Prediction of late-pregnancy pelvic floor dysfunction: a nomogram model based on proton density fat fraction and synthetic magnetic resonance imaging Authors: Yanni Zeng, Juanting Wang, Haiyan Shi, Ruomi Guo, Xueying Zhao, Yan Li, Yu Zhou, Hongsheng Wu, Yong Hong Published: 2026 ABSTRACT AbstractBackgroundPelvic floor dysfunction (PFD) is a highly prevalent condition that significantly compromises women’s quality of life. To date, the majority of studies and clinical rehabilitation interventions have focused on the assessment and management of pelvic floor muscle (PFM) function in the postpartum period, whereas investigations into the monitoring of pelvic floor function during the third trimester of pregnancy remain relatively understudied. This study aimed to develop a radiology-clinical nomogram integrating proton density fat fraction (PDFF; Iterative Decomposition of water and fat with Echo Asymmetry and Least-squares estimation, IDEAL-IQ sequence), synthetic magnetic resonance imaging (MAGnetic Resonance Image Compilation, MAGiC sequence), and diffusion tensor imaging (DTI) for predicting PFD in late pregnancy.MethodsA total of 114 third-trimester pregnant women were classified into PFD and non-PFD groups. Fat fraction (FF), T1, T2, proton density (PD), and fractional anisotropy (FA) of bilateral levator ani muscles (LAMs) were measured from IDEAL-IQ, MAGiC, and DTI by two radiologists (with 15 years of experience). Measurement reliability was assessed using intraclass correlation coefficients (ICCs). Univariate analysis screened potential predictors [age, gestational week, gravida, parity, delivery mode, fetal weight, pre- and post-pregnancy body mass index (BMI), and quantitative imaging parameters]. Significant variables were included in a multivariate logistic regression to build the nomogram. Model performance was assessed using receiver operating characteristic (ROC) curves, calibration plots, and decision curve analysis (DCA).ResultsOf 114 participants, 52 were assigned to the PFD group and 62 to the non-PFD group. Univariate analysis showed significant between-group differences in gestational week (P=0.025), gravida (P=0.032), parity (P=0.023), fetal weight (P<0.001), left LAM FF (L-LAM FF, P=0.005), and right LAM T2 (R-LAM T2, P=0.004). The final Akaike information criterion (AIC)-selected multivariable prediction model, gravida, L-LAM FF, and R-LAM T2 were retained as nomogram predictors. The nomogram achieved moderate discrimination with an area under the curve (AUC) of 0.726 [95% confidence interval (CI): 0.632–0.815]. Calibration indicated good model fit, and DCA demonstrated positive net benefit within a threshold probability range of 25–83%. Interobserver agreement was moderate/excellent (ICC 0.49–0.86). FA was only an exploratory imaging biomarker and was not retained in the final combined prediction model.ConclusionsThe nomogram incorporating gravida, L-LAM FF, and R-LAM T2 effectively predicts PFD in late pregnancy.Keywords: Pelvic floor dysfunction (PFD), third trimester, MAGnetic Resonance Image Compilation (MAGiC), Iterative Decomposition of water and fat with Echo Asymmetry and Least-squares estimation (IDEAL-IQ), quantitative imaging Introduction Pelvic floor dysfunction (PFD), encompassing urinary incontinence (UI), pelvic organ prolapse (POP), and defecatory dysfunction (1), is a highly prevalent disorder that substantially impairs women’s quality of life. Pregnancy and childbirth are well-established risk factors for PFD development (2,3). As the principal biomechanical support of the pelvic floor, the levator ani muscle (LAM) is particularly susceptible to injury from prolonged gestational load and the mechanical stretch of vaginal delivery (4). Epidemiological studies indicate that levator ani avulsion, present in up to 55% of patients with POP, is a significant risk factor for PFD (5). Therefore, accurate antenatal identification of women at high risk for PFD is crucial for early counseling, risk stratification, and preventive strategies. Current clinical tools for evaluating pelvic floor function include physical examination, ultrasonography, static/dynamic magnetic resonance imaging (MRI), and physiologic tests (6-9). However, many of these modalities are operator dependent, challenging to implement during late pregnancy, or lack objective and quantitative biomarkers for early muscular degeneration detection. Although conventional MRI provides excellent anatomic description, it lacks quantitative metrics for early deterioration and cannot distinguish early fat infiltration from muscle atrophy. Recent quantitative MRI techniques, such as synthetic MRI (MAGnetic Resonance