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

Abstract

Background and Aims. Diagnosing pediatric intussusception from ultrasound images can be a difficult task in many primary care hospitals that lack experienced radiologists. To address this challenge, this study developed an artificial intelligence- (AI-) based system for automatic detection of “concentric circles” signs on ultrasound images, thereby improving the efficiency and accuracy of pediatric intussusception diagnosis. Methods. A total of 440 cases (373 pediatric intussusception and 67 normal cases) were retrospectively collected from Children’s Hospital affiliated to Zhejiang University School of Medicine from January 2020 to December 2020. An improved Faster RCNN deep learning framework was used to detect “concentric circle” signs. Finally, independent validation set was used to evaluate the performance of the developed AI tool. Results. The data of pediatric intussusception were divided into a training set and validation set according to the ratio of 8 : 2, with training set (298 pediatric intussusception) and validation set (75 pediatric intussusception and 67 normal cases). In the “concentric circle” detection model, the detection rate, recall, specificity, and F 1 score assessed by the validation set were 92.8%, 95.0%, 92.2%, and 86.4%, respectively. Pediatric intussusception was classified by “concentric circle” signs, and the accuracy, recall, specificity, and F 1 score were 93.0%, 92.0%, 94.1%, and 93.2% on the validation set, respectively. Conclusion. The model established in this paper can realize the automatic detection of “concentric circle” signs in the ultrasound images of abdominal intussusception in children; the AI tool can improve the diagnosis speed of pediatric intussusception. It is necessary to further develop an artificial intelligence system for real-time detection of “concentric circles” in ultrasound images for the judgment of children with intussusception.

Role of artificial intelligence in oncologic emergencies: a narrative review

“…Indeed, it is not surprising that the sensitivity of the algorithm was demonstrated higher than that of the radiologists, but it is particularly interesting that the specificity was almost superimposable (0.92 vs. 0.96) [ 54 ]. Another study by Li et al [ 55 ] assessed the performance of DL for the automatic detection of “concentric circle” signs in US images.…”

“…Many studies have been carried out to investigate the potential role of AI to increase sensitivity [52] in the diagnosis of intestinal intussusception. DL-based algorithms have been developed for the detection of ileocolic intussusception on abdominal radiography and abdominal US [53][54][55]. Kim et al [53] demonstrated that a DL may increase radiologists' diagnostic performance in detecting intestinal intussusception.…”