Morphomics in esophageal cancer: Validation and association with muscular and cardiorespiratory fitness

Abstract

BACKGROUND

Morphomics, a computed tomography-based body composition assessment, helps predicting esophageal cancer outcomes, but its link to bioelectrical impedance analysis (BIA) and functional assessments such as hand grip strength (HGS) and cardiopulmonary exercise testing (CPET) remains unclear.

AIM

To investigate correlations between morphomics and BIA, HGS, CPET, and assess its ability to predict low cardiorespiratory fitness (CRF).

METHODS

Fifty esophageal cancer patients underwent multi-level morphomics, BIA, HGS, and CPET. Correlations were analyzed using heatmaps and scatter plots, and logistic regression assessed morphomic predictive value for low CRF.

RESULTS

T11 is the only level with complete morphomic data, making it the most applicable. To ensure reliability, T11 and its adjacent levels, T10-12, were included in the subsequent analysis. Dorsal muscle group volume from T10-12 morphomics all correlated positively with BIA muscle components (r = 0.56-0.68, all P < 0.001), HGS (r = 0.4-0.48, all P < 0.001), and CPET variables (r = 0.43-0.51, all P < 0.001). Subcutaneous fat area and visceral fat area from morphomics correlated with body fat percentage (r = 0.58-0.67, all P < 0.001) and negatively with CPET parameters (r = -0.33 to -0.52, all P < 0.05). Morphomics also showed potential in identifying low CRF, with an area under the receiver operating characteristic curve of 0.778.

CONCLUSION

T11 morphomics shows strong correlation with BIA, HGS, and CPET, and may serve as a practical tool for preoperative risk assessment in esophageal cancer patients.

Key Words: Morphomics; Body composition; Bioelectrical impedance analysis; Hand grip strength; Cardiopulmonary exercise testing; Computed tomography

Core Tip: This study explored the relationship between computed tomography-based morphomics and physical assessments, including bioelectrical impedance analysis, hand grip strength, and cardiopulmonary exercise testing in esophageal cancer patients. Morphomics at the eleventh thoracic vertebra level had the most complete data and showed strong correlations with muscle mass, strength, and exercise capacity. Fat measurements from morphomics were inversely related to exercise performance. Morphomics also demonstrated potential to identify patients with low cardiorespiratory fitness, highlighting its clinical potential as a reliable tool for assessing physical fitness and guiding personalized care in esophageal cancer.

INTRODUCTION

Body composition is an essential part in the evaluation of physical health status, and studies have demonstrated its capacity to serve as a prognostic indicator[1]. Conventional approaches to body composition assessment, such as bioelectrical impedance analysis (BIA), have been widely adopted as one of the gold standards due to their accuracy and cost-effectiveness. In recent years, as computed tomography (CT) has become a standard procedure for patients with esophageal cancer, morphomics-a CT-derived body composition measurement-has garnered attention for its precision and its associations with surgical risks, outcome predictions, and long-term survival in patients with esophageal cancer[2-5]. However, most previous studies have focused on a single vertebral level for deriving morphomic parameters, comply L3[6-8]. Nevertheless, this level is not universally conducted in esophageal cancer cases. This study aimed to identify the spine level with the most comprehensive information for CT analysis as a feasible marker in esophageal cancer and to assess its validity by BIA using a novel multi-level approach.

Secondly, the prognostic predictive power of morphomics, is expected to align with muscular fitness and cardiorespiratory fitness (CRF)[9,10]. Hand grip strength (HGS) and cardiopulmonary exercise testing (CPET) are the most common tests for assessing these factors, both of which have well-established prognostic value[11-15]. However, no study to date has explored the association between morphomics and CRF in esophageal cancer patients[16]. This study aimed to strengthen this link.

