LAUSR

Artificial intelligence (AI) in medicine has been widely explored, Specifically, deep learning has been reported to gain excellent achievement on computed tomography (CT) and magnetic resonance (MR) images. Deep learning models can learn the most predictive features directly from raw image pixels and avoid the subjective impressions. At present, there are many AI algorithms, including feed-forward neural networks, regularized regression, gradient-boostered trees, and random forests are one of the most used machine learning methods. In this study, the authors sought to optimize random forest prediction performance using combinations of clinical and medical imaging data, and constructed a nomogram. This is the first article creating a new contrast-enhanced sonography (CEUS)-related nomogram for differentiating benign from malignant lesions in the periampullary area. The periampullary region is a complex area composed of three histologically and physiologically different anatomical structures, namely, ampulla of Vater, pancreatic duct, and common bile duct. Theoretically, clinical and biological behaviors of the diseases depend on their primary pathological structure. However, determining the disease nature arising in this narrow and complex area is usually not so easy in the clinical practice. The current gold standard before surgery is endoscopic retrograde cholangiopancreatography (ERCP) with biopsy (or endoscopic ultrasonography [EUS] with fine needle biopsy). However, this procedure is slightly invasive and not complication-free. To solve this diagnostic problem and establish the baselines for a standardized assessment of the disease, the authors tried to combine and analyze multiple non-invasive imaging examinations (gray-scale ultrasound [US], CT, magnetic resonance cholangiopancreatography [MRCP], and CEUS) and clinical data. At present, CEUS is frequently used for visualizing fine vascular structure of the disease, but, in this study, the authors worked out a way of improving its diagnostic ability; the patients were asked to drink oral agent CEUS to remove gastroduodenal gas and to demarcate clearly the periampullary area. In this study, the authors retrospectively analyzed the data of 124 patients with periampullary disease who underwent histological examination and they identified that patient's age, CBD diameter, enhancement pattern of arterial phase, wash-out pattern of venous phase, lesion size after CEUS as five most important indicator of disease characteristics, and, based on this, created a nomogram that predict the disease differentiation between benignity and malignancy. The authors should be praised for suggesting a new nomogram that showed nomogram to be highly diagnostic of periampullary diseases: the area under curves (AUC) and 95% confidence intervals (CI) for nomogram, magnetic resonance imaging (MRI) +MRCP+CEUS (combination of subjective diagnosis by the sonographer and radiologist), CEUS (subjective diagnosis by sonographer), MRI + MRCP (subjective diagnosis by radiologist) were 0.98, 0.91, 0.89, and 0.68 respectively. However, there are some points to consider about this study. First, the study subjects need to be reviewed. There were only 13 patients of benign disease group. Second, the diseases were roughly divided into two groups (benign and malignant). Strictly speaking, it is important to distinguish ampullary carcinoma (in the strict sense of the term) from bile duct carcinoma or pancreas carcinoma due to wide variation in clinical manifestations and outcomes. It is also important to divide the malignant lesions into well, moderate, and poorly differentiated ones. Third, in this study, the enhancement pattern in arterial phase was divided subjectively into “hyper-enhancement, hypo-enhancement, and iso-enhancement,” but this evaluation should be performed through time-intensitycurve analysis. Finally, this clinical study was a single-institution Received: 22 April 2022 Revised: 3 June 2022 Accepted: 6 June 2022