A newly developed neural network is highly accurate at detecting important landmarks in breast surgery—unlocking the potential for objective assessment of breast symmetry, suggests a study in the February issue. Plastic and Reconstructive Surgery®, Official medical journal of the American Society of Plastic Surgeons (ASPS). The journal is published by Wolters Kluwer in the Lippincott Portfolio.
Neural networks and machine learning have the potential to improve assessment of breast symmetry in reconstructive and cosmetic breast surgery by enabling rapid, automatic detection of features used by plastic surgeons.”
Lead author Nitzan Kenig, MD, Albacete University Hospital, Spain
Developing neural networks for objective breast assessment
Breast symmetry is a key concern in breast surgery, and is usually assessed by simple subjective assessment by both patient and surgeon. Computer programs can provide more objective assessments, but with limitations such as the need to enter data manually and long calculation times.
Neural networks – an artificial intelligence technique that seeks to mimic the way the human brain processes data – are being explored for their potential to improve care in various areas of medical practice. Dr. Koenig and colleagues developed an “ad hoc convolutional neural network” to identify key breast features used in breast symmetry assessment.
Using an open-source algorithm called YOLOV3 (“You Only Look Once,” version 3), the researchers trained their neural network to identify three anatomical features used in female animal evaluation: the breast border, the nipple-areola complex (the nipple and surrounding tissue), and the suprasternal groove (at the base of the neck, the depression at the top of the breastbone).
The neural network was trained using 200 frontal photographs of breast surgery patients. Its effectiveness in identifying key features of the breast was tested using an additional set of 47 photographs of patients undergoing breast reconstruction after breast cancer surgery.
Potential for ‘rapid, automated, objective’ assessment of breast symmetry
After training, the neural network was highly accurate in localizing the three features, with a total detection rate of 97.7%. For right and left breast boundaries and the nipple-areola complex, accuracy was 100%. For the suprasternal groove, the detection rate dropped to 87%. Processing was fast, with an average detection time of 0.52 seconds.
The neural network was able to detect and localize key features even in visibly asymmetric breast reconstructions. The high success rate confirmed that the training data set was sufficient, not requiring data augmentation techniques
“Neural networks and machine learning have the potential to improve the assessment of breast symmetry in plastic surgery through automatic and rapid detection of features used by practicing surgeons,” said Dr. Koenig and co-authors conclude. They believe that, with further advances in image recognition capabilities and their application in breast surgery, neural networks can play a role in breast symmetry assessment and planning aesthetic and reconstructive plastic surgery.