Developing a Novel Deep Learning Model for Pediatric Chest Radiograph Analysis

Pediatric thoracic diseases pose significant challenges to healthcare systems worldwide, particularly in regions with limited access to specialized medical resources. Lower respiratory infections, in particular, contribute significantly to childhood morbidity and mortality. This research aims to address the diagnostic challenges associated with pediatric thoracic diseases by harnessing the power of artificial intelligence. The study utilizes the PediCXR dataset, composed of pediatric chest radiographs, and employs a Faster R-CNN model pretrained on ImageNet data for lesion detection and classification. The model is trained to identify abnormal lesions indicative of various thoracic diseases, with a focus on achieving high accuracy and minimizing loss. Experimental results demonstrate the model's capability to achieve an accuracy exceeding 90% and a loss under 0.1, meeting predefined objectives.