I can't help but feel that a discussion on the merit of BMI is a
digression, from the OP's question.
In addition, to being of no relevance to "R Programming".
In relation to Richard's technical comments:
As per my previous post, it is possible to get *relative" measures.
(Assuming the images are not on a standardized scale, which they could be).
In relation to Jim's anatomical comments:
An improved algorithm could evaluate a subject's body type.
In addition to factoring in gender, ethnic characteristics and age.
In any case, I don't see why image-based classification is any less
worthwhile than conclusions based on timeseries plots and
scatterplots...