Introduction: Beyond the body mass index paradigm

For decades, clinical obesity assessment has relied almost exclusively on body mass index (BMI), a crude proxy that conflates adiposity with lean mass, fails to capture fat distribution, and overlooks metabolic health heterogeneity. The limitations of BMI—particularly its inability to distinguish visceral adiposity from subcutaneous fat, its poor sensitivity in sarcopenic obesity, and its ethnic variability—have driven a paradigm shift toward multi-dimensional, mechanism-informed assessment. Recent advances span three interconnected domains: (1) imaging-based phenotyping with artificial intelligence (AI), (2) molecular biomarkers derived from multi-omics, and (3) digital health technologies enabling continuous, real-world metabolic monitoring. Together, these innovations are redefining obesity not as a single disease, but as a spectrum of adiposity-associated dysfunctions requiring personalized staging and intervention.

Imaging-based precision phenotyping: From DXA to AI-driven body composition

Dual-energy X-ray absorptiometry (DXA) and magnetic resonance imaging (MRI) have long provided gold-standard measurements of fat mass and lean mass. However, their clinical utility has been hampered by cost, accessibility, and the need for manual segmentation. Recent breakthroughs in deep learning have automated this process with remarkable accuracy. In a 2023 study published inRadiology, Pickhardt and colleagues validated a fully automated CT-based tool that quantifies visceral adipose tissue (VAT), subcutaneous adipose tissue (SAT), and skeletal muscle area from routine abdominal scans, achieving intra-class correlation coefficients above 0.98 against manual annotations. More importantly, the AI-derived VAT volume outperformed BMI in predicting all-cause mortality in a cohort of over 10,000 patients (hazard ratio 1.67 per standard deviation increase vs. 1.21 for BMI). This "opportunistic screening" approach—extracting body composition metrics from images acquired for unrelated indications—represents a low-cost, high-throughput strategy for population-level obesity assessment.

Simultaneously, bioelectrical impedance analysis (BIA) has evolved from single-frequency devices to multi-frequency segmental systems that estimate extracellular and intracellular water, enabling phase angle measurements as a proxy for cellular health. A 2024 meta-analysis inObesity Reviewsdemonstrated that phase angle <5.4° independently predicted frailty and sarcopenic obesity, adding a functional dimension to static fat measurements.

Molecular biomarkers: The multi-omics revolution

Adipose tissue is no longer viewed as inert energy storage; it is a dynamic endocrine organ secreting adipokines, exosomes, and free nucleic acids. The field has moved from single biomarkers (leptin, adiponectin) toward composite molecular signatures. The most striking advance is the use of circulating microRNAs (miRNAs) as stable, minimally invasive indicators of adipose tissue dysfunction. A landmark study by Lorente-Cebrián et al. (2023,Nature Metabolism) identified a panel of 14 miRNAs differentially expressed in subcutaneous versus visceral adipose tissue, with miR-148a-3p and miR-26b-5p showing strong correlation with insulin resistance independent of BMI. These findings suggest that molecular assessment can identify "metabolically unhealthy obesity" (MUO) at a stage when BMI is still within normal range.

Proteomics and metabolomics have added further granularity. Using untargeted plasma metabolomics, the Framingham Heart Study third-generation cohort identified a 21-metabolite signature—including branched-chain amino acids, acylcarnitines, and ceramides—that predicted incident type 2 diabetes with an area under the curve of 0.89, outperforming fasting glucose and HbA1c. This signature effectively captures the "lipotoxicity" pathway linking ectopic fat deposition to insulin resistance, providing a biological readout of obesity's systemic burden. Moreover, epigenetic clocks (e.g., PhenoAge, GrimAge) have been applied to adipose tissue, revealing that accelerated epigenetic aging in visceral fat predicts cardiovascular events beyond chronological age and BMI (Horvath et al., 2024,Aging Cell).

