A Deep Learning–Based Approach to Reduce Rescan and Recall Rates in Clinical MRI Examinations

Abstract

BACKGROUND AND PURPOSE: MR imaging rescans and recalls can create large hospital revenue loss. The purpose of this study was to develop a fast, automated method for assessing rescan need in motion-corrupted brain series. MATERIALS AND METHODS:A deep learning-based approach was developed, outputting a probability for a series to be clinically useful. Comparison of this per-series probability with a threshold, which can depend on scan indication and reading radiologist, determines whether a series needs to be rescanned. The deep learning classification performance was compared with that of 4 technologists and 5 radiologists in 49 test series with low and moderate motion artifacts. These series were assumed to be scanned for 2 scan indications: screening for multiple sclerosis and stroke. RESULTS:The image-quality rating was found to be scan indication-and reading radiologist-dependent. Of the 49 test datasets, technologists created a mean ratio of rescans/recalls of (4.7 Ϯ 5.1)/(9.5 Ϯ 6.8) for MS and (8.6 Ϯ 7.7)/(1.6 Ϯ 1.9) for stroke. With thresholds adapted for scan indication and reading radiologist, deep learning created a rescan/recall ratio of (7.3 Ϯ 2.2)/(3.2 Ϯ 2.5) for MS, and (3.6 Ϯ 1.5)/(2.8 Ϯ 1.6) for stroke. Due to the large variability in the technologists' assessments, it was only the decrease in the recall rate for MS, for which the deep learning algorithm was trained, that was statistically significant (P ϭ .03). CONCLUSIONS:Fast, automated deep learning-based image-quality rating can decrease rescan and recall rates, while rendering them technologist-independent. It was estimated that decreasing rescans and recalls from the technologists' values to the values of deep learning could save hospitals $24,000/scanner/year. ABBREVIATIONS: CB ϭ clinically bad; CG ϭ clinically good; CNN ϭ convolutional neural network; DL ϭ deep learning; D0 -D5 ϭ radiologists; IQ ϭ image quality; R0 ϭ radiologist; ROC ϭ receiver operating characteristic; T1-T4 ϭ MR imaging technologists

Economic evaluations of big data analytics for clinical decision-making: a scoping review

“…Eight studies presented predictive analytics for 7 different types of imaging data (CT, MRI, Chest radiographs, digital cervical smears, mammographies, digital photographs, and ventilation-perfusion lung scans). 72–79 The number of full economic evaluations 73 , 75–77 , 79 and studies performed after development 73–76 , 78 were both higher than the first group of papers that used medical history data. Four studies measured effects in (quality-adjusted) life-years, 73 , 75–77 and more than half of the studies included implementation costs of analytics.…”

“… 73–77 The number of studies that found the analytics could lead to cost-savings was once again quite high (63%). 72–74 , 78 , 79 Just like the studies that used medical history data, authors of studies in this category emphasized the need for further validation prior to implementation. However, several studies also emphasized the balance between the requirements of the technologies (ie, test sensitivity) and potential health benefits and cost-savings.…”

“…Most papers were included because their volume might be large enough to be considered big data (ie, N > 100 000, imaging data) and studies that used monitoring data were included because of the potential speed with which the data is collected (velocity). Eight of these papers used medical history data, 32 , 37 , 40 , 44 , 45 , 50 , 60 , 68 5 used imaging data, 72–74 , 76 , 78 4 used monitoring data, 80 , 82 , 83 , 87 and 3 used omics data. 89 , 94 , 44 , 96 Most were partial economic evaluations (n = 15) and 12 were performed after development.…”

Emerging deep learning techniques using magnetic resonance imaging data applied in multiple sclerosis and clinical isolated syndrome patients (Review)

“…Sreekumari et al ( 68 ) developed an automated method for assessing the need of rescan, in motion corrupted brain scans. Authors developed a CNN with 7 convolutional layers, 4 max pooling layers, and 3 batch normalization layers that computed the probability for a MR series to be clinically useful, while by combining this probability with a scan dependent and radiologist defined threshold, they determined whether a series need to be rescanned.…”

Opportunities for Understanding MS Mechanisms and Progression With MRI Using Large-Scale Data Sharing and Artificial Intelligence

“… 20 , 21 The majority of studies to date used classic machine learning techniques such as support vector machines or random forests, where features have to be defined and extracted from the data a priori, while more recent studies also use deep learning methods, allowing automated detection of relevant features in the data. Deep learning has now been used not only to segment WM lesions 22 - 24 or their enhancing subset, 17 but also to quantify lesion changes, 25 , 26 detect the central vein sign, 27 classify different lesion types based on diffusion basis spectrum imaging, 28 predict gadolinium enhancement from other image types, 29 perform MRI-based diagnosis, 30 , 31 segment and analyze nonlesion structures, 32 , 33 analyze myelin water fraction 34 or quantitative susceptibility mapping data, 35 synthesize absent image types, 36 perform automatic QC, 37 improve image quality, 38 or correct intensity differences between scanners 39 (additional references in eAppendix 1).…”