Related Experiment Video
Updated: Jun 4, 2025
13:44
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
42.8K
The Surgeon's Digital Eye: Assessing Artificial Intelligence-generated Images in Breast Augmentation and Reduction
Arsany Yassa1, Arya Akhavan1, Solina Ayad2
1From Division of Plastic and Reconstructive Surgery, Rutgers New Jersey Medical School, Newark, N.J.
Plastic and Reconstructive Surgery. Global Open
|December 23, 2024
Summary
Artificial intelligence (AI) image generation shows potential for plastic surgery consultations, but current models struggle with realistic nipple-areola complex visualization and diverse body types. Further development is needed before AI can fully replace patient photographs for surgical planning.
Area of Science:
- Plastic Surgery
- Medical Imaging
- Artificial Intelligence
Background:
- Public overestimation of artificial intelligence (AI) capabilities in plastic surgery outcomes.
- Need to assess AI-generated image accuracy for breast augmentation and reduction.
- Determine AI's utility in setting realistic surgical expectations.
Purpose of the Study:
- Evaluate the accuracy and realism of AI-generated pre- and post-surgery images for breast augmentation and reduction.
- Assess the clinical value of AI in plastic surgery consultations.
- Compare the performance of different AI platforms (GetIMG, Leonardo, Perchance).
Main Methods:
- Utilized AI platforms (GetIMG, Leonardo, Perchance) to create surgical images.
- Evaluated images using 11 metrics by board-certified plastic surgeons and residents.
- Categorized images based on realism and clinical value.
- Excluded nipple-areolar complex images due to AI restrictions.
Main Results:
- AI models demonstrated comparable realism and clinical value scores.
- GetIMG showed superior surgical relevance compared to other models.
- Healing and scarring prediction was a consistently underperforming metric across all models.
- Some AI images were perceived as "cartoonish" with unrealistic skin textures.
Conclusions:
- AI models show promise in predicting outcomes like breast size and volume.
- Significant limitations include the absence of nipple-areola complex visualization and inconsistent representation of diverse body types and skin tones.
- Current AI-generated images are not yet a substitute for actual patient photographs in consultations.
