Using Breastgan v2.0 to Improve Patient Selection in Single- Stage Augmentation/ Mastopexy

Goals/Purpose: The controversy surrounding single- stage breast augmentation mastopexy is well- described in the plastic surgery literature. When performed individually, breast augmentation and mastopexy are safe and reliable, incurring relatively few complications and achieving predictably high patient satisfaction scores. However, the single- stage combined augmentation mastopexy juxtaposes two procedures exerting opposing forces on the skin envelope of the breast: expansion by way of breast volume augmentation but reduction concurrent with pedicled nipple repositioning to lift the breast. This opposition has earned the augmentation mastopexy its degree of difficulty and revision rate, both among the highest across all plastic surgery procedures. Carefully balancing these conflicting vectors may be challenging for plastic surgeons. The authors of the present study propose BreastGAN (Breast Generative Adversarial Network), an artificial intelligence- equipped tool designed to simulate results of breast surgery from a single frontal preoperative image. It outputs separate images reflecting plausible breast augmentation and combined augmentation mastopexy results. This may provide patients with more tangible insight into the proposed procedures and allow them to provide more informed consent.

Methods/Technique: All patients with signed research consents who underwent primary breast augmentation or combined augmentation mastopexy performed by one of the authors (E.H.F.) between 2003 and 2018 were included. In total, before and after image pairs were collected from 1,235 breast augmentation patients and 389 augmentation mastopexy patients, constituting two separate databases. BreastGAN was evaluated on all images in both test sets, or 309 pairs of augmentation patient images and 97 pairs of augmentation mastopexy patient images. Images generated by the tool were compared to the corresponding true postoperative images.

Results/Complications: BreastGAN (trained to output augmentation and augmentation mastopexy results, respectively) was deployed across preoperative images of patients presenting with a wide array of breast morphologies who each underwent either breast augmentation or augmentation mastopexy. See Figure 1 for a sample of such patients, including preoperative and true postoperative images, alongside BreastGAN- generated surgical simulation results.

Conclusion: This study features a potential low-cost alternative to costly surgical simulators. If adopted, it may provide surgeons with a tool with which to obtain more informed consent by giving them a tangible example of a plausible postoperative result for multiple procedures. GAN training on more images provided by a complete distribution of plastic surgeons will be the subject of future study.

Figure 1. Sample of BreastGAN testing results. Importantly, poor nipple placement in the “AI-generated augmentation” columns of rows 3 and 4 reflects a key difference between augmentation and augmentation mastopexy in ptotic breasts. *This illustrates an interesting example of BreastGAN used to simulate a breast augmentation in a patient with ptotic breasts.