Lastly, low CRF has been associated with increased risk of complications following major surgery[14,15,17]. In elective major intra-abdominal procedures, a ventilatory anaerobic threshold normalized by body weight (VAT/BW) below 10.9 mL/kg/min has been shown to predict all-cause hospital and 90-day mortality[15]. In the context of esophageal cancer surgery, a VAT/BW below 11 mL/kg/min has been linked to postoperative cardiopulmonary complications[14]. Accordingly, our study investigated whether morphomic parameters can effectively identify esophageal cancer patients with VAT/BW < 11 mL/kg/min.

To the best of our knowledge, this is the first investigation to establish morphomics using BIA as the reference standard in esophageal cancer patients examining the association with CRF. The present study explored the relationships between multi-level morphomics and BIA, HGS and CPET, as well as its potential predictive value.

MATERIALS AND METHODS

Participants

The data were retrospectively collected from 50 newly diagnosed esophageal cancer patients at Chang Gung Memorial Hospital at Linkou (Taoyuan, Taiwan), between July 2019 and July 2022. All patients underwent CT of the chest and abdomen for esophageal cancer staging, along with assessments of muscular and CRF including body composition from morphomics and BIA, HGS and CPET. CT and CPET were performed within a 2-week interval, with HGS measured just prior to CPET. The study was conducted in accordance with the ethical standards of the institutional review board (IRB number: 202500080B0) at Chang Gung Memorial Hospital at Linkou.

Body composition

Morphomics: Analytic morphomics processing was conducted following previously described methods[2,4,5]. All CT scans were processed using a fully automated, artificial intelligence-based algorithm implemented in MATLAB R2016b (The MathWorks, Inc, Natick, MA). Morphomic variables were analyzed at the cross-sectional level of the bottom of the specific vertebra, excluding bone mineral density (BMD). The specific analytic morphomics were as follows: (1) BMD: Average pixel intensity Hounsfield units (HU) within trabecular bone of vertebra at mid-vertebra level; (2) Subcutaneous fat (SF) area: Area of fat-intensity pixels (–205 to –51 HU) between the fascia and skin; (3) SF density: Median pixel intensity of fat-intensity pixels in SF area; (4) Visceral fat (VF) area: Area of fat-intensity pixels within the fascial envelope; (5) VF density: Median pixel intensity of fat-intensity pixels in VF area; (6) Dorsal muscle group (DMG) area: Area of muscle-intensity pixels (-29 to 150 HU) contained in the triangular region between the spinal canal and bilateral lateral seams; (7) DMG volume: Volume of DMG between this vertebra and its superior neighbor; and (8) DMG density: Median pixel intensity of muscle-intensity pixels in DMG area.

BIA: Whole-body composition was assessed using the InBody S10 device (Seoul, Korea), which measures electrical impedance at four frequencies (5, 50, 250, and 500 kHz)[18,19]. Participants remained seated upright on a non-conductive chair throughout the entire 10-minute assessment. Electrodes were positioned on each body segment according to the manufacturer’s guidelines to measure electrical impedance. Participants were instructed to fast for at least 2 hours prior to testing. The following parameters were recorded: Appendicular skeletal muscle index (ASMI), body cell mass (BCM), fat-free mass (FFM), lean body mass (LBM), and body fat percentage (BF).

Muscular fitness

HGS was measured using a dynamometer (Tsutsumi company, TTM, Tokyo). Participants stood naturally with palms facing their bodies. Grip distance was optimized for force. They squeezed the dynamometer with maximum force for about 3 seconds. Two trials were done for each hand, alternating sides. The average of the two highest values, one from each side, was calculated.

Cardiopulmonary fitness

A symptom-limited incremental exercise test was conducted in an upright position using a calibrated cycle ergometer (Ergoselect 150P, Germany) to evaluate cardiopulmonary fitness. The CPET was performed 2-4 hours after a light meal. The protocol included 2 minutes of rest, a 1-minute warm-up at 10 watts, and a ramp increase of 10 watts per minute until the participant reached exhaustion. The participant was requested to keep the pedaling frequency at 60 ± 3 revolutions per minute.