Image Compilation, MAGiC), proton density fat fraction (PDFF; Iterative Decomposition of water and fat with Echo Asymmetry and Least-squares estimation, IDEAL-IQ), and diffusion tensor imaging (DTI), show promise for microstructure characterization but have not been integrated into routine PFD assessment. MAGiC enables a rapid, single acquisition that replaces multiple conventional sequences, providing simultaneous quantification of T1, T2, and PD with reduced examination time and improved reproducibility (10,11). IDEAL-IQ is validated for fat quantification (12-14), and DTI evaluates muscle fiber continuity and microstructural integrity by quantifying the directional anisotropy of water diffusion within tissues (15,16). Despite these advances, the potential of multiparametric quantitative MRI for assessing pelvic floor muscle (PFM) condition remains largely unexplored, highlighting the need for noninvasive, quantitative tools to assess PFM status pre-delivery. Pregnancy- and delivery-related tensile injury disrupts the ordered architecture of pelvic floor myofibers. Accordingly, we hypothesize that quantitative metrics including PDFF, fractional anisotropy (FA), and T1, T2, and proton density (PD) values can act as biomarkers to evaluate microstructural alterations of PFMs. Therefore, this study aimed to quantitatively evaluate third-trimester PFM integrity using multiparametric quantitative MRI, to identify imaging markers associated with antenatal PFM function. We present this article in accordance with the TRIPOD reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1-0235/rc). Methods Study population and clinical assessments This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committees of The Third Affiliated Hospital of Sun Yat-sen University (No. II2023-007-01) and Huadu District People’s Hospital of Guangzhou (No. 2026-0027). The requirement for informed consent was waived due to the retrospective nature of this study. Only de‑identified retrospective clinical and MRI data were analyzed in this work. This retrospective study was conducted at Huadu District People’s Hospital of Guangzhou between January and October 2025. A total of 123 pregnant women (gestational weeks 28–40) who underwent MRI examination were enrolled based on predefined inclusion and exclusion criteria. The chief complaints of these patients included constipation, stress urinary incontinence (SUI), pelvic floor discomfort, and lower back pain. Patients were allocated to the PFD group according to the presence of clinical symptoms (SUI or vaginal wall prolapse) and confirmed abnormalities in PFM strength as assessed by the Oxford Grading System. In contrast, those without these features were assigned to the non-PFD group. All pelvic floor evaluations were completed by one gynecologist with 10 years of experience in PFD. This study used the 6-point Modified Oxford Grading Scale (MOGS), and a score ≤ 3 was defined as abnormal PFM function (17). The scale ranges from 0 to 5, with clear definitions for each grade: 0= no contraction, 1= flicker, 2= weak, 3= moderate, 4= good (with lift), and 5= strong. Higher scores reflect stronger contractile strength of the PFMs. Participants lay supine with hips and knees flexed. They were instructed on proper pelvic floor contraction before assessment. The examiner placed two fingers in the anterior one-third of the vagina and required participants to maximally contract and elevate the PFMs. Three 5-second maximal contractions were performed at 30-second intervals, and the best MOGS score from three trials was recorded as the final result. The clinical assessor undertaking MOGS grading was blinded to MRI data, and radiologists analyzing MRI parameters were blinded to all clinical grouping information. Exclusion criteria included MRI contraindications, pre-pregnancy body mass index (BMI) >30 kg/m2, pelvic tumors, prior pelvic floor surgery, and incomplete/poor-quality images, and SUI patients with a MOGS score >3. After exclusions, 114 women were analyzed: PFD group (n=52) and non-PFD group (n=62). Participant selection is detailed in Figure 1. Flowchart of the study datasets. AUC, area under the curve; BMI, body mass index; MRI, magnetic resonance imaging; PFD, pelvic floor dysfunction; ROC, receiver operating characteristic. MRI data acquisition MRI was performed on a 3.0 T scanner (SIGNA™ Architect, GE Healthcare, Milwaukee, WI, USA) using a 30-channel body coil. Participants were scanned supine with legs slightly flexed after bladder emptying; no contrast was used. Parameters for each sequence are shown in Table 1. Table 1MRI scan sequence and parametersVariableIDEAL-IQMAGiCDTITR (ms)9.74,0008,000TE (ms)3.512.373FOV (mm2)320×220320×320320×220Thickness (mm)333Spacing (mm)000NEX111Slices363636Echoes681Flip angle (°)49090Bandwidth (kHz)111.1162.5250Voxel size (mm3)2.0×2.0×3.01.0×1.2×3.02.5×2.5×3.0Scan time1 min 17 s4 min 0 s2 min 32 sDTI, diffusion tensor imaging; FOV, field of view; IDEAL-IQ, Iterative Decomposition of water and fat with Echo Asymmetry and Least-squares estimation; MAGiC, MAGnetic Resonance Image Compilation; MRI, magnetic resonance imaging; NEX, number of excitations; TE, echo time; TR, repetition time. MRI scan sequence and parameters DTI, diffusion tensor imaging; FOV, field of view; IDEAL-IQ, Iterative Decomposition of water and fat with Echo Asymmetry and Least-squares estimation; MAGiC, MAGnetic Resonance Image Compilation; MRI, magnetic resonance imaging; NEX, number of excitations; TE, echo time; TR, repetition time. IDEAL-IQ was applied to perform quantitative water‑fat decomposition and PDFF quantification. This chemical-shift‑based multi‑echo spoiled gradient‑recalled echo sequence samples progressive phase disparities between water and fat signals across multiple distinct echo times. Briefly, IDEAL-IQ applies iterative decomposition of water and fat with echo asymmetry and least-squares estimation, incorporating multi-peak fat spectrum modeling and R2* correction to separate water and fat components and to generate quantitative maps. IDEAL-IQ was applied to perform quantitative water‑fat decomposition and PDFF quantification. This chemical-shift‑based multi‑echo spoiled gradient‑recalled echo sequence samples progressive phase disparities between water and fat signals across multiple distinct echo times. Briefly, IDEAL-IQ applies iterative decomposition of water and fat with echo asymmetry and least-squares estimation, incorporating multi-peak fat spectrum modeling and R2* correction to separate water and fat components and to generate quantitative maps. MAGiC is based on a multi-dynamic multi-echo sequence, in which images are acquired at multiple saturation/inversion delay times and multiple echo times within a single scan. This acquisition provides multi-contrast signal evolution that enables simultaneous estimation of tissue relaxation parameters. Image reconstruction was performed using the MAGiC reconstruction software on the scanner console. The reconstruction algorithm fits the acquired multi-dynamic multi-echo signal intensities to a relaxation model to estimate voxel-wise T1 relaxation time, T2 relaxation time, and proton density. Based on these quantitative maps, synthetic T1-weighted, T2-weighted, proton density-weighted, and fluid-attenuated inversion recovery (FLAIR) images were generated by specifying virtual repetition time (TR), echo time (TE), and inversion time (TI) parameters. MAGiC is based on a multi-dynamic multi-echo sequence, in which images are acquired at multiple saturation/inversion delay times and multiple echo times within a single scan. This acquisition provides multi-contrast signal evolution that enables simultaneous estimation of tissue relaxation parameters. Image reconstruction was performed using the MAGiC reconstruction software on the scanner console. The reconstruction algorithm fits the acquired multi-dynamic multi-echo signal intensities to a relaxation model to estimate voxel-wise T1 relaxation time, T2 relaxation time, and proton density. Based on these quantitative maps, synthetic T1-weighted, T2-weighted, proton density-weighted, and fluid-attenuated inversion recovery (FLAIR) images were generated by specifying virtual repetition time (TR), echo time (TE), and inversion time (TI) parameters. DTI characterizes tissue microarchitecture by quantifying the magnitude and directionality of water diffusion. By modeling diffusion as a tensor, DTI provides indices of diffusion anisotropy and principal diffusion orientation, which can be used to assess the organization and integrity of fiber-like tissue structures. In this study, DTI was acquired with one non-diffusion-weighted image at b=0 s/mm2 and diffusion-weighted images at b=1,000 s/mm2 along 15 non-collinear diffusion-encoding directions. DTI data were preprocessed with correction for motion and eddy-current distortions, followed by PFM masking and diffusion tensor fitting. FA maps were then generated from the eigenvalues of the fitted diffusion tensor. Regions of interest (ROIs) were manually delineated on co-registered anatomical images or b0 images and subsequently applied to the FA maps to extract quantitative FA values. DTI characterizes tissue microarchitecture by quantifying the magnitude and directionality of water diffusion. By modeling diffusion as a tensor, DTI