Digital phenotyping: Continuous and contextual assessment

Traditional obesity assessment captures a single time-point snapshot in a clinical setting, missing the dynamic fluctuations of energy balance, eating behavior, and physical activity. Wearable devices—smartwatches, continuous glucose monitors (CGMs), and smart scales—are filling this gap. The integration of CGM data with AI-driven meal detection has enabled "glycemic phenotyping," classifying individuals into distinct postprandial response clusters. Zeevi et al.'s landmark 2015 work on personalized nutrition has been expanded by the 2023 PREDICT study, which used machine learning on CGM, gut microbiome, and lifestyle data to predict postprandial triglyceride and glucose responses with high accuracy, demonstrating that identical meals elicit vastly different metabolic responses across individuals.

A more recent breakthrough is the use of smartphone-based digital biomarkers. Deep learning models applied to facial photographs and voice recordings have shown promise in estimating central adiposity. A 2024 proof-of-concept study innpj Digital Medicineused a convolutional neural network on 2D frontal and lateral facial images to estimate waist-to-hip ratio with a mean absolute error of 2.1 cm, exploiting facial features associated with cortisol-related fat deposition. While not yet diagnostic, this technology could enable low-cost screening in low-resource settings. Additionally, smart toilets and urine sensors are being developed to measure urinary metabolomics (e.g., ketones, uric acid) on a daily basis, potentially capturing real-time shifts in lipid oxidation.

Precision staging: The Edmonton Obesity Staging System and beyond

Recognizing that BMI alone cannot guide treatment intensity, the clinical community has embraced the Edmonton Obesity Staging System (EOSS), which classifies obesity into five stages based on metabolic, mechanical, mental, and social comorbidities. Recent refinements have integrated imaging and molecular data into this framework. The 2024 European Association for the Study of Obesity (EASO) consensus statement proposed a new "adiposity-based chronic disease" (ABCD) model, replacing the term "obesity" with a diagnostic label that specifies etiology, phenotype, and complications. For instance, "ABCD with visceral adiposity, prediabetes, and sarcopenia" provides a actionable precision medicine description. This staging approach has been shown to predict treatment response: patients with EOSS stage 2–3 and high VAT respond significantly better to GLP-1 receptor agonists than those with stage 0–1 and low VAT, as demonstrated in a post-hoc analysis of the STEP-4 trial (2024,The Lancet Diabetes & Endocrinology).

Future directions: Integration, equity, and longitudinal multi-scale modeling

The next frontier lies in fusing these disparate data streams into a unified, longitudinal "digital twin" of the patient. Multi-scale models integrating genomic risk scores, epigenetic clocks, gut microbiome composition, continuous glucose data, and AI-derived imaging metrics could simulate individual responses to diet, exercise, pharmacotherapy, or bariatric surgery before actual implementation. Preliminary work in this direction—using Bayesian networks and reinforcement learning—has been shown to predict 12-month weight loss trajectories with 82% accuracy in a pilot cohort (2025,Obesity).

However, three critical challenges remain. First, algorithmic bias: AI models trained predominantly on Caucasian populations may misclassify body composition in Asian, African, and Latinx individuals, where fat distribution and muscle density differ. Second, accessibility: advanced imaging and multi-omics remain prohibitively expensive for global health settings. Low-cost alternatives—such as point-of-care ultrasound for subcutaneous fat thickness and dried blood spot metabolomics—are under active development. Third, ethical integration: continuous digital monitoring raises privacy concerns, and the definition of "metabolic health" requires culturally sensitive, patient-centered validation.

Conclusion

Obesity assessment has evolved from a static anthropometric measurement to a dynamic, multi-parametric, and personalized discipline. The convergence of AI-driven imaging, molecular signatures, and wearable technology offers an unprecedented opportunity to identify high-risk phenotypes, tailor interventions, and monitor therapeutic efficacy in real time. The future of obesity management lies not in a single number, but in a comprehensive, continuously updated biological and behavioral profile that respects individual variability and promotes health equity.