Assessing Postpartum Levator Ani Muscle Recovery: A Feasibility Study on Automated Texture Analysis in Transvaginal Ultrasound.

objectiveLevator ani (LA) muscle injury during childbirth is an important contributor to the development of pelvic floor disorders (PFDs). Understanding the mechanisms of LA muscle damage and healing is crucial for developing individualized recovery plans and preventive measures. Although muscle avulsion has been extensively studied, the internal rearrangement of muscle structure has not. Current evaluation by palpation or force measurement can be limited due to pain and motivation. An objective evaluation of internal muscle structure has the advantage of being independent of these subjective factors. The objective of this study was to evaluate the feasibility of using 3-D transvaginal ultrasound (US) imaging combined with quantitative texture analysis for assessing deep LA muscle injuries and monitoring tissue microstructural changes involved in healing over time in postpartum women. By providing insights into recovery mechanisms, this approach has the potential to track muscle repair and quantify the progression of tissue healing.

methodsThis is a secondary analysis of 3-D transvaginal US images performed on women who were at high risk for LA muscle injuries following vaginal delivery but did not have muscle avulsion (n = 15), and on women who underwent cesarean section (C-section) before the second stage of labor (n = 10) as controls. Imaging data were collected at 2 wk, 6 wk and 6 mo postpartum for the vaginal delivery group, and at 6 wk and 6 mo for the C-section group. Texture "features" (93 in total) that result from image characteristics (i.e., local pixel-intensity differences and adjacency relationships) were extracted from the US images using six classical texture analysis techniques. Feature reduction was implemented by calculating the Pearson correlation coefficient (PCC) and removing features with high similarity (|PCC| >0.8). Three feature-selection methods were employed: linear mixed effects analysis of variance, receiver operator characteristic analysis and least absolute shrinkage and selection operator regression. Features selected by at least two methods were used to train a random Forest classification model, with the dataset split into 70% training and 30% testing sets.

resultsFour texture features were identified as potential imaging biomarkers associated with structural changes in the LA muscle during recovery. Average percentage changes of these four features (12.2% ± 2.3%) were observed between 2 and 6 wk postpartum in the vaginal delivery group. In contrast, smaller changes between 6 wk and 6 mo were noted in the vaginal delivery group and C-section group over the same period (6.0% ± 1.2% and 6.6% ± 2.3%, respectively). The random Forest model achieved a classification accuracy of 78% at distinguishing different time points based on the selected features.

conclusionUS image texture analysis detected systematic changes in muscle echotexture between 2 and 6 wk postpartum, recognized as a crucial period for healing. This suggests that the proposed approach could meaningfully complement current evaluation techniques. By providing objective imaging metrics, this method may be useful in studying rates of recovery independent of current subjective methods that are based on voluntary contraction. It can also be used to compare different recovery protocols and ultimately improve patient outcomes.