Minute ventilation (Ve), oxygen consumption (VO₂), and carbon dioxide production (VCO₂) were measured breath by breath using a computerized system (MasterScreen CPX, Cardinal Health, Germany). Heart rate was monitored via R–R intervals on a 12-lead electrocardiogram, while arterial blood pressure was assessed with an automated system (Tango, SunTech Medical, United Kingdom). Arterial oxygen saturation was continuously monitored using finger pulse oximetry (Model 9500, Nonin Onyx, Plymouth, Minnesota). The exercise test was terminated under any of the following conditions: (1) The participant could no longer maintain a pedaling cadence of at least 50 rpm; (2) The participant experienced volitional fatigue and asked to stop; (3) Peak VO₂ plateaued or declined despite continued effort; or (4) An adverse cardiovascular event occurred[20]. The data were averaged every 15 seconds.

The ventilatory anaerobic threshold (VAT) was primarily identified using the V-slope method and confirmed through the following ventilatory criteria: (1) A nonlinear increase in VCO₂ relative to VO₂; (2) An increase in the Ve/VO₂ ratio without a corresponding rise in the Ve/VCO₂ ratio; and (3) A rise in end-tidal oxygen pressure without a simultaneous drop in end-tidal carbon dioxide pressure. Ventilatory efficiency was evaluated using four parameters: (1) EqCO₂ nadir: The lowest value of the ventilatory equivalent for CO₂ (Ve/VCO₂) during the incremental exercise test; (2) EqO₂ nadir: The lowest value of the ventilatory equivalent for O₂ (Ve/VO₂) throughout The test; (3) Ve-VCO₂ slope: Calculated using linear regression (Y = mX + b), where Y represents Ve, X represents VCO₂, and m is the slope. Ve and VCO₂ values were taken from the start of exercise to peak effort. A flatter slope indicates more efficient Ve; and (4) Oxygen uptake efficiency slope: Calculated from the slope of VO₂ plotted against the logarithm of Ve [(L/min)/Log(L/min)], reflecting the efficiency of oxygen uptake during exercise. A higher slope indicates better oxygen utilization efficiency[21].

Statistical analysis

All statistical analyses were conducted using R (version 4.4.1) and RStudio (version 2014.12.0). Continuous variables were presented as mean ± SD, whereas categorical variables, such as gender, were reported as counts.

This study employed a novel multi-level approach, ultimately selecting levels with the least information loss to ensure feasibility. Pearson correlation coefficients were calculated to evaluate univariable relationships between BIA, CPET, HGS, and morphomics. Only those with correlation coefficients greater than 0.3 and reaching significance at all levels are considered reliable correlations.

Multivariable logistic regression was used to predict the likelihood of patients having an VAT/BW below 11 mL/kg/min, which has been identified as a clinically meaningful cutoff for predicting complications in major surgery[14,15]. Multicollinearity was examined with variance-inflation factors before model fitting. All eight morphomic variables were initially entered as covariates, including BMD, DMG area, DMG volume, DMG density, SF area, SF density, VF area and VF density. We then adopted a bidirectional stepwise variable-selection procedure and yielded a parsimonious model comprising BMD, VF area, and DMG volume. Logistic regression models were fitted with both the full and reduced predictor sets, and odds ratios (OR) with 95% confidence intervals (CI) were reported to quantify the contribution of each variable. To ensure validity, internal validation was done through 1000 bootstrap resamples. We reported OR of each variable in both full model and the parsimonious model to assess the importance of each variable. Receiver operating characteristic (ROC) analysis was then employed to assess the predictive performance of these variables in distinguishing individuals with VAT/BW ≤ 11 mL/kg/min. Significance of the area under the ROC curve (AUC) was tested using the Wilcoxon test, with statistical significance set at P value < 0.05. Both apparent ROC and optimism-corrected ROC were presented.

RESULTS

Participant characteristics

The majority of the cohort was male (98%), with a mean age of 57.3 years. The summary of variables of morphomics, BIA, HGS and CPET were recorded in Supplementary Table 1.

