Development and validation of artificial intelligence to detect and diagnose liver lesions from ultrasound images

Abstract

Artificial intelligence (AI) using a convolutional neural network (CNN) has demonstrated promising performance in radiological analysis. We aimed to develop and validate a CNN for the detection and diagnosis of focal liver lesions (FLLs) from ultrasonography (USG) still images. The CNN was developed with a supervised training method using 40,397 retrospectively collected images from 3,487 patients, including 20,432 FLLs (hepatocellular carcinomas (HCCs), cysts, hemangiomas, focal fatty sparing, and focal fatty infiltration). AI performance was evaluated using an internal test set of 6,191 images with 845 FLLs, then externally validated using 18,922 images with 1,195 FLLs from two additional hospitals. The internal evaluation yielded an overall detection rate, diagnostic sensitivity and specificity of 87.0% (95%CI: 84.3–89.6), 83.9% (95%CI: 80.3–87.4), and 97.1% (95%CI: 96.5–97.7), respectively. The CNN also performed consistently well on external validation cohorts, with a detection rate, diagnostic sensitivity and specificity of 75.0% (95%CI: 71.7–78.3), 84.9% (95%CI: 81.6–88.2), and 97.1% (95%CI: 96.5–97.6), respectively. For diagnosis of HCC, the CNN yielded sensitivity, specificity, and negative predictive value (NPV) of 73.6% (95%CI: 64.3–82.8), 97.8% (95%CI: 96.7–98.9), and 96.5% (95%CI: 95.0–97.9) on the internal test set; and 81.5% (95%CI: 74.2–88.8), 94.4% (95%CI: 92.8–96.0), and 97.4% (95%CI: 96.2–98.5) on the external validation set, respectively. CNN detected and diagnosed common FLLs in USG images with excellent specificity and NPV for HCC. Further development of an AI system for real-time detection and characterization of FLLs in USG is warranted.

The feasibility to use artificial intelligence to aid detecting focal liver lesions in real-time ultrasound: a preliminary study based on videos

“…We acknowledged this limitation and addressed it by using our previous large dataset of image snapshots to pre-train the AI system in the first training step. This dataset contained ultrasound images collected from multiple institutions in Thailand and had a large variety of FLL characteristics and images from 17 ultrasound machine models 7 . We believe that the pre-training process using this dataset allowed the AI system to better handle variation in ultrasound exams.…”

“…Model A corresponded to the RetinaNet pre-trained on the MS-COCO dataset. We then further trained the Model A with a subset of our previously collected dataset 7 of 8510 ultrasound snapshot images of FLLs comprising 1114 HCCs, 2155 cysts, 1375 hemangiomas, 3303 FFSs, 563 FFIs and 17,047 snapshot images without FLL, resulting in Model B . Our previous study 7 used ultrasound snapshot images to train a RetinaNet model to detect FLLs.…”

“…We then further trained the Model A with a subset of our previously collected dataset 7 of 8510 ultrasound snapshot images of FLLs comprising 1114 HCCs, 2155 cysts, 1375 hemangiomas, 3303 FFSs, 563 FFIs and 17,047 snapshot images without FLL, resulting in Model B . Our previous study 7 used ultrasound snapshot images to train a RetinaNet model to detect FLLs. In Step 1 of the current study, we trained the Model A with the same training hyperparameters (Supplementary Method S1 ).…”

Artificial intelligence in liver ultrasound

“…Increasing the types of FLLs may confuse the diagnosis and reduce the accuracy. Similar to the previous study, a multicenter study estimating internal validation and external validation cohorts had a larger volume of training data and involved more varieties of FLLs, including cysts, HCC, hemangiomas, focal fatty infiltration and focal fatty sparing[ 39 ]. Although they obtained a lower sensitivity due to more types of diseases, the performance in the external validation cohorts was still satisfactory.…”

Performance of Deep Learning-Based Algorithm for Detection of Pediatric Intussusception on Abdominal Ultrasound Images

“…Sui et al [14] developed a deep-learning AI model (ThyNet) to differentiate between malignant tumours and benign thyroid nodules and aimed to investigate how ThyNet could help radiologists improve diagnostic performance and avoid unnecessary fine needle aspiration. Tiyarattanachai et al [15] developed a deep learning network for the detection and diagnosis of focal liver lesions from ultrasound images, AI model detected and diagnosed common focal liver lesions. For diagnosis of hepatocellular carcinomas, the AI model yielded sensitivity, specificity, and negative predictive value of 73.6%, 97.8%, and 96.5% on the internal validation set.…”