Evaluation of liver tumors by using artificial intelligence in contrast-enhanced ultrasound

Objectives Contrast-enhanced ultrasound (CEUS) is an imagistic method commonly used in clinical practice to characterize focal liver lesions, but it has some limitations that can lead to misdiagnosis. We developed a deep learning system to detect and classify liver tumors based on standard and contrast-enhanced ultrasound, together with clinical data. Materials The dataset contained 49 patients with focal liver lesions, both benign and malignant, evaluated in the Department of Gastroenterology and Hepatology from the Emergency Clinical County Hospital of Craiova between February 2018 and December 2020. For dataset preparation, a region of interest (ROI) was drawn manually around the tumor borders by two experienced doctors. Time-intensity curve (TIC) was computed in order to describe the enhancement of contrast agent in all three vascular phases. The proposed system contained two artificial intelligence (AI) models. The first model was trained for image segmentation in order to extract the time-intensity curve, while the second deep learning model was a fully connected neural network which was trained on clinical data along with features extracted from the TIC. We assessed the sensitivity and specificity of the proposed system and compared it with the diagnostic performance of two clinicians, one of them blinded to the clinical informations and the patient's final diagnosis. Results For the blinded evaluation, we have obtained a sensitivity of 0,81 and a specificity of 1, while the clinician who had access to the clinical information obtained a sensitivity of 0,87 and a specificity of 1. The AI-based software obtained a sensitivity of 0,82 and specificity of 0,93. Conclusions Imagistic assessment based on AI has been introduced in the US field and can prevent human error and improve the accuracy of the diagnosis. However, depending on the samples from the dataset used, an AI model can have difficulties in classifying particular cases.