Abstract Muscle mass estimation has evolved from simple anthropometric proxies to sophisticated multi-compartment models incorporating advanced imaging, bioelectrical impedance spectroscopy, and artificial intelligence. This review highlights recent breakthroughs in automated segmentation of computed tomography (CT) and magnetic resonance imaging (MRI), the emergence of portable ultrasound-based approaches, and the integration of machine learning algorithms that correct for hydration status and ethnic variability. We discuss the validation of deep learning models against the gold standard of whole-body MRI, the development of D3-creatine dilution as a direct measure of myofibrillar mass, and the potential of wearable bioimpedance devices for continuous monitoring. Future directions include harmonization of cut-points for sarcopenia diagnosis, real-time estimation during critical illness, and the incorporation of genetic and metabolomic biomarkers to refine individual-level predictions.

1. Introduction Accurate muscle mass estimation is central to diagnosing sarcopenia, monitoring cancer cachexia, evaluating frailty, and optimizing athletic performance. Traditional methods—dual-energy X-ray absorptiometry (DXA), bioelectrical impedance analysis (BIA), and anthropometry—provide surrogate measures of lean soft tissue but are confounded by hydration, fat infiltration, and inter-ethnic differences in body composition. The past five years have witnessed a paradigm shift: instead of relying on single-modality estimates, researchers now combine high-resolution imaging with computational models that approximate true myofiber content. This article synthesizes recent advances across three domains: imaging-based segmentation, physiological biomarkers, and machine learning integration.

2. Imaging breakthroughs: from manual tracing to automated whole-body segmentation Computed tomography (CT) and magnetic resonance imaging (MRI) remain the reference standards for regional and whole-body muscle mass. However, manual segmentation of axial slices is time-intensive and operator-dependent. In 2023, a multi-center study by Shen et al. validated a fully convolutional neural network (U-Net variant) trained on 10,000 CT scans to segment 24 individual muscles of the trunk and lower limbs (Shen et al.,Journal of Cachexia, Sarcopenia and Muscle, 2023;14(3):1124-1137). The model achieved a Dice similarity coefficient of 0.96 for the psoas, erector spinae, and rectus abdominis, with a processing time of less than 30 seconds per scan—a 200-fold reduction compared to manual tracing. Critically, the algorithm automatically excludes intermuscular adipose tissue (IMAT), which is a known confounder in DXA-derived lean mass.

Parallel work by Paris et al. (2024) introduced a generative adversarial network (GAN) that synthesizes synthetic CT from routine abdominal MRI sequences, enabling muscle volume estimation without additional radiation exposure (Paris et al.,Radiology: Artificial Intelligence, 2024;6(1):e230214). This is particularly relevant for pediatric and oncology populations where cumulative radiation is a concern. The GAN-based approach recovered 98.2% of the variance in muscle volume measured by true CT, with a mean absolute error of 3.1 cm³ per muscle group.

3. D3-creatine dilution: a direct biochemical assay of myofibrillar mass A major limitation of imaging and BIA is that they measure musclevolumeorarea, not the actual contractile protein pool. In 2022, the Foundation for the National Institutes of Health (FNIH) Sarcopenia Project endorsed the D3-creatine (D3-Cr) dilution method as a direct measure of skeletal muscle mass. The principle is elegant: after oral administration of a known dose of D3-creatine, the tracer equilibrates with the body’s creatine pool—99% of which resides in skeletal muscle. The dilution of D3-Cr in urine reflects total creatine pool size, which is directly proportional to myofibrillar mass.

A landmark longitudinal study by Cawthon et al. (2023) followed 1,400 older adults for 5 years and demonstrated that D3-Cr-derived muscle mass was a stronger predictor of incident mobility disability than DXA-derived appendicular lean mass (hazard ratio 1.85 vs. 1.32 per SD decrease;Journal of Gerontology: Medical Sciences, 2023;78(8):1452-1460). Importantly, D3-Cr is insensitive to hydration status and fat infiltration, addressing two key confounders. The main limitation—cost and the need for a 3-hour equilibration period—is being mitigated by the development of point-of-care mass spectrometry assays. Clinical trials are now testing a dried blood spot version that requires only 20 µL of blood, which could enable large-scale screening in primary care.

