Automated Aesthetic Evaluation After Breast Reconstruction Using Routine Postoperative Photographs: A No-Code AutoML...

Aesthetic assessment after breast reconstruction remains subjective and is often affected by inter-rater variability. A practical and reproducible evaluation method is needed for routine clinical use. This study aimed to develop an automated algorithm for classifying postoperative breast aesthetics into three categories: excellent, fair, and bad using Google Cloud AutoML Vision, without requiring advanced programming expertise.

A total of 519 frontal two-dimensional photographs obtained 1 year after breast reconstruction were retrospectively collected. Images were annotated into three aesthetic categories based primarily on the Harris method, with additional semi-quantitative elements from the Sawai aesthetic scale. The original dataset was divided into training (n = 336), validation (n = 84), and test (n = 99) sets. Data augmentation was applied only to the training set, increasing it to 1,344 images. In total, 1,527 images were uploaded to AutoML Vision to build a single-label image-classification model. Model performance was compared with that of plastic surgeons, nurses, and laypersons using the same test set.

The algorithm achieved a precision of 62.1%, recall of 59.6%, and overall accuracy of 60.6% on the test set. The weighted Cohen's kappa coefficient was 0.47. Mean classification accuracy was 78.9% for plastic surgeons, 60.2% for nurses, and 59.3% for laypersons. The algorithm showed numerically higher agreement with the reference labels than nurses and laypersons, although its overall performance remained moderate.

An automated aesthetic assessment model for breast reconstruction was successfully developed using AutoML Vision. Although performance remained inferior to that of plastic surgeons, the model demonstrated moderate agreement and may serve as a practical adjunctive tool for standardized postoperative aesthetic evaluation.

This journal requires that authors assign a level of evidence to each article. For a full description of these Evidence-Based Medicine ratings, please refer to the Table of Contents or the online Instructions to Authors www.springer.com/00266 .