Code-Free Classification of Pediatric Pneumonia on Chest Radiographs Using Google Cloud Vertex AI AutoML: A...

Pneumonia is a clinically significant respiratory infection for which timely and accurate diagnosis is essential. Chest radiography is commonly used in the evaluation of suspected pneumonia; however, radiographic interpretation may be affected by reader experience, workload, image quality, and interobserver variability. This study evaluated whether a commercially available, code-free automated machine learning platform could distinguish pediatric chest radiographs labeled as normal from those labeled as pneumonia. A publicly available pediatric chest radiograph dataset was uploaded to Google Cloud Vertex AI AutoML Image Classification. Of 5,863 source images, 5,824 were successfully imported, including 1,579 normal radiographs and 4,245 pneumonia radiographs. Vertex AI randomly divided the imported images into training, validation, and testing subsets using an 80-10-10 allocation. A single-label binary image-classification model was trained using an eight node-hour budget, with all remaining user-configurable settings left at platform defaults. The model achieved an average precision of 0.999. At a confidence threshold of 0.50, platform-reported aggregate precision and recall were both 98.6%. The row-normalized confusion matrix demonstrated correct classification of 97% of normal radiographs and 99% of pneumonia radiographs. Model training was completed in two hours and eight minutes without custom neural-network programming or dedicated local high-performance computing hardware. These findings demonstrate the technical feasibility of developing a high-performing medical image-classification model using a code-free, cloud-based AutoML platform. However, the results represent internal benchmark performance on a single curated pediatric dataset. Image-level randomization, unavailable patient identifiers, class imbalance, proprietary model development, and the absence of external validation limit clinical generalizability. Independent multi-institutional evaluation is required before clinical implementation can be considered.