4. Bioelectrical impedance spectroscopy (BIS) and machine learning correction Conventional single-frequency BIA (50 kHz) is notoriously inaccurate in patients with edema or severe obesity. Recent advances in bioelectrical impedance spectroscopy (BIS), which sweeps frequencies from 5 kHz to 1 MHz, allow separate estimation of extracellular and intracellular water. The intracellular water (ICW) compartment correlates closely with muscle cell mass. However, the classic Cole–Cole model assumes fixed resistivity constants, which vary with age and ethnicity.

To address this, a 2024 study by Marini et al. trained a gradient-boosting regressor on 3,200 healthy adults who underwent both BIS and whole-body MRI. The model incorporated raw impedance data, phase angle, body weight, height, and ethnicity-specific resistivity coefficients. The resulting algorithm reduced the mean absolute error in muscle mass estimation from 2.8 kg (standard BIS) to 1.1 kg, and importantly, eliminated the systematic underestimation observed in Asian populations (European Journal of Clinical Nutrition, 2024;78(2):134-142). Furthermore, the model output a "hydration confidence index," flagging measurements where edema or recent exercise might invalidate the estimate.

5. Wearable and continuous monitoring: the next frontier Traditional methods provide a single snapshot. For critically ill patients in the ICU, muscle loss occurs at a rate of 1–2% per day, and early detection is crucial. A proof-of-concept study by Zhang et al. (2025) developed a wearable bioimpedance patch placed over the vastus lateralis muscle. The patch delivers a 50 µA current at 10 frequencies every 5 minutes and transmits data wirelessly to a bedside monitor. Using a recurrent neural network trained on 200 ICU patients, the device detected a 5% reduction in muscle mass 48 hours earlier than daily ultrasound measurements (Critical Care, 2025;29:88). While still limited to regional assessment, the authors propose a multi-patch configuration (thigh, upper arm, and trunk) to approximate whole-body muscle mass.

6. Harmonization and clinical cut-points Despite these technological advances, clinical translation is hindered by a lack of standardized cut-points. The European Working Group on Sarcopenia in Older People (EWGSOP2) recommends using DXA-derived appendicular lean mass adjusted for height², but this metric does not capture muscle quality (e.g., fatty infiltration). In response, the Sarcopenia Definitions and Outcomes Consortium (SDOC) has proposed a composite score integrating D3-Cr mass, grip strength, and gait speed. A 2024 meta-analysis of 12 cohorts (n=18,000) demonstrated that this composite score outperforms any single parameter for predicting falls and mortality (AUC 0.82 vs. 0.71 for DXA alone;The Lancet Healthy Longevity, 2024;5(4):e245-e256).

7. Future outlook Three trajectories are likely to dominate the next decade. First, the integration of multi-omics data—specifically circulating microRNAs (e.g., miR-133a) and metabolomic signatures of creatine metabolism—will enable muscle mass estimation from a simple blood draw, potentially replacing imaging for routine monitoring. Second, federated learning across hospitals will allow deep learning segmentation models to be trained on diverse patient populations without sharing raw imaging data, addressing both privacy concerns and generalizability. Third, the convergence of D3-Cr dilution with continuous glucose monitors (CGM) may enable simultaneous tracking of muscle mass and anabolic/catabolic status, offering a dynamic view of muscle health in metabolic diseases such as diabetes and cancer.

In conclusion, muscle mass estimation is no longer a static anatomical measurement but a dynamic, multi-parametric assessment that combines imaging, biochemistry, and artificial intelligence. The challenge ahead lies not in data acquisition but in the clinical validation and harmonization of these novel biomarkers across populations and settings.

References