Spine level selection

Morphomic variables were extracted from cross-sectional levels spanning the thoracic to lumbar vertebrae. To ensure data reliability, vertebral levels with more than 5% missing values were excluded from the analysis. Consequently, T10, T11, and T12 were selected, as their missing data rates were 0.5%, 0%, and 0.5%, respectively. In contrast, the missing data rates at L1, L2, and L3 were significantly higher (15.5%, 21.5%, and 34%, respectively), making them unsuitable for further analysis.

Body composition correlation between morphomics and BIA

For muscle component, DMG volume from morphomics at T10-T12 Level had positive correlation with skeletal muscle mass (SMM) (r = 0.6-0.65), FFM (r = 0.6-0.64), LBM (r = 0.56-0.58), BCM (r = 0.6-0.68) and appendicular skeletal muscle (ASM) (r = 0.6-0.67) (all P < 0.001). A similar relationship was also observed between DMG area and SMM (r = 0.55-0.65), FFM (r = 0.53-0.64), LBM (r = 0.5-0.61), BCM (r = 0.54-0.66) and ASM (r = 0.54-0.65) (all P < 0.001) (Table 1; Figure 1A, Figure 2).

Figure 1 Heatmap of correlation matrix between morphomics. A: Bioelectrical impedance analysis related body composition; B: Cardiopulmonary exercise testing and hand grip strength. DMG: Dorsal muscle group; BMD: Bone mineral density; SF: Subcutaneous fat; VF: Visceral fat; BCM: Body cell mass; SM: Skeletal muscle; LBM: Lean body mass; FFM: Fat free mass; ASM: Appendicular skeletal muscle; ASMI: Appendicular skeletal muscle index; LA: Left arm; LL: Left leg; MM: Muscle mass; RA: Right arm; RL: Right leg; TR: Trunk; HGS: Hand grip strength; OUES: Oxygen uptake efficiency slope; WR: Work rate; VAT: Ventilatory anaerobic threshold; BW: Body weight; EqO₂: Ventilatory equivalent for O₂; VO₂: Oxygen consumption; EqCO₂: Ventilatory equivalent for CO₂.

Figure 2 Scatterplots of selected parameters between morphomics and bioelectrical impedance analysis/ hand grip strength/ cardiopulmonary exercise testing (orange line: Linear regression; grey line: Locally estimated scatterplot smoothing curve; blue shading = 95%CI; n = 50). DMG: Dorsal muscle group; BMD: Bone mineral density; SF: Subcutaneous fat; VF: Visceral fat; BW: Body weight; LBM: Lean body mass; FFM: Fat free mass; ASM: Appendicular skeletal muscle; VAT: Ventilatory anaerobic threshold; HGS: Hand grip strength; SM: Skeletal muscle; VO₂: Oxygen consumption.

Table 1 Body composition correlation between T10-12 morphomics and bioelectrical impedance analysis.

