ReMAP-PET: Beyond Visual Understanding -- Learning Region-Guided Metabolic Alignment Semantics from Brain PET

2026-06-28

AI Summary

This work addresses the limitation of existing 3D brain foundation models that treat PET scans as generic volumetric data, thereby neglecting their regional metabolic characteristics and hindering effective assessment of neurodegenerative diseases. To overcome this, the study introduces regional standardized uptake value ratio (SUVR) as explicit metabolic semantic supervision for the first time. By jointly optimizing SUVR regression and contrastive learning objectives, the authors partially fine-tune a MedicalNet 3D ResNet-50 backbone and align it with BioClinicalBERT to enable language-compatible radiology report generation. The resulting metabolism-aware embedding space exhibits clear structure and interpretability, achieving a SUVR mean absolute error of 0.070 and 77.8% Recall@1 on a dataset of 1,015 PET-SUVR samples—significantly outperforming frozen baselines. Notably, the model demonstrates clinical utility in diagnostic classification and cognitive scoring tasks without any task-specific fine-tuning.

Abstract

Positron Emission Tomography (PET) reveals brain metabolism and is clinically central to neurodegenerative disease assessment, yet existing 3D brain foundation models treat PET as generic volumetric data, missing the structured regional metabolic information that distinguishes it from structural neuroimaging. To address these limitations, we propose ReMAP-PET, a framework that moves beyond visual encoding by supervising a partially-tuned MedicalNet 3D ResNet-50 with brain regional standardized uptake value ratio (SUVR) profiles through joint regression and contrastive objectives, enabling the encoder to learn the metabolic semantics underlying PET modality. On 1015 paired PET--SUVR samples, ReMAP-PET achieves 0.070 SUVR MAE and 77.8% PET SUVR Recall@1, substantially outperforming five frozen pretrained baselines. We further connect the metabolic embedding to clinical language via contrastive alignment with frozen BioClinicalBERT and demonstrate end-to-end PET-to-report generation through SUVR-constrained verbalization. Linear probing on diagnostic classification and cognitive regression tasks confirms that the embeddings retain clinically relevant information without task-specific fine-tuning. Our results show that grounding PET encoders in regional metabolic semantics -- rather than treating PET as generic volumetric data -- yields representations that are structured, interpretable, and language-compatible, pointing to a new direction for metabolic-aware PET understanding.

Problem

• PET
• metabolic semantics
• regional SUVR
• neurodegenerative disease
• brain imaging

Innovation

• metabolic semantics
• region-guided learning
• SUVR supervision
• contrastive alignment
• PET-language grounding