JMIR Publications

ABSTRACT

Background:

Artificial intelligence (AI) is widely utilized in the field of radiology; however, its development in pediatric imaging has been slow, with little to no application in pediatric emergency conditions. In the case of intussusception, an emergent pediatric abdominal condition, AI could facilitate early screening and triage using abdominal radiographs. However, further technical advancements and validation are needed to ensure its clinical utility.

Objective:

This study aimed to upgrade and externally validate an AI model for screening ileocolic intussusception on pediatric abdominal radiographs using multicenter data.

Methods:

This retrospective study included pediatric patients (≤5 years old) who underwent abdominal radiographs and US for suspected intussusception. Based on the preliminary study from Hospital A, the AI model was retrained using data from Hospital B and validated with external datasets from Hospital C and D. The upgraded AI’s diagnostic performance was compared with that of radiologists with different experience levels using sensitivity, specificity, and the area under the receiver operating characteristic curve (AUC) analysis.

Results:

Based on the previously developed AI model trained on 746 patients from Hospital A, additional 431 patients, including 143 cases of intussusception, from Hospital B were used for further training to develop an upgraded AI model. A total of 68 patients, including 19 cases of intussusception, from Hospital C and 90 patients, including 30 cases of intussusception, from Hospital D were included for external validation. The upgraded AI model demonstrated a sensitivity of 81.7% (95% CI 68.6-90%) and a specificity of 81.7% (95% CI 73.3-87.8%), outperforming overall radiologists without AI assistance (AUC 86.2% vs. 64%, p=0.014). AI assistance improved radiologists' specificity (93%, difference +15.9%, p<0.001) and AUC (79.2%, difference +15.2%, p=0.054), particularly benefiting less experienced radiologists.

Conclusions:

The upgraded AI model enhanced screening for ileocolic intussusception on pediatric abdominal radiographs, improving radiologists' specificity and accuracy, especially for those with less pediatric experience. A user-friendly software platform was introduced for broader validation and clinical use, underscoring AI's potential as a screening tool for triaging patients needing further US evaluation.