Anatomic Features of the Abdomen Predict Donor Site Complications in Deep Inferior Epigastric Artery Perforator Flap Breast Reconstruction: A Machine Learning Approach.

BACKGROUND

Abdominal donor site complications remain significant concerns in Deep Inferior Epigastric Perforator (DIEP) flap breast reconstruction. In this study, we pair computed tomography (CT) morphometric analysis of the abdominal wall with machine learning approaches to identify features of the abdomen that signal increased risk for donor site complications.

METHODS

Morphometric analysis of the abdominal wall was conducted on pre-operative abdominal CT scans from 107 patients who underwent DIEP flap breast reconstruction by a single surgeon (TMM). Standard statistical approaches were used to correlate these anatomic measurements to adverse donor site outcomes. Multilayer perceptron (MLP) classifier machine learning algorithms based on morphometric data, BMI, or both were used to generate prediction models for donor site outcomes.

RESULTS

BMI correlated with features of the abdominal wall including abdominal wall protrusion, subcutaneous and visceral fat volumes, rectus abdominis width and abdominal wall thickness. Using quartile analysis, abdominal bulge correlated with BMI and inter-ASIS distance. Incision dehiscence correlated with abdominal wall protrusion, rectus abdominus width, visceral fat, and mean rectus abdominis density. Using MLP approaches, prediction models for bulge based on CT morphometric measurements outperformed those based on BMI alone whereas these models performed similarly in predicting donor site dehiscence. Abdominal wall protrusion was the most influential variable in both models.

CONCLUSIONS

Our study is the first to couple comprehensive morphometric analysis with machine learning approaches to delineate anatomic risk factors for abdominal donor site outcomes. Our multi-feature donor site outcome prediction models may be more informative than BMI in pre-operative risk assessment.