Bioelectrical impedance analysis	Morphomics	Spine level	Correlation	P value
Body cell mass	DMG volume	T10	0.594	< 0.001
		T11	0.647	< 0.001
		T12	0.684	< 0.001
	DMG area	T10	0.647	< 0.001
		T11	0.543	< 0.001
		T12	0.664	< 0.001
Body fat percentage	SF area	T10	0.666	< 0.001
		T11	0.672	< 0.001
		T12	0.674	< 0.001
	VF area	T10	0.576	< 0.001
		T11	0.578	< 0.001
		T12	0.606	< 0.001
	DMG density	T10	-0.469	< 0.001
		T11	-0.457	< 0.001
		T12	-0.513	< 0.001
Fat free mass	SF area	T10	0.597	< 0.001
		T11	0.599	< 0.001
		T12	0.590	< 0.001
	SF area	T10	-0.244	0.09
		T11	-0.281	0.048
		T12	-0.420	0.002
	VF area	T10	0.491	< 0.001
		T11	0.568	< 0.001
		T12	0.491	< 0.001
	DMG volume	T10	0.589	< 0.001
		T11	0.642	< 0.001
		T12	0.589	< 0.001
	DMG area	T10	0.643	< 0.001
		T11	0.528	< 0.001
		T12	0.643	< 0.001
Lean body mass	DMG volume	T10	0.556	< 0.001
		T11	0.584	< 0.001
		T12	0.556	< 0.001
	DMG area	T10	0.613	< 0.001
		T11	0.459	< 0.001
		T12	0.613	< 0.001
Skeletal muscle mass	DMG volume	T10	0.598	< 0.001
		T11	0.650	< 0.001
		T12	0.598	< 0.001
	DMG area	T10	0.650	< 0.001
		T11	0.546	< 0.001
		T12	0.650	< 0.001
Appendicular skeletal muscle	DMG volume	T10	0.603	< 0.001
		T11	0.670	< 0.001
		T12	0.671	< 0.001
	DMG area	T10	0.638	< 0.001
		T11	0.542	< 0.001
		T12	0.653	< 0.001
Muscle mass, left arm	DMG volume	T10	0.540	< 0.001
		T11	0.679	< 0.001
		T12	0.540	< 0.001
	DMG area	T10	0.628	< 0.001
		T11	0.605	< 0.001
		T12	0.628	< 0.001
Muscle mass, left leg	DMG volume	T10	0.594	< 0.001
		T11	0.629	< 0.001
		T12	0.594	< 0.001
	DMG area	T10	0.601	< 0.001
		T11	0.494	< 0.001
		T12	0.601	< 0.001
Muscle mass, right arm	DMG volume	T10	0.540	< 0.001
		T11	0.677	< 0.001
		T12	0.540	< 0.001
	DMG area	T10	0.623	< 0.001
		T11	0.571	< 0.001
		T12	0.623	< 0.001
Muscle mass, right leg	DMG volume	T10	0.605	< 0.001
		T11	0.637	< 0.001
		T12	0.605	< 0.001
	DMG area	T10	0.621	< 0.001
		T11	0.497	< 0.001
		T12	0.621	< 0.001
Muscle mass, trunk	DMG volume	T10	0.568	< 0.001
		T11	0.705	< 0.001
		T12	0.568	< 0.001
	DMG area	T10	0.646	< 0.001
		T11	0.598	< 0.001
		T12	0.646	< 0.001

DMG: Dorsal muscle group; SF: Subcutaneous fat; VF: Visceral fat.

For fat component, BF showed moderate to strong positive relationship with SF area (r = 0.67) and VF area (r = 0.58-0.61) across the T10-T12 spinal levels (all P < 0.001). Conversely, a negative correlation was observed between BF and DMG density (r = -0.46 to -0.51, P < 0.001) (Table 1; Figure 1A, Figure 2). As for bone component, BMD exhibited little to no correlation with most BIA parameters, showing only a weak association with ASMI (r = -0.24, P = 0.099) (Figure 1A).

Morphomics and HGS

HGS demonstrated positive correlation with DMG volume at T10 (r = 0.48, P < 0.001), T11 (r = 0.40, P < 0.001) and T12 (r = 0.45, P < 0.001) and DMG area at T10 (r = 0.51, P < 0.001), T11 (r = 0.36, P = 0.011) and T12 (r = 0.35, P = 0.013) (Table 2; Figure 1B, Figure 2).

Table 2 Correlation between T10-12 morphomics and muscular and cardiorespiratory fitness.