provides indices of diffusion anisotropy and principal diffusion orientation, which can be used to assess the organization and integrity of fiber-like tissue structures. In this study, DTI was acquired with one non-diffusion-weighted image at b=0 s/mm2 and diffusion-weighted images at b=1,000 s/mm2 along 15 non-collinear diffusion-encoding directions. DTI data were preprocessed with correction for motion and eddy-current distortions, followed by PFM masking and diffusion tensor fitting. FA maps were then generated from the eigenvalues of the fitted diffusion tensor. Regions of interest (ROIs) were manually delineated on co-registered anatomical images or b0 images and subsequently applied to the FA maps to extract quantitative FA values. Image analysis Analysis was conducted on an AW4.7 workstation (GE Healthcare) using Volume Viewer software. Quantitative MAGiC parameters were measured using the built-in console module. Two radiologists (15 years’ experience each) evaluated bilateral LAM, measuring FF, FA, T1, T2, and PD. Image reconstruction was performed using the IDEAL-IQ algorithm on the scanner console/workstation. The reconstructed outputs included water-only images, fat-only images, in-phase images, out-of-phase images, PDFF maps, and T2* maps. Two consistent ellipsoid ROIs per side were placed on two axial slices, avoiding adjacent structures to minimize artifacts (Figure 2). The mean from four ROIs per side was recorded. Comparison of LAM fat fraction values between two representative patients. Patient 1 (A,B): the fat content of bilateral LAM in a late pregnant woman with PFD (35 weeks, G5P3) was measured by delineating/positioning the ROIs on IDEAL-IQ. L-LAM FF =85.09%, R-LAM FF =56.31%. Patient 2 (C,D): pregnant woman without PFD (33 weeks, G3P1), L-LAM FF =16.53%, R-LAM FF =14.36%. FF, fat fraction; IDEAL-IQ, Iterative Decomposition of water and fat with Echo Asymmetry and Least-squares estimation; L-LAM, left levator ani muscle; LAM, levator ani muscle; PFD, pelvic floor dysfunction; R-LAM, right levator ani muscle; ROI, region of interest. For T1, T2, and PD (from MAGiC maps) values, ROIs were manually drawn along the inner LAM contours on high-resolution T2-weighted images. The contour was copied and the corresponding ROI was pasted onto the quantitative maps (Figure 3). DTI sequences, T2-weighted images, and FA maps were linked. After delineating the inner contour of the LAM on T2-weighted images, FA values within the corresponding ROI were extracted from the matched FA maps. Two independent measurements per side were averaged. ROIs were manually delineated along the inner contours of the left and right LAM. Patient 1 (A-D): pregnant woman with PFD (35 weeks, G5P3), L-LAM T2 =109 ms, R-LAM T2 =98 ms. Patient 2 (E-H): pregnant woman without PFD (33 weeks, G3P1), L-LAM T2 =62 ms, R-LAM T2 =61 ms. LAM, levator ani muscle; L-LAM, left levator ani muscle; PFD, pelvic floor dysfunction; R-LAM, right levator ani muscle; ROI, region of interest. Statistical analysis Analyses were conducted using R software (v4.1.3, R Foundation for Statistical Computing, Vienna, Austria). Missing data were handled by multiple imputation (“Multivariate Imputation by Chained Equations, mice” package). Data are presented as mean ± standard deviation (SD) for normally distributed continuous variables and median (interquartile range, IQR; Q1, Q3) for non-normally distributed continuous variables. Univariate group comparisons employed the chi-square test for categorical variables, the independent samples t-test for normally distributed continuous variables, and the Mann-Whitney U test for non-normally distributed continuous variables. Reproducibility was evaluated using the intraclass correlation coefficient (ICC). The corresponding classification criteria were set as: poor (ICC ≤0.40), fair (0.40< ICC <0.59), good (0.59≤ ICC <0.75), and excellent (ICC ≥0.75). All ICC measurements were acquired at the identical anatomical location for each patient. Multivariate logistic regression used the “glm” function with backward elimination based on the minimum Akaike information criterion (AIC). Model discrimination was evaluated via receiver operating characteristic (ROC) analysis (“pROC” package), reporting the area under the curve (AUC). Calibration was assessed using the “calibration” function (“rms” package) and the Hosmer-Lemeshow test. Clinical utility was estimated via decision curve analysis (DCA, “ggDCA” package). Results Participant characteristics and univariate analysis This study included 114 third-trimester pregnant women. Univariate analysis of 18 potential risk factors [age, gestational week, gravida, parity, delivery