You knew AI was coming for medicine, just not exactly how.
One of the biggest issues I have as a breast surgeon is trying to get breast symmetry. Breasts are not even. The areolas are different sizes, heights. The nipples are not centered. The breasts droop. The underlying ribcage may stick out farther on one side which makes that side look larger. The fat out into the armpit (which I liposuction with my breast lifts and reductions and isn’t breast tissue- it is truly just fat) distorts the width and bra band size. Stretch marks? Cysts? Where does the inframammary fold start? The list goes on and on.
So I got Vectra 3D imaging years and years ago. I thought finally the holy grail of modern medicine- I will take a 3D photo of the breast and walah! it will tell me the right breast is 37cc larger than the left. If someone wants to be 1 cup size smaller, I will need to remove exactly 248 grams from the right and 211 grams from the left.
But sadly no.
I love my Vectra, and I think it can be helpful, but it is frequently not accurate. 3D photos cannot capture details we still find doing old fashioned exams. How stretchy is the skin? How heavy is it the breast? Breast tissue weighs differently than fat, and your breast is a mix of the two. Oh and this mixture, which differs between people and between the same person changes as they age, have weight changes, menopause, have pregnancies, and breastfeed. So reducing someone the exact same 2 cup sizes may remove 150 grams in one patient and 500 grams in another.
This journal article in the Plastic and Reconstructive Surgery Journal February 2024 “Identification of Key Breast Features Using a Neural Network: Applications of Machine learning in the Clinical Setting of Plastic Surgery.” This study looked at using computer programs to use AI to evaluate the breast. Can it help us assess, plan, and improve symmetry?
This is in the beginning stages. Their goal was just for the program to detect key features: boundary of the breast, nipple areola complex, and suprasternal notch.
It was fast- it found these key landmarks in 0.52 seconds (faster than I am with my measuring tape).
They conclude, “Neural networks and machine learning have the potential to improve the evaluation of breast symmetry in plastic surgery by automated and quick detection features.”
My thoughts?
I agree! To be honest, I am stunned this assessment of breast symmetry- of nipple position, fold position, areola size, degree of drooping, and VOLUME difference- has not been cracked sooner. But as I talk about in my clinic, the body is dynamic. Until we have a computer AI program that can assess skin tension and stretchiness, volume and how it correlates to weight (one cup of marshmallows does not weigh the same as a cup of stones though their volumes are equal) and can anticipate tissue forces and how they change and heal, there is still an art to plastic surgery.
There is still an art to plastic surgery. I find a beauty in knowing the body defies being completely predictable.