Deep learning-based computer-aided diagnosis for parotid gland tumors on MRI

Area of Science:

• Radiology
• Artificial Intelligence
• Oncology

Background:

• Accurate differentiation of benign and malignant parotid tumors is crucial for appropriate patient management.
• Magnetic Resonance Imaging (MRI) is a key modality for parotid tumor evaluation.
• Deep learning (DL) offers potential for improving diagnostic accuracy in medical imaging.

Purpose of the Study:

• To assess the incremental clinical value of a DL-based computer-aided diagnosis (CAD) system.
• To evaluate the impact of DL-CAD on reader performance in MRI-based parotid tumor assessment.
• To determine if DL-CAD improves the differentiation between benign and malignant parotid tumors.

Main Methods:

• Development of a DL model (EfficientNet-based CNN) using MRI data from 170 patients.
• Rigorous five-fold cross-validation with patient-wise splits for model evaluation.
• Reader study involving four readers classifying 134 parotid tumors on MRI, with and without DL-CAD assistance.

Main Results:

• The DL model achieved high performance (accuracy: 0.85, AUC: 0.93).
• Readers using DL-CAD demonstrated significantly improved diagnostic accuracy (0.86 vs 0.76) and AUC (0.94 vs 0.82).
• Performance gains were consistent across experienced radiologists and residents, particularly for higher-grade and locally advanced tumors.

Conclusions:

• DL-based CAD enhances diagnostic performance in MRI-based parotid tumor evaluation.
• The system improves accuracy irrespective of reader experience.
• DL-CAD shows particular promise for managing high-grade and locally advanced parotid gland malignancies.