pattern, fetal weight, pre- and post-pregnancy BMI, and imaging parameters (FF, FA, T1, T2, PD) of the bilateral LAM] was conducted between the PFD and non-PFD groups. Seven factors demonstrated statistically significant differences (P<0.05, Table 2): three imaging parameters (L-LAM FF, L-LAM T2, R-LAM T2) and four clinical indicators (gestational week, gravida, parity, fetal weight). Table 2Univariate analysis of risk factors between the PFD group and the non-PFD groupCharacteristicAll patients (n=114)Non-PFD group (n=62)PFD group (n=52)χ2/t/UP valueAge (years)32.00 (28.25–35.00)32.00 (29.00–35.00)32.50 (28.00–36.00)0.3720.542Gestational week32.00 (30.00–34.00)32.00 (29.25–34.00)33.00 (31.00–35.25)5.0590.025Gravida3.00 (1.25–4.00)2.00 (1.00–3.00)3.00 (2.00–4.00)4.6150.032Parity1.00 (0.00–2.00)1.00 (0.00–2.00)1.00 (0.75–2.00)5.1490.023Fetal weight (g)2,163.30 (1,611.26–2,650.57)1,964.59 (1,406.66–2,403.76)2,412.19 (1,892.38–2,774.14)11.613<0.001BMI (kg/m2) Pre-pregnancy21.55 (19.35–24.03)21.63 (19.27–23.41)21.48 (19.99–25.85)1.2560.262 Post-pregnancy26.20 (23.93–28.96)26.10 (24.03–28.46)26.39 (23.42–29.63)0.7090.4FF (%) R-LAM7.10 (4.11–11.77)7.05 (3.87–10.60)7.21 (4.13–13.33)1.1260.289 L-LAM8.77 (4.70–17.35)6.99 (4.29–12.79)12.41 (5.67–22.68)8.0110.005FA R-LAM0.42 (0.36–0.46)0.43 (0.37–0.47)0.41 (0.34–0.45)2.620.106 L-LAM0.42±0.070.43±0.080.41±0.071.6480.202R-LAM (ms) T1924.00 (801.50–1,055.75)893.00 (797.00–1,008.00)965.00 (832.00–1,115.75)1.9110.167 T277.00 (71.25–84.00)74.00 (70.25–81.00)80.00 (74.75–89.00)8.4810.004 PD74.00 (69.73–77.40)73.50 (69.03–75.65)74.90 (70.07–79.30)2.2730.132L-LAM (ms) T1887.00 (776.50–1,038.25)869.00 (770.50–1,033.25)918.50 (805.75–1,050.75)0.7880.375 T278.00 (72.00–88.00)76.00 (72.00–84.75)82.50 (74.75–89.25)6.2460.012 PD68.20 (64.62–74.55)67.65 (64.43–70.65)69.55 (64.70–74.90)0.8340.361Natural delivery0.2320.630 No83 (72.81)44 (70.97)39 (75.00) Yes31 (27.19)18 (29.03)13 (25.00)Data are presented as median (interquartile range) or n (%). BMI, body mass index; FA, fractional anisotropy; FF, fat fraction; L-LAM, left levator ani muscle; PD, proton density; PFD, pelvic floor dysfunction; R-LAM, right levator ani muscle. Univariate analysis of risk factors between the PFD group and the non-PFD group Data are presented as median (interquartile range) or n (%). BMI, body mass index; FA, fractional anisotropy; FF, fat fraction; L-LAM, left levator ani muscle; PD, proton density; PFD, pelvic floor dysfunction; R-LAM, right levator ani muscle. Multivariate logistic regression and nomogram development The seven significant indicators were included in a multivariate logistic regression model. Following AIC-based backward selection, three variables were retained in the final prediction model: gravida, L-LAM FF, and R-LAM T2. Among these variables, L-LAM FF reached conventional statistical significance, whereas gravida and R-LAM T2 showed borderline associations. A nomogram was constructed by converting the regression coefficients (β values) from the final model into a visual scoring system for individualized PFD risk prediction (Figure 4). The β values, odds ratios (ORs), and corresponding 95% confidence intervals (CIs) are detailed in Table 3. The integrated nomogram incorporates the gravida, R-LAM T2, and L-LAM FF. Sum the points of three variables to match the individual PFD risk probability on the total points scale. FF, fat fraction; L-LAM, left levator ani muscle; PFD, pelvic floor dysfunction; R-LAM, right levator ani muscle. Table 3Multivariate logistic regression analysis of risk factors between two groupsCharacteristicBSEORCIZP valueGravida0.2360.1251.270.99-1.621.8930.058R-LAM T20.0360.0181.041-1.071.9450.052L-LAM FF0.0620.0241.061.02-1.122.5670.01B, regression coefficient; CI, confidence interval; FF, fat fraction; L-LAM, left levator ani muscle; OR, odds ratio; R-LAM, right levator ani muscle; SE, standard error. Multivariate logistic regression analysis of risk factors between two groups B, regression coefficient; CI, confidence interval; FF, fat fraction; L-LAM, left levator ani muscle; OR, odds ratio; R-LAM, right levator ani muscle; SE, standard error. Model validation Model performance was assessed by discrimination (ROC curves), calibration, and clinical utility (DCA). The ROC analysis for predicting PFD yielded an AUC of 0.726 (95% CI: 0.632–0.815; Figure 5). The calibration plot, based on 500 bootstrap repetitions, showed close alignment between predicted and observed probabilities (Figure 6). The calibration curve was plotted to assess the consistency between predicted and actual probabilities. The concordance index (C-index) was 0.733, indicating moderate