥ 5 ng/L had a higher incidence of bilateral pneumonia and longer hospital length of stay.
Researchers say their study is likely the first to demonstrate the value of a cutoff lower than the 99th percentile of hs-cTnI to identify early cardiac damage in nonsevere patients with COVID-19. Results are similar to those of earlier studies that show the level of hs-cTnI increased significantly from 10 to 13 days after symptom onset in COVID-19 patients and that the cardiac biomarkers were highly related to lymphocyte count and CRP. These finding suggest that cardiac damage from COVID-19 is related to viral response and hyperinflammation.