AI for childhood interstitial lung disease (chILD) – AI-based support tool for the diagnosis of persistent tachypnea of infancy (PTI)

Background: Childhood interstitial lung diseases (chILD) encompass a diverse range of rare pediatric conditions affecting the parenchyma, conducting airways, and alveolar spaces of the lungs. One subgroup is PTI, also known as neuroendocrine cell hyperplasia infancy (NEHI), which emerges predominantly in the first years of life. Diagnosis and differentiation of PTI from other chILD conditions often rely on CT scanning, revealing characteristic patterns of ground-glass opacities. However, identification of PTI poses challenges for non-specialized pediatric radiologists, particularly in centers lacking dedicated expertise.

Method: To train and validate our Machine Learning (ML) model, we will use CT images sourced from the chILD register [1], in which data from pediatric patients with chILD has been collected across Europe since 2012. We developed a preprocessing pipeline built on Orthanc [2] to construct a ML-ready dataset from this register. We included baseline axial CT series reconstructed with a lung kernel from patients under 3 years of age. Exploratory data analysis was used to assess quality of the dataset.

Results: The final dataset comprised 454 (819) patients (series). Among these, 160 patients were diagnosed with PTI. Our dataset is notably heterogeneous and unique in size, given its origin from 88 centers with various types of CT scanners, and the inherent variability in lung appearance among young children. Our next step involves training and evaluating a neural network using this dataset.

Publication History

Article published online:
28 February 2025

• Literatur

• 1 Jodogne S.. The Orthanc ecosystem for medical imaging. Journal of digital imaging 2018; 31 (03) 341-352