Musclar/cardiorespiratory fitness	Morphomics	Spine level	Correlation	P value
Hand grip strength	DMG volume	T10	0.482	< 0.001
		T11	0.396	0.004
		T12	0.455	< 0.001
	DMG area	T10	0.514	< 0.001
		T11	0.358	0.01
		T12	0.350	0.01
Peak VO2	DMG volume	T10	0.504	< 0.001
		T11	0.483	< 0.001
		T12	0.466	< 0.001
	DMG area	T10	0.460	< 0.001
		T11	0.435	0.002
		T12	0.424	0.002
Peak WR	DMG volume	T10	0.435	0.002
		T11	0.430	0.002
		T12	0.436	0.002
	DMG area	T10	0.380	0.006
		T11	0.394	0.005
		T12	0.338	0.016
Peak VO2/BW	SF area	T10	-0.469	< 0.001
		T11	-0.457	< 0.001
		T12	-0.461	< 0.001
	VF area	T10	-0.290	0.041
		T11	-0.400	0.004
		T12	-0.436	0.002
	DMG density	T10	0.385	0.006
		T11	0.319	0.024
		T12	0.408	0.003
%predicted peak VO2	SF area	T10	-0.486	< 0.001
		T11	-0.478	< 0.001
		T12	-0.471	< 0.001
	VF area	T10	-0.335	0.017
		T11	-0.455	< 0.001
		T12	-0.464	< 0.001
	DMG density	T10	0.341	0.015
		T11	0.373	0.008
		T12	0.462	< 0.001
VAT	DMG area	T10	0.356	0.011
		T11	0.324	0.022
		T12	0.327	< 0.001
VAT WR	BMD	T10	0.341	< 0.001
		T11	0.328	< 0.001
		T12	0.330	< 0.001
VAT/BW	BMD	T10	0.374	0.007
		T11	0.361	0.01
		T12	0.314	0.026
	SF area	T10	-0.373	0.008
		T11	-0.383	0.006
		T12	-0.373	0.008
	VF area	T10	-0.430	0.002
		T11	-0.522	< 0.001
		T12	-0.520	< 0.001
EqO2nadir	SF area	T10	-0.350	0.013
		T11	-0.341	0.016
		T12	-0.334	0.018
Oxygen uptake efficiency slope	DMG volume	T10	0.449	0.001
		T11	0.483	< 0.001
		T12	0.509	< 0.001
	DMG area	T10	0.449	0.001
		T11	0.435	0.002
		T12	0.451	0.001

BMD: Bone mineral density; BW: Body weight; EqO₂: Ventilatory equivalent for O₂; VO₂: Oxygen consumption; VAT: Ventilatory anaerobic threshold; WR: Work rate; DMG: Dorsal muscle group; SF: Subcutaneous fat; VF: Visceral fat.

Morphomics and CPET

Morphomic analysis of the DMG highlighted a clear link between muscle mass and cardiopulmonary fitness. DMG area at T10-T12 showed positive correlation with several CPET variables, including peak VO₂ (r = 0.42-0.46, P < 0.001), peak WR (r = 0.34-0.39, P < 0.05), peak VO₂/BW (r = 0.32-0.39, P < 0.05), VAT (r = 0.32-0.36, P = 0.01-0.02) and oxygen uptake efficiency slope (OUES) (r = 0.43-0.45, P < 0.001). Besides, DMG volume showed similar positive associations with peak VO₂ (r = 0.47-0.50), peak WR (r = 0.43-0.44) and OUES (r = 0.45-0.51) (all P < 0.001) (Table 2; Figure 1B, Figure 2).

In contrast to muscular morphomics, the fat component from morphomics showed negative correlation with CPET variables. SF area exhibited a negative correlation with peak VO₂/BW (r = -0.46 to -0.47, P < 0.001), VAT/BW (r = -0.37 to -0.38, P < 0.001) and EqO₂ nadir (r = -0.33 to -0.35, P < 0.05). VF area demonstrated a similar correlation with peak VO₂/BW (r = -0.3 to -0.43, P < 0.05) and VAT/BW (r = -0.43 to -0.52, P < 0.001).

Furthermore, BMD demonstrated a positive correlation with VAT WR (r = 0.33-0.34, P = 0.02) and VAT/BW (r = 0.31-0.37, P < 0.05).

Morphomic analysis for predicting low CRF

Since all the morphomic parameters were derived from the same anatomical region, we tried to detect potential multicollinearity using variance inflation factors. We found little to moderate collinearity among all the eight morphomic predictors. (Supplementary Table 2) Therefore, all morphomic parameters were suitable for logistic regression analysis.