discrimination ability. The logistic calibration intercept and slope were 0.000 and 1.000 (apparent), respectively. DCA indicated a positive net benefit for the model within a threshold probability range of 25% to 83% (Figure 7). ROC curve for predicting late-pregnancy PFD using nomogram, the AUC is 0.726 (95% CI: 0.632–0.815). AUC, area under the curve; CI, confidence interval; PFD, pelvic floor dysfunction; ROC, receiver operating characteristic. Calibration plot: prediction model exhibited moderate discrimination and calibration performance. ROC, receiver operating characteristic. DCA curve: positive net benefits across a 25–83% threshold probability. DCA, decision curve analysis. Consistency of measurements Excellent agreement (ICC ≥0.75) was observed in four parameters: R-LAM FF (ICC =0.84), L-LAM FA (ICC =0.76), R-LAM PD (ICC =0.86), and L-LAM PD (ICC =0.83). Good agreement (0.59≤ ICC <0.75) was found in four parameters: R-LAM FA (ICC =0.66), R-LAM T1 (ICC =0.64), R-LAM T2 (ICC =0.72), and L-LAM T2 (ICC =0.62). Only L-LAM FF (ICC =0.49) and L-LAM T1 (ICC =0.54) showed fair agreement (0.40< ICC <0.59). Overall, interobserver agreement for LAM quantitative MRI parameters was predominantly good to excellent, indicating acceptable reproducibility of the measurements. Discussion A nomogram for noninvasive prediction of third-trimester PFD risk was successfully developed and internally validated. The model demonstrated moderate discriminative performance (AUC: 0.726), moderate calibration, and positive clinical utility across threshold probabilities of 25–83%. Based on prior literature (18,19), we evaluated potential risk factors including age, gestational week, gravida, parity, delivery pattern, fetal weight, BMI, and bilateral LAM parameters (FF, FA, T1, T2, PD). By integrating clinical variables and bilateral LAM quantitative MRI parameters, multivariate logistic regression identified three nomogram predictors—gravida, L-LAM FF, and R-LAM T2—allowing individualized and visual risk estimation for early screening and targeted intervention. The imaging biomarkers L-LAM FF and R-LAM T2 were critical for accurate risk estimation. Univariate analysis showed significantly higher L-LAM FF in the PFD group, and it remained a nomogram predictor. Pregnancy and vaginal delivery are established triggers for PFD (20,21). The PFMs consist predominantly of type I slow-twitch muscle fibers, with a minor component of type II fast-twitch muscle fibers—the former constituting approximately 70–80% of the LAM fibers. Hamrick et al. (22) demonstrated that with the decline in muscle function, type I muscle fibers accumulate more intramuscular fat than type II muscle fibers. Therefore, intramuscular fat infiltration within the PFMs may serve as an indicator of mechanical muscular injury, denervation, or degeneration of the muscle (23). FF, quantified via IDEAL-IQ, enables early detection of this pathological hallmark (24,25). In addition, our finding of elevated FF specifically in the left LAM (PFD vs. non-PFD; right side P=0.289) may reflect asymmetric mechanical stress from fetal descent, often in the left occiput anterior (LOA) position (26), justifying focused clinical monitoring T2 value reflects water mobility and microenvironmental homogeneity. Under resting conditions, water molecules in the extracellular space and intracellular compartment of normal skeletal muscle exist predominantly as free water and bound water, respectively. Increased intracellular water content in skeletal muscle elicits alterations in the tissue’s T2 relaxation time (27). Esposito et al. (28) demonstrated a positive correlation between the T2 value and two core metrics for assessing skeletal muscle injury: the number of inflammatory leukocytes (CD45+) and the scores of tissue injury and inflammatory severity. This finding indicates that the T2 value can reflect the inflammatory expression profile of injured skeletal muscle (29). It can detect pathological and physiological changes that are invisible on conventional MRI (30). In our cohort, elevated LAM T2 values were associated with PFD, advanced gestational age, and higher fetal weight. These findings align with the known mechanisms whereby late-pregnancy hormonal changes and mechanical stress induce edema, fiber disruption, and perimysial disruption, which widen inter-fascicular spaces, increase free water, and elevate T2 values (31,32). Additionally, Kim et al. noted that T2 values correlate with FF (33). Fat infiltration, edema, and inflammation in skeletal muscle all exert an impact on its T2 value, with potential mutual interaction and masking effects between fat infiltration and T2 