[Paper Review] Meningioma Analysis and Diagnosis using Limited Labeled Samples

Jiamiao Lu, Wei Wu+9 more|arXiv (Cornell University)|Feb 11, 2026

Meningioma and schwannoma management0 citations

TL;DR

The paper introduces an adaptive multi-scale spatial-frequency fusion network (AMSF-Net) for few-shot meningioma learning, integrating spatial and frequency-domain features to improve MRI-based grading with limited labeled data. A new XJTU Meningioma dataset is presented for evaluation.

ABSTRACT

The biological behavior and treatment response of meningiomas depend on their grade, making an accurate diagnosis essential for treatment planning and prognosis assessment. We observed that the weighted fusion of spatial-frequency domain features significantly influences meningioma classification performance. Notably, the contribution of specific frequency bands obtained by discrete wavelet transform varies considerably across different images. A feature fusion architecture with adaptive weights of different frequency band information and spatial domain information is proposed for few-shot meningioma learning. To verify the effectiveness of the proposed method, a new MRI dataset of meningiomas is introduced. The experimental results demonstrate the superiority of the proposed method compared with existing state-of-the-art methods in three datasets. The code will be available at: https://github.com/ICL-SUST/AMSF-Net

Motivation & Objective

• Address the challenge of grading meningiomas with limited labeled data.
• Leverage both spatial-domain and frequency-domain features to enhance classification.
• Develop an adaptive fusion architecture that can learn from few-shot MRI data.
• Introduce and validate on a new MRI dataset (XJTU Meningioma) to demonstrate effectiveness.

Proposed method

• Propose AMSF-Net with two modules: AMFF for multi-scale wavelet-based feature extraction and ACA-SFF for adaptive cross-attention fusion of spatial and frequency information.
• Apply a cascaded discrete wavelet transform to obtain low- and high-frequency subbands across multiple scales (L-level DWT) for frequency-aware representations.
• Use adaptive fusion gates to weigh LH, HL, HH subbands and fuse multi-scale features into a unified representation.
• Incorporate a reconstruction-based episodic classifier for final few-shot decision making within an N-way K-shot setting.
• Utilize an episodic training strategy with task sampling to simulate few-shot learning during training; ensure subject-disjoint splits for evaluation.

Experimental results

Research questions

• RQ1 How can limited labeled samples be effectively used to grade meningiomas from MRI?
• RQ2 Can integrating spatial-domain and frequency-domain features improve few-shot meningioma classification?
• RQ3 Does an adaptive multi-scale spatial-frequency fusion approach outperform existing methods on new and existing datasets?

Key findings

• AMSF-Net demonstrates improved classification performance over state-of-the-art methods on three datasets (as reported in the abstract).
• The method reveals that adaptive fusion of frequency bands from DWT contributes variably across images, underscoring the need for adaptive weighting.
• A new MRI dataset (XJTU Meningioma) was introduced to evaluate few-shot learning for meningioma grading.
• The framework supports few-shot learning by using episodic training and cross-domain attention to fuse spatial and frequency features.
• The approach integrates an end-to-end architecture with a reconstruction-based episodic classifier for final decisions.