Search

Liver biopsy is the reference standard test to differentiate between non-alcoholic steatohepatitis (NASH) and simple steatosis (SS) in non-alcoholic fatty liver disease (NAFLD), but noninvasive diagnostics are warranted. The diagnostic accuracy in NASH using MR imaging modality have not yet been clearly identified. This study was assessed the accuracy of magnetic resonance imaging (MRI) method for diagnosing NASH. Data were extracted from research articles obtained after a literature search from multiple electronic databases. Random-effects meta-analyses were performed to obtain overall effect size of the area under the receiver operating characteristic(ROC) curve, sensitivity, specificity, likelihood ratios(LR), diagnostic odds ratio(DOR) of MRI method in detecting histopathologically-proven SS(or non-NASH) and NASH. Seven studies were analyzed 485 patients, which included 207 SS and 278 NASH. The pooled sensitivity was 87.4% (95% CI, 76.4–95.3) and specificity was 74.3% (95% CI, 62.4–84.6). Pooled positive LR was 2.59 (95% CI, 1.96–3.42) and negative LR was 0.17 (95% CI, 0.07–0.38). DOR was 21.57 (95% CI, 7.27–63.99). The area under the curve of summary ROC was 0.89. Our meta-analysis shows that the MRI-based diagnostic methods are valuable additions in detecting NASH.

The Observational Medical Outcomes Partnership-Common Data Model (OMOP-CDM) used in distributed research networks has low coverage of clinical data and does not reflect the latest trends of precision medicine. Radiology data have great merits to visualize and identify the lesions in specific diseases. However, radiology data should be shared to obtain the sufficient scale and diversity required to provide strong evidence for improving patient care. Our study was to develop a web-based management system for radiology-CDM (R-CDM), as an extension of the OMOP-CDM, and to assess the feasibility of an R-CDM dataset for application of radiological image data in AI learning. This study standardized a cirrhosis of liver (LC) R-CDM dataset consisting of CT data (LC 40,575 images vs. non-LC 33,565 images). With use of modified AI learning algorithm, the diagnostic accuracy was 0.99292 (error rate = 0.00708), and its sensitivity was 0.99469 for LC and specificity was 0.99115 for non-LC. We developed a web-based management system for searching and downloading standardized R-CDM dataset and constructed a liver cirrhosis R-CDM dataset for clinical practice. Our management system and LC dataset would be helpful for multicenter study and AI learning research.

Recently, there have been considerable efforts to develop non-invasive image-based method for diagnosis, staging, and monitoring of chronic liver disease (CLD). This study developed a deep learning (DL) algorithm for discriminating the CLD using non-contrast abdominal CT images. This study enrolled 499 patients with CLD and 122 healthy controls. The main structure of DL algorithm used GoogLeNet-V3 and several modules fine-tuned for this study. In the test data sets, the DL algorithm had a discrimination accuracy of 99.4% (loss is about 0.7%) and an AUROC of 0.998, for diagnosis between normal controls and CLD patients. In the validation test, we achieve good validation result as follows: specificity (for healthy controls) 0.98292 (error rate: 0.01708) and sensitivity (for CLD) 0.99469 (error rate: 0.00531), respectively. Our deep learning algorithm would be useful for accurate discrimination in CLD from the abdominal CT images without the use of contrast agent. Further study is needed to diagnose the disease severity within CLD patient group for clinical application.

Recent advances in artificial intelligence have been used for a variety of clinical interests, such as lesion detection, segmentation, classification, and differentiation in medical images, and a great deal of research is underway. A good algorithm is also needed to obtain meaningful artificial intelligence research results, but a correctly created data set is also important. To provide these data sets, we proposed the radiology-common data model (R-CDM), as an expansion of OMOP-CDM. With the use of R-CDM, this study created a data set to distinguish urinary diseases for 873 patients and showed significance using an artificial intelligence algorithm that modified GoogleNet, a deep-learning algorithm based on CNN. In this experiment, 99% accuracy and AUROC 0.9 were obtained. External validation was performed using urinary stones data sets for 200 patients at other hospitals to increase confidence in the results of deep learning. In external validation, the results of 90% sensitivity, 97% specificity, and AUROC 0.984 confirm that the data sets and algorithms have very high efficiency. The study will help researchers who perform artificial intelligence through medical imaging.

In sarcopenic obesity, the importance of evaluating muscle and fat mass is unquestionable. There exist diverse quantification methods for assessing muscle and fat mass by imaging techniques; thus these methods must be standardized for clinical practice. This study developed a quantification software for the body composition imaging using abdominal magnetic resonance (MR) images and compared the difference between sarcopenic obesity and healthy controls for clinical application. Thirty patients with sarcopenic obesity and 30 healthy controls participated. The quantification software was developed based on an ImageJ multiplatform and the processing steps are as follows: execution, setting, confirmation, and extraction. The variation in the muscle area (MA), subcutaneous fat area (SA), and visceral fat area (VA) was analyzed with an independent two sample T-test. There were significant differences in SA (p