The multivariable logistic regression analysis indicated that a larger VF area significantly increased the likelihood of an VAT/BW ≤ 11 mL/min/kg (OR 5.01, 95%CI: 1.91-18.72, P = 0.0046). Conversely, higher BMD demonstrated a protective effect against low VAT/BW (OR 0.42, 95%CI: 0.17-0.90, P = 0.038). Similarly, although not reaching statistical significance, a higher DMG volume showed a trend toward an inverse association with low VAT/BW (OR 0.41, 95%CI: 0.15-0.94, P = 0.051) (Table 3).

Table 3 Full logistic regression model and parsimonious model after stepwise variable selection for ventilatory anaerobic threshold normalized by body weight ≤ 11 mL/kg/min with morphomics.

Predictor	Full model		Parsimonious model
	OR (95%CI)	P value	OR (95%CI)	P value
Bone mineral density	0.37 (0.14-0.85)	0.03	0.42 (0.17-0.90)	0.038
Subcutaneous fat area	0.94 (0.25-4.55)	0.93
Subcutaneous fat density	0.93 (0.26-2.97)	0.90
Visceral fat area	5.73 (1.39-33.37)	0.03	5.01 (1.91-18.72)	0.0046
Visceral fat density	1.17 (0.45-3.09)	0.75
Dorsal muscle group volume	0.27 (0.05-1.12)	0.10	0.41 (0.15-0.94)	0.051
Dorsal muscle group area	1.59 (0.38-7.52)	0.54
Dorsal muscle group density	1.22 (0.48-3.16)	0.67

OR: Odds ratio; CI: Confidence interval.

Subsequently, we employed a ROC analysis to examine the efficacy of morphomics in differentiating participants with an VAT/BW ≤ 11 mL/kg/min. The results indicated a robust performance, with an apparent AUC of 0.840 (95%CI: 0.72-0.96) and an optimism-corrected AUC of 0.778 (95%CI: 0.63-0.88, P < 0.001) (Figure 3). The model's sensitivity was 64.3%, its specificity was 78.7%, and it had a positive predictive value of 66.8% as well as a negative predictive value of 76.8%.

Figure 3 Receiver operator curve analysis for predicting patients with low cardiorespiratory fitness from the morphomics. The area under the receiver operator curve curves (AUC) was 0.840, and optimism-corrected AUC was 0.778 under the setting of 1000 bootstrap resamples internal validation (95% confidence interval 0.628-0.0.882, P < 0.001). AUC: Area under the receiver operator curve; ROC: Receiver operator curve.

DISCUSSION

The present study showed that T11 Level exhibits zero missing information in morphomics derived from CT in this esophageal cancer cohort. Body composition measurements obtained from morphomics are highly correlated with those from BIA, including muscle and fat composition, supporting their validity. In the muscle component, the DMG volume and area measured by morphomics exhibited strong positive correlations with BCM, FFM, LBM, SMM, and ASM obtained from BIA. Regarding the fat component, BF measured via BIA positively correlated with SF and VF areas but negatively with DMG density. Moreover, a moderate linear correlation is observed among morphomic parameters, CRF, and HGS. Specifically, BMD, VF area and DMG volume from morphomics were found to be predictive of low VAT/BW.

The measurement of core muscle size using morphomics at the lumbar level was initially introduced as a predictor of surgical outcomes[4]. Since then, most studies have focused on morphomic variables derived from a single lumbar vertebral level. However, research has demonstrated that CT-derived body composition assessments at the cervical, thoracic, and lumbar levels exhibit comparable reliability[7,8]. Furthermore, integrating CT imaging from multiple vertebral levels has been shown to improve correlation strength[22]. Given the limited availability of lumbar-level CT imaging in esophageal cancer patients and the need for more precise assessments, this study utilized multi-level morphomics to enhance feasibility and validity.

