Artificial Intelligence for Contrast-Enhanced Ultrasound of the Liver: A Systematic Review

Abstract

Introduction: The research field of artificial intelligence (AI) in medicine and especially in gastroenterology is rapidly progressing with the first AI tools entering routine clinical practice, for example, in colorectal cancer screening. Contrast-enhanced ultrasound (CEUS) is a highly reliable, low-risk, and low-cost diagnostic modality for the examination of the liver. However, doctors need many years of training and experience to master this technique and, despite all efforts to standardize CEUS, it is often believed to contain significant interrater variability. As has been shown for endoscopy, AI holds promise to support examiners at all training levels in their decision-making and efficiency.

Methods: In this systematic review, we analyzed and compared original research studies applying AI methods to CEUS examinations of the liver published between January 2010 and February 2024. We performed a structured literature search on PubMed, Web of Science, and IEEE. Two independent reviewers screened the articles and subsequently extracted relevant methodological features, e.g., cohort size, validation process, machine learning algorithm used, and indicative performance measures from the included articles.

Results: We included 41 studies with most applying AI methods for classification tasks related to focal liver lesions. These included distinguishing benign versus malignant or classifying the entity itself, while a few studies tried to classify tumor grading, microvascular invasion status, or response to transcatheter arterial chemoembolization directly from CEUS. Some articles tried to segment or detect focal liver lesions, while others aimed to predict survival and recurrence after ablation. The majority (25/41) of studies used hand-picked and/or annotated images as data input to their models. We observed mostly good to high reported model performances with accuracies ranging between 58.6% and 98.9%, while noticing a general lack of external validation.

Conclusion: Even though multiple proof-of-concept studies for the application of AI methods to CEUS examinations of the liver exist and report high performance, more prospective, externally validated, and multicenter research is needed to bring such algorithms from desk to bedside.

Keywords: Artificial intelligence; Contrast-enhanced ultrasound; Hepatology; Liver; Sonography.

Fig. 1.

Common ML algorithms shown schematically. a SVM, data (2 classes – crosses, blue and plusses, orange) is moved from the initial single dimension to a higher one (dashed arrows) where it can be classified (gray dashed line). b RF, in a single decision tree, data (blue circle) passes through a series of questions (yellow circles) and, depending on the result, travels to classification (blue or orange circles). The RF is composed of multiple decision trees with different questions. c k-NNs, new data (blue circle) are classified according to the majority class of the k nearest data points. Orange dashed circle shows classification for k = 3 and blue for k = 5. d Logistic regression (LR), a curve (yellow line) is fitted to the training data that split it into 2 classes (blue and orange). New data are classified by applying the curve function and a threshold (gray dashed line). e NN, an NN is made up of many neurons (circles) arranged in layers, the input to each neuron comes from the layer before it. Data enter in the input layer (blue) and – in this case – a classification is output at the end (blue and orange).

Fig. 2.

Flowchart of the identification and selection process. CEUS, contrast-enhanced ultrasound; AI, artificial intelligence.

Fig. 3.

Flowchart showing different paths the data take in the reviewed studies. From the original data, hand-picked images and/or videos are selected with or without expert-drawn ROI. These then undergo feature extraction, and features can be either handcrafted (based on TIC, ultrasomics, optical flow, and/or textural analyses) or DL-based (e.g., using the features after passing an image through a CNN). Features are then fed into an algorithm that can be a classical method (SVM, RF, k-NN, or LR) or a DL method (simple NN, CNN, LSTM). Finally, the prediction is obtained for the task, broadly grouped into classification (of malignancy, lesion, or MVI) or segmentation (for which only SVMs and CNNs were used in the reviewed studies). ROI, region of interest; DL, deep learning; TIC, time-intensity curve; SVM, support vector machine; RF, random forest; LR, logistic regression; (C)NN, (convolutional) neural network; LSTM, long short-term memory network.

Fig. 4.

Comparison of methodology and performance of studies investigated. a Input data to models split into hand-picked images and videos, grouped by additional information used. b Cohort sizes in the reviewed studies. c Validation methods used; cross-validation is further split by the number of folds used. d Accuracies reported, grouped by method (classical or DL) and features used (hand-crafted or DL) for the tasks of malignancy (blue) and lesion (red) classification.