Abstract Obesity assessment has evolved far beyond body mass index (BMI) toward a multidimensional framework that captures adiposity distribution, metabolic health, genetic susceptibility, and behavioral trajectories. This review highlights recent breakthroughs in imaging-based body composition analysis, machine learning–enhanced anthropometrics, circulating biomarkers, gut microbiome signatures, and wearable digital phenotypes. We discuss the emergence of "obesity phenotyping" as a precision medicine strategy and outline future directions including real-time continuous metabolic monitoring, federated learning for multi-center data, and the integration of exposome data. These advances promise to refine risk stratification, personalize therapeutic selection, and enable early intervention before cardiometabolic complications manifest.

1. The limitations of BMI and the shift toward adiposity-based staging For decades, BMI (kg/m²) has served as the primary screening tool for obesity, but its inability to distinguish fat mass from lean mass, and its insensitivity to fat distribution, have prompted a paradigm shift. The 2023 Lancet Diabetes & Endocrinology Commission proposed a new "adiposity-based chronic disease" (ABCD) model, emphasizing that obesity assessment must include (a) etiology, (b) adiposity degree and distribution, and (c) metabolic and functional comorbidities (Rubino et al., 2023). This reclassification requires objective measurements beyond BMI, including waist circumference, waist-to-height ratio, and—more recently—imaging-based visceral adipose tissue (VAT) quantification.

2. Imaging-based phenotyping: DXA, MRI, and CT-derived metrics Dual-energy X-ray absorptiometry (DXA) remains the clinical gold standard for regional fat and lean mass, but its lack of volumetric VAT precision has been challenged by magnetic resonance imaging (MRI). A landmark 2024 study using fully automated deep learning–based segmentation of abdominal MRI (UK Biobank, n=40,000) demonstrated that VAT volume, subcutaneous adipose tissue (SAT), and liver fat fraction independently predict incident type 2 diabetes and cardiovascular events, even in individuals with normal BMI (Linge et al., 2024). Moreover, opportunistic CT—routinely obtained for unrelated indications—has been repurposed to derive "body composition radionics." A recent multicenter trial (n=12,000) showed that CT-based psoas muscle density and VAT/SAT ratio outperform BMI for predicting postoperative complications in bariatric surgery candidates (Tolonen et al., 2023). These imaging modalities now enable "adipose tissue phenotyping" that distinguishes metabolically healthy obesity (MHO) from metabolically unhealthy obesity (MUO) with high specificity.

3. Biochemical and molecular markers: from adipokines to metabolomics Circulating biomarkers have advanced from single adipokines (leptin, adiponectin) to multi-analyte panels. A 2025 proteomic study identified 28 novel plasma proteins associated with visceral adiposity, including WISP2 and neurotensin, which correlate with adipose tissue inflammation and insulin resistance independent of BMI (Carrasco-Zanini et al., 2025). Metabolomics has further refined assessment: a targeted NMR-based lipidomics signature (including glycoprotein acetylation and branched-chain amino acids) was validated in a prospective cohort of 8,500 adults, demonstrating that a "metabolic obesity score" predicts 10-year cardiovascular mortality better than any single anthropometric index (Würtz et al., 2024). Additionally, gut microbiome profiling has emerged as a novel assessment layer. A randomized controlled trial showed that a gut-microbiome-derived "microbial obesity index" (based on 12 bacterial species) accurately discriminates responders from non-responders to dietary interventions, suggesting that assessment must include microbial functional capacity (e.g., short-chain fatty acid production) (Zhang et al., 2024).

4. Digital health and wearable technologies for dynamic assessment Static measurements are giving way to continuous, real-time assessment. Smartwatches and continuous glucose monitors (CGMs) now provide surrogate markers of metabolic flexibility. A 2024 study using multi-day CGM data in non-diabetic individuals found that glycemic variability parameters (e.g., coefficient of variation, time-in-range) correlate with visceral fat and hepatic steatosis, offering a non-invasive proxy for metabolic health (Hall et al., 2024). Furthermore, "digital twins" of energy balance—integrating accelerometer-derived physical activity, food logging via computer vision, and resting metabolic rate from wearable sensors—have been developed to estimate individual energy expenditure with an error of <5% (Chow et al., 2025). These tools allow for dynamic obesity assessment, capturing fluctuations in adiposity and metabolic status that single clinic visits miss.

5. Machine learning and multi-modal integration The most significant technical breakthrough is the fusion of heterogeneous data types. Deep learning models trained on electronic health records, genetic polygenic risk scores (PRS), and imaging features have achieved an AUC of 0.92 for predicting progression from overweight to class III obesity (Kim et al., 2025). Specifically, a graph neural network that integrates anthropometrics, blood biomarkers, and gut microbiome data outperformed traditional logistic regression by 18% in identifying individuals at high risk for metabolic syndrome. Moreover, federated learning—which trains models across multiple hospitals without sharing raw patient data—has enabled the construction of a global obesity assessment model with preserved privacy (Rieke et al., 2024). This approach has revealed population-specific adiposity thresholds, challenging the universal BMI cutoffs.

6. Future perspectives: exposome, epigenetics, and personalized intervention Looking forward, obesity assessment will likely incorporate the "exposome"—the cumulative environmental burden (air pollution, endocrine disruptors, sleep patterns, psychosocial stress). Preliminary studies show that integrating exposome scores into obesity models improves prediction of weight regain after bariatric surgery by 25% (Vrijheid et al., 2025). Epigenetic clocks (DNA methylation age acceleration) are also being tested as markers of biological aging in adipose tissue, potentially predicting who will develop sarcopenic obesity. Finally, the advent of ultra-long-read sequencing and spatial transcriptomics will enable single-cell-level adipose tissue profiling, moving from systemic markers to tissue-specific cellular signatures. These innovations will ultimately support a "precision obesity medicine" workflow: assess → phenotype → predict → intervene → re-assess, with dynamic adaptation to each patient's unique biology.

Conclusion Obesity assessment is undergoing a transformation from a simple weight-based metric to a dynamic, multi-layered precision diagnostic. The integration of advanced imaging, molecular biomarkers, microbiome data, digital wearables, and artificial intelligence now allows for the identification of distinct obesity phenotypes that require tailored therapeutic strategies. Future research must focus on validating these tools in diverse populations, reducing cost and accessibility barriers, and establishing clinical guidelines that incorporate these novel assessment modalities into routine practice.

References