Clinical Scorecard: Evaluating the Diagnostic Utility of Advanced CT Radiomics and Deep Learning for Distinguishing Pediatric Peripheral Neuroblastoma from Ganglioneuroblastoma

At a Glance

Key Highlights

• Peripheral NB is the most common extracranial solid tumor in children with diagnosis typically before age five.
• Conventional contrast-enhanced CT has limited ability to differentiate NB from GNB due to overlapping clinical and imaging features.
• Radiomics and deep learning models based on contrast-enhanced CT, clinical data, and biochemical markers (NSE, SF) improve differential diagnosis accuracy.

Guideline-Based Recommendations

Diagnosis

• Use contrast-enhanced CT with arterial and venous phase imaging for initial tumor evaluation.
• Incorporate radiomic feature extraction and deep learning-based segmentation to improve lesion characterization.
• Combine imaging findings with clinical data (age, sex) and biochemical indicators (neuron-specific enolase, serum ferritin) for comprehensive assessment.

Management

• Accurate histological subtyping via imaging supports pretreatment risk stratification and personalized therapy planning.
• Use imaging and biochemical markers to guide selection, dosage, and duration of chemotherapy and radiotherapy.

Monitoring & Follow-up

• Apply deep learning models for consistent tumor segmentation to monitor treatment response and progression.
• Regularly assess biochemical markers alongside imaging to evaluate tumor behavior.

Risks

• Serum ferritin levels may be elevated due to inflammation or cancer, reducing specificity.
• Manual segmentation of tumors is time-consuming and subject to inter-observer variability; automated deep learning segmentation mitigates this.

Patient & Prescribing Data

Pediatric patients newly diagnosed with peripheral NB or GNB without prior treatment

Imaging-based risk stratification enables tailored chemotherapy and radiotherapy regimens, potentially improving outcomes.

Clinical Best Practices

• Ensure high-quality contrast-enhanced CT imaging with arterial and venous phases for optimal radiomic analysis.
• Utilize automated 3D-UNet deep learning segmentation models to standardize ROI delineation and reduce variability.
• Integrate clinical, biochemical, and advanced imaging data for comprehensive diagnostic evaluation.
• Exclude patients with prior treatment or poor image quality to maintain diagnostic accuracy.
• Sedate pediatric patients as needed to ensure image acquisition quality.

References

• Neuroblastoma overview and tumor spectrum
• Contrast-enhanced CT in peripheral neuroblastoma diagnosis
• Radiomics and deep learning applications in pediatric tumors
• Neuron-specific enolase and serum ferritin in neuroblastoma risk stratification
• 3D-UNet architecture for medical image segmentation