value, consistent with our observation of concurrent L-LAM FF and T2 elevation in PFD. Notably, although univariate analysis implicated bilateral LAM T2 and L-LAM FF, only R-LAM T2 and L-LAM FF were retained as nomogram predictors. This selection reflects an integration of statistical rigor with the anatomical and biological specificity of the imaging biomarkers. In late pregnancy, bilateral T2 elevation arises from hormonally driven fluid accumulation (31). Given the marked correlation between FF and T2 (33), L-LAM FF and T2 exhibited substantial collinearity. Wang et al. (34) demonstrated that PDFF serves as a valuable biomarker for evaluating PFM dysfunction. This parameter reflects fat infiltration and myogenic microstructural changes, which are consistent with the pathogenic hypothesis of PFD (35). In the multivariable model, L-LAM FF was retained because it demonstrated greater statistical sensitivity to chronic structural alterations, consistent with the above study. From an anatomical perspective, chronic structural lesions including fatty infiltration in the LAM during pregnancy may be associated with sustained mechanical compression on the left LAM caused by the LOA fetal position, which is the most common presentation in obstetric practice (22,36). Biologically, although fat infiltration and edema coexist in the third trimester, elevated right LAM T2 may better reflect acute functional perturbations (e.g., edema, inflammation) rather than chronic structural change. A plausible hypothesis is that chronic left-sided damage elicits compensatory, acute functional adaptation in the right side, sharing load via edema and widened intermuscular spaces (elevated T2). This complementary pattern—chronic structural alteration on the left and acute functional perturbation on the right—may enhance discriminatory performance and aligns with evidence of asymmetric pelvic floor loading and injury in pregnancy (37,38). Accumulating clinical evidence (31,37) has firmly established that LAM injury presents an asymmetric pattern in both postpartum populations and patients with SUI. However, limited research has focused on LAM asymmetry in pregnant women. Our study detected prominent asymmetry in quantitative parameters between the bilateral LAM of pregnant participants. Given the currently unclear pathogenesis underlying this phenomenon, further investigations are urgently required. Univariate analysis identified gestational week, fetal weight, gravidity, and parity as significant risk factors. Higher gravidity/parity is established risk for disorders such as SUI and POP (39,40). Fetal weight showed univariate significance (P<0.001) but was excluded from the multivariate model, likely due to collinearity with other factors that synergistically increase intra-abdominal pressure. Gravidity was the only clinical variable retained as a nomogram factor, underscoring the substantial impact of pregnancy on pelvic floor integrity (3). Although vaginal delivery is recognized as a risk factor (2), no significant difference in delivery pattern was observed between groups, potentially reflecting postpartum recovery of PFMs. Pessoa et al. (41) reported LAM injury in 11–42% of primiparous women postpartum, influenced by fetal weight, labor duration, and mode. PFM strength recovers to ~70% at 6 months and 85–90% at 12 months postpartum (42). The limited number of participants with prior vaginal delivery (n=31/114) may have reduced statistical power for this variable. Any further subgrouping based on number of previous vaginal births would render these subgroups statistically underpowered. FA reflects the anisotropic characteristics of microstructural organization within muscle fibers, thereby indirectly indicating their structural integrity. The lack of significant FA differences in this study may be attributed to LAM’s complex anatomy, with interwoven fibers poorly delineated on MRI. Elevated abdominal pressure in pregnancy stretches and thins the muscle, potentially reducing diffusion contrast sensitivity, aligning with Zijta et al. (43). However, Schmid et al. (16) showed that FA fails to effectively reflect functional changes in the LAM. This discrepancy may be attributed to the fact that the participants in this study were pregnant women in the third trimester, whose PFD were primarily manifested as functional changes without significant structural abnormalities of muscle fibers. Our predictive model followed a rigorous workflow and was validated across discrimination, calibration, and clinical utility. Routine clinical variables only reflect obstetric features, whereas quantitative MRI can detect early microscopic lesions of LAM that