Scholarlink

Purpose of Review <p>Bone marrow adipose tissue (BMAT) is a significant fat depot with distinct skeletal, haematological and metabolic roles. It increases with ageing, osteoporosis, metabolic disease, and cancer, and is emerging as a biomarker for fracture risk. Quantification of bone marrow adiposity (BMA) has relied on MRI, proton spectroscopy, computed tomography, or histology, but large-scale studies have been limited by requiring labour-intensive analysis. Deep learning (DL) now enables scalable, automated BMA measurement, with greatest progress in MRI. This review highlights recent advances.</p> Recent Findings <p>DL models, especially U-Nets, have been applied to UK Biobank MRI data, allowing site-specific BM fat fraction (BMFF) measurement or calvarial BMA estimation in tens of thousands of participants. These breakthroughs have revealed robust associations between BMA and age, sex, ethnicity, bone mineral density, adiposity, and metabolic traits, while uncovering site-specific patterns. Genome-wide association studies of BMFF and calvarial BMA have defined their genetic architecture, identifying hundreds of loci enriched for pathways in oestrogen signalling, adipogenesis, and skeletal remodelling. Phenome-wide association studies demonstrate links between altered BMFF and osteoporosis, fracture, type 2 diabetes, cardiovascular disease, and diverse other conditions, with Mendelian randomisation providing the first causal evidence that increased femoral BMFF contributes to osteoporosis. Despite successes, challenges remain, including extending analyses to non-European ancestries and validating DL pipelines in clinical settings.</p> Summary <p>Collectively, DL-enabled BMA quantification has established BMAT as a clinically relevant, genetically tractable fat depot and provides new opportunities for mechanistic insight, risk prediction, and therapeutic targeting in musculoskeletal, metabolic, and other diseases.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep Learning for Analysis of Bone Marrow Adiposity: Breakthroughs from Recent Large-Scale Analyses in the UK Biobank

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月10日

查看全文
  • Wei Xu,
  • Chengjia Wang,
  • William P Cawthorn

摘要

Purpose of Review

Bone marrow adipose tissue (BMAT) is a significant fat depot with distinct skeletal, haematological and metabolic roles. It increases with ageing, osteoporosis, metabolic disease, and cancer, and is emerging as a biomarker for fracture risk. Quantification of bone marrow adiposity (BMA) has relied on MRI, proton spectroscopy, computed tomography, or histology, but large-scale studies have been limited by requiring labour-intensive analysis. Deep learning (DL) now enables scalable, automated BMA measurement, with greatest progress in MRI. This review highlights recent advances.

Recent Findings

DL models, especially U-Nets, have been applied to UK Biobank MRI data, allowing site-specific BM fat fraction (BMFF) measurement or calvarial BMA estimation in tens of thousands of participants. These breakthroughs have revealed robust associations between BMA and age, sex, ethnicity, bone mineral density, adiposity, and metabolic traits, while uncovering site-specific patterns. Genome-wide association studies of BMFF and calvarial BMA have defined their genetic architecture, identifying hundreds of loci enriched for pathways in oestrogen signalling, adipogenesis, and skeletal remodelling. Phenome-wide association studies demonstrate links between altered BMFF and osteoporosis, fracture, type 2 diabetes, cardiovascular disease, and diverse other conditions, with Mendelian randomisation providing the first causal evidence that increased femoral BMFF contributes to osteoporosis. Despite successes, challenges remain, including extending analyses to non-European ancestries and validating DL pipelines in clinical settings.

Summary

Collectively, DL-enabled BMA quantification has established BMAT as a clinically relevant, genetically tractable fat depot and provides new opportunities for mechanistic insight, risk prediction, and therapeutic targeting in musculoskeletal, metabolic, and other diseases.