A Non-enhanced CT Dataset for Differential Diagnosis of Pediatric Extracranial Germ Cell Tumors.
The paper reports a standardized dataset of 642 non-enhanced CT scans spanning six pediatric extracranial germ cell tumor subtypes, with expert-annotated tumor masks and preliminary radiomics-based machine-learning experiments for differential diagnosis.
Open original publication →What the AI sees
The paper reports a standardized dataset of 642 non-enhanced CT scans spanning six pediatric extracranial germ cell tumor subtypes, with expert-annotated tumor masks and preliminary radiomics-based machine-learning experiments for differential diagnosis.
Research significance
The record supports the dataset's potential for developing subtype-classification tools; it is an untested inference that improved preoperative classification could eventually guide treatment selection or reduce inappropriate interventions, as no therapeutic decisions or patient outcomes were evaluated.
Source abstract
Extracranial germ cell tumors (EGCTs) are rare pediatric neoplasms characterized by significant histological heterogeneity and variability in clinical behavior. This highlights the necessity for precise preoperative differential diagnosis. While computed tomography (CT) provides essential imaging information for differentiation, conventional visual assessments are often limited due to overlapping radiological features. Although artificial intelligence (AI) has demonstrated potential in enhancing diagnostic accuracy in medical imaging, its application to EGCTs has been constrained by the scarcity of publicly available imaging datasets. To address this gap, we present the CT Pediatric EGCTs Diagnosis (CT-PEGCT-Diag) dataset, which consists of 642 non-enhanced CT scans representing six distinct histological subtypes: mature teratoma, immature teratoma, yolk sac tumor, mixed germ cell tumor, dysgerminoma, and embryonal carcinoma. The dataset encompasses standardized preprocessing protocols, rigorous quality control measures, and expert-annotated tumor masks. Preliminary experiments employing radiomics-based machine learning models demonstrate the dataset's utility in aiding the development of diagnostic tools. The CT-PEGCT-Diag dataset is designed to facilitate the validation of AI models and advance research into imaging biomarkers for the subtype classification of pediatric EGCTs.