A Deep Learning Model for Classification of Parotid Neoplasms Based on Multimodal Magnetic Resonance Image Sequences

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Parotid neoplasms require accurate classification for effective treatment.
• Current diagnostic methods may have limitations in differentiating benign and malignant tumors.
• Multimodal magnetic resonance imaging (MRI) offers rich information for tumor characterization.

Purpose of the Study:

• To develop and validate a deep learning model for automatic parotid neoplasm classification using multimodal MRI.
• To enhance diagnostic accuracy and decision-making in clinical practice.
• To compare the model's performance against experienced radiologists.

Main Methods:

• A deep learning model was designed combining Resnet and Transformer networks.
• Multimodal MRI sequences from 266 patients were utilized.
• Multi-modality fusion strategies were optimized to improve model effectiveness.

Main Results:

• The fused multimodal deep learning model achieved an accuracy of 0.85 and an ROC AUC of 0.96.
• The model demonstrated high sensitivity (0.90) and specificity (0.84) in differentiating benign and malignant parotid neoplasms.
• Performance was superior to single-sequence MRI models and comparable to or better than radiologists for certain tumor types.

Conclusions:

• An accurate and efficient AI-based classification model for parotid neoplasms was developed using fused multimodal MRI.
• The multimodal approach significantly outperformed single-image or single-sequence inputs.
• This AI model shows promise for improving the diagnosis of parotid gland tumors.