Muscle and fat are the two primary components of body composition, and numerous studies have compared different methods for their assessment. Previous research has demonstrated a positive correlation between muscle mass measured by CT-derived morphomics and BIA[6,7,23,24]. Consistent with these findings, our study revealed that the DMG area and DMG volume correlated positively with muscle-related components from BIA, including SMM, FFM, LBM, BCM, and ASM. Similarly, previous studies have reported comparable fat measurements between morphomics and BIA[7,8,22,23,25]. Our findings align with these reports, demonstrating a moderate to strong positive correlation between BF and both SF area and VF area.

Beyond muscle and fat mass, we also investigated the relationship between muscle quality and fat composition. Our results indicated a negative correlation between BF and DMG density, suggesting that higher fat accumulation may be associated with less dense muscle. These findings highlight the added value of morphomics in providing deeper insights into tissue integrity beyond traditional body composition assessments.

Low HGS is associated with increased hospitalization rates, poor nutritional status, higher overall mortality, and diminished quality of life[26]. Previous studies have reported that muscle strength is positively correlated with muscle mass and inversely correlated with fat mass[27-30]. However, our study found that although HGS was positively associated with DMG volume and area, it did not show a significant correlation with SF and VF areas. These findings suggest that, in esophageal cancer patients, muscle strength is primarily influenced by muscle mass rather than fat mass.

CRF is a critical prognostic factor in esophageal cancer, and several studies have explored its relationship with body composition[12,14,31,32]. While previous research has consistently shown a positive correlation between muscle mass and CRF, findings regarding fat mass have been inconclusive[16,33,34]. In this cohort of esophageal cancer patients, we observed a positive correlation between DMG and key CRF indicators, including peak VO₂, peak WR, peak VO₂/BW, VAT, and OUES. Skeletal muscle is the primary site of VO₂ during exercise, and individuals with greater muscle mass demonstrate higher oxygen demand and utilization efficiency[35]. Additionally, muscle mass correlates with central circulatory factors such as blood volume and stroke volume, which are key determinants of maximal aerobic capacity[36]. Additionally, we found a negative correlation between SF and VF area and both peak VO₂/BW and VAT/BW. This could be attributed to the role of adipose tissue in promoting systemic inflammation through the secretion of pro-inflammatory cytokines, which can impair endothelial function and oxygen delivery[37,38].

Apart from soft tissues, BMD was positively associated with aerobic capacity, particularly VAT/BW. This finding aligns with previous evidence suggesting that individuals with higher BMD may exhibit improved mechanical efficiency and physical performance[39,40].

Low CRF, such as VAT/BW, has been associated with poor outcomes after major surgery[14,15]. In the present study, we utilized morphomic analysis to predict low CRF and found that low BMD and high VF area were associated with reduced VAT/BW. Although low DMG volume only showed a trend toward association with low VAT/BW, these findings underscore the significant role of the musculoskeletal system in CRF. Furthermore, we demonstrated that with only three morphomic variables could predict VAT/BW ≤ 11 mL/kg/min, with an AUC of 0.778, a sensitivity of 64.3%, and a specificity of 78.7% in the model.

While this study provides valuable insights, several limitations should be acknowledged. The limited sample size, lack of external validation for the prediction model of low VAT/BW, and the homogeneity of the Taiwanese cohort may restrict the generalizability of the findings. Additionally, as the participants were initially recruited for esophageal cancer evaluation, the results may not necessarily be applicable to other diseases or the general population. These limitations highlight the need for further research to validate and expand upon our findings.

CONCLUSION

In morphomics derived from CT in patients with esophageal cancer, T11 is the only level with complete data in the present cohort, making it the most applicable in clinical practice. Body composition measurements obtained from morphomics are highly correlated with those from BIA, supporting their validity. Moreover, morphomic variables demonstrated moderate correlations with HGS and CRF, while also showing potential for predicting low VAT/BW. Future prospective studies with larger and more diverse cohorts are warranted to validate and implement morphomics-guided assessment in surgical planning.

ACKNOWLEDGEMENTS

The authors acknowledge the administrative support and statistical assistance provided by the Morphomic Analysis Group at the University of Michigan, Ann Arbor, United States.