Introduction: The evolving paradigm of sarcopenia screening

Sarcopenia, the progressive loss of skeletal muscle mass, strength, and physical performance, is now recognized as a muscle disease (ICD-10-CM M62.84) with profound implications for falls, frailty, metabolic dysfunction, and mortality. Historically, screening relied on dual-energy X-ray absorptiometry (DXA) or bioelectrical impedance analysis (BIA) combined with grip strength and gait speed. However, these tools suffer from limited accessibility, operator dependence, and poor sensitivity for early-phase muscle deterioration—particularly in individuals with obesity (sarcopenic obesity) or those with normal mass but reduced muscle quality (myosteatosis). The past three years have witnessed a paradigm shift toward opportunistic, automated, and biologically informed screening approaches. This review highlights recent breakthroughs in imaging-based deep learning, point-of-care ultrasound, digital gait and wearable sensors, and circulating biomarkers, while addressing the challenges of standardization and implementation.

Deep learning on routine CT and MRI: From opportunistic to prospective screening

One of the most transformative advances is the use of artificial intelligence (AI) to quantify muscle mass and quality from routine abdominal or thoracic computed tomography (CT) scans, obtained for unrelated indications (e.g., cancer staging, trauma). A landmark multicenter study by Kim et al. (2024,Journal of Cachexia, Sarcopenia and Muscle) trained a 3D convolutional neural network on 12,000 CT scans to automatically segment the L3 vertebral level and derive skeletal muscle index (SMI) and mean muscle attenuation (a proxy for intramuscular fat). The model achieved a Dice coefficient of 0.96 and, crucially, identified sarcopenia in 31% of patients who were previously classified as “normal” by DXA-based mass thresholds, because it captured poor muscle quality. This has led to the concept of “opportunistic sarcopenia screening” embedded in routine radiology workflows, with automated reports flagging high-risk patients without additional radiation or cost.

More recently, the integration of radiomics—extracting hundreds of texture and shape features from muscle regions—has improved prognostic accuracy. A 2025 study inEuropean Radiologydemonstrated that a combined deep-learning and radiomics signature predicted 5-year all-cause mortality better than SMI alone (AUC 0.84 vs. 0.71). However, generalizability across different CT scanners and protocols remains a barrier, prompting the development of harmonization techniques such as cycle-consistent generative adversarial networks (CycleGAN) to normalize image intensities.

Ultrasound: A portable, radiation-free alternative with AI assistance

Point-of-care ultrasound (POCUS) has emerged as a promising screening tool, particularly in primary care and geriatric wards. The key parameters are rectus femoris cross-sectional area (RF-CSA), muscle thickness, and echo intensity (EI). The major breakthrough is the development of automated, real-time AI algorithms that guide novice operators to the correct anatomical plane and perform instant segmentation. A prospective trial by Perkisas et al. (2024,Ultrasound in Medicine & Biology) compared AI-assisted POCUS against MRI as the reference standard in 220 community-dwelling older adults. The AI system reduced inter-operator variability from an ICC of 0.78 to 0.95 and achieved sensitivity of 89% for sarcopenia diagnosis. Furthermore, the addition of shear-wave elastography (SWE) to measure muscle stiffness—which correlates with fibrosis and reduced contractile function—has been shown to detect sarcopenic changes up to 18 months before clinically evident weakness (Narici et al., 2025,Age and Ageing). This positions ultrasound not merely as a diagnostic tool but as a longitudinal monitoring modality for pre-sarcopenia.

Digital biomarkers: Gait, grip, and wearable-derived metrics

The ubiquity of smartphones and smartwatches has catalyzed the use of digital biomarkers for remote, continuous screening. A multi-cohort study published inNature Medicine(2024) used inertial measurement units (IMUs) from wrist-worn devices to extract gait features (stride time variability, double support time, and gait speed) during unsupervised daily activities. The algorithm, trained on 8,000 participants, detected sarcopenia with an AUC of 0.82, comparable to clinic-based gait speed tests, but with the advantage of capturing real-world performance over weeks. Notably, the study introduced a novel “muscle fatigue index” derived from the decay in acceleration amplitude during a 6-minute walk, which outperformed static grip strength for predicting incident disability.

Another innovative approach is the use of smartphone camera-based timed up-and-go (TUG) tests with pose estimation (MediaPipe or OpenPose). A 2025 validation study inJournal of Medical Internet Researchdemonstrated that a 30-second chair-stand test, recorded on a smartphone and analyzed by a pose-estimation algorithm, could estimate lower-limb muscle power with a correlation coefficient of 0.87 against isokinetic dynamometry. These digital assessments enable self-administered screening, reducing clinician burden and enabling large-scale population screening.

Circulating biomarkers and multi-omics: Toward molecular staging

While imaging and function capture the phenotype, blood-based biomarkers offer mechanistic insight and potential for early detection. The most promising candidates include:

Future directions: Multimodal fusion, personalized thresholds, and primary care integration

The future of sarcopenia screening lies in multimodal fusion—combining imaging, functional, and molecular data into a single risk algorithm. The European Working Group on Sarcopenia in Older People (EWGSOP3) is currently evaluating a digital decision-support tool that integrates AI-analyzed CT, smartphone gait metrics, and a dried-blood-spot miRNA panel. Early simulation data suggest that this approach could reduce false-positive referrals for DXA by 40%.

Another critical frontier is the establishment of ethnicity- and sex-specific cutoff values for AI-derived imaging parameters. Current thresholds derived from Caucasian cohorts misclassify Asian populations, where lower muscle mass is offset by higher muscle quality. Ongoing work using federated learning across Asian and European cohorts aims to develop harmonized, population-adjusted algorithms without sharing raw patient data.

Finally, implementation science must address equity. Low-cost ultrasound probes that plug into smartphones, combined with open-source AI models, could enable screening in low-resource settings. Pilot programs in rural India and sub-Saharan Africa have demonstrated feasibility, with community health workers achieving adequate image acquisition after 4 hours of training.

Conclusion

Sarcopenia screening is rapidly evolving from a single-timepoint, mass-focused measurement to a dynamic, multimodal, and molecularly informed process. AI-driven imaging and ultrasound enable opportunistic and portable detection, while digital wearables provide continuous real-world functional data. Circulating biomarkers promise earlier and more mechanistically specific detection. The next decade will likely see the consolidation of these tools into a unified, risk-based screening pathway embedded within routine health checks—ultimately enabling timely interventions that preserve independence and quality of life in aging populations.

References