cannot be found by clinical evaluation, thus providing extra predictive information. A key innovation is the focus on the third-trimester window, enabling “early warning” via quantitative biomarkers (L-LAM FF, R-LAM T2) that reflect microscopic structural changes, offering greater objectivity and earlier detection than clinical symptoms alone. Persistent and compliant rehabilitation training enables full recovery of PFM function (33). Hence, our results offer practical guidance for pelvic floor rehabilitation in late pregnancy. Interobserver agreement for L-LAM FF only reached a fair level owing to several inherent constraints. First, the LAM was remarkably thinned and atrophic among third-trimester participants, some of whom exhibited severe intramuscular fatty replacement. As the IDEAL-IQ sequence featured high sensitivity to adipose tissue, voxels encompassing atrophic muscle bundles contained mixed muscle and fat tissue; subtle positional deviations during ROI delineation could alter FF value. Second, small ROIs mitigate peripheral fat contamination but reduce tissue sampling representativeness. In addition, a single outlier case with extensive fatty infiltration further diminished the overall inter-rater concordance. Future investigations could implement high-resolution thin-slice MRI sequences and standardized semi-automated segmentation pipelines to minimize measurement bias arising from subjective inter-observer variability. Several limitations of this study should be acknowledged. First of all, it was a single-center study with a relatively small sample size (n=114), and the events-per-variable (EPV) ratio was approximately 7.4. Secondly, only internal validation was performed, and formal quantitative comparisons among clinical‑only, imaging‑only, and combined prediction models using delta‑AUC, net reclassification index (NRI), integrated discrimination improvement (IDI), calibration assessment, and DCA were not performed in this study. Thirdly, the direct correlation between imaging markers and PFM function (e.g., electromyography, muscle strength tests) was not explored. Fourth, we conducted 18 distinct univariate regression analyses to screen candidate predictors, with no formal correction for multiple comparisons (e.g., Bonferroni or Benjamini-Hochberg adjustment) implemented. Future directions include: (I) multi-center prospective studies for model optimization and external validation to verify our conclusions and improve the generalizability of the model. Calibration regression will be performed on independent external datasets to derive external‑validation calibration intercept and slope for objective assessment of model calibration across diverse populations; (II) integration of multimodal imaging (e.g., dynamic MRI) for improved accuracy; (III) long-term follow-up to assess predictive value for postpartum PFD; and (IV) future investigations may either predefine a concise set of candidate predictors or employ formal multiplicity correction for exploratory variable screening. Conclusions The developed nomogram, which integrates gravida, L-LAM FF, and R-LAM T2 value, demonstrates moderate discriminative predictive performance for prenatal PFD in late pregnancy. This is a preliminary model that has undergone only internal validation, and further external validation is therefore required. It serves as an internally validated clinical tool to guide decision-making in antenatal screening and risk stratification for PFD. Supplementary The article’s supplementary files as Acknowledgments The authors would like to thank all the participants of the present study and the workers in The Third Affiliated Hospital of Sun Yat-sen University and Huadu District People’s Hospital of Guangzhou. Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committees of The Third Affiliated Hospital of Sun Yat-sen University (No. II2023-007-01) and Huadu District People’s Hospital of Guangzhou (No. 2026-0027). The requirement for informed consent was waived due to the retrospective nature of this study. Only de‑identified retrospective clinical and MRI data were analyzed in this work. Footnotes Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1-0235/rc Funding: None. Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1-0235/coif). X.Z. is an employee of GE Healthcare, Beijing. The other authors have no conflicts of interest to declare. Data Sharing Statement Available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1-0235/dss References Associated Data Supplementary Materials The article’s supplementary files as Data Availability Statement Available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1-0235/dss