Intelligent meningioma grading based on medical features

Area of Science:

• Neuro-oncology
• Medical imaging analysis
• Machine learning in medicine

Background:

• Meningiomas are common primary brain tumors, with high-grade types being aggressive and recurrent.
• Accurate pathological grading is vital for treatment, follow-up, and prognosis.
• Existing radiomics and deep learning methods have limitations in reliability due to pixel-level features or ambiguous image interpretations.

Purpose of the Study:

• To validate the effectiveness of integrating medical features with deep neural networks for enhanced meningioma grading accuracy and reliability.

Main Methods:

• A novel SNN-Tran model was developed for meningioma grading.
• The model analyzes diverse medical features: tumor volume, peritumoral edema, dural tail sign, tumor location, edema-to-tumor ratio, age, and gender.
• This approach captures complex feature interactions for improved prediction reliability.

Main Results:

• The SNN-Tran model achieved high performance: 0.875 accuracy, 0.886 sensitivity, 0.847 specificity, and 0.872 AUC.
• The proposed method outperformed traditional deep learning, radiomics, and state-of-the-art approaches.

Conclusions:

• Combining medical features with the SNN-Tran model significantly enhances meningioma grading accuracy and reliability.
• The SNN-Tran model demonstrates superior capability in identifying long-range dependencies within medical feature sequences.