Thomas WJ Ash1, Guy B. Williams1,
Fiona Regan2, Sally Georgia Harding1, Tero Saukkonen2,
David B. Dunger2, T Adrian Carpenter1, Alison Sleigh1
1Wolfson Brain Imaging Centre,
University of Cambridge, Cambridge, Cambridgeshire, United Kingdom; 2Department
of Paediatrics, University of Cambridge, Cambridge, Cambridgeshire, United
Kingdom
Using
machine learning (support vector machines) we construct a model that
classifies soleus muscle spectra of adolescents into Type 1 diabetes /
control groups with 95% accuracy in leave one out tests. This is an improvement on previous analysis
of the data using solely IMCL peak area calculations which achieves 84% accuracy. The model confirms that IMCL height
increases in type one diabetics, as well as showing that creatine and choline
peaks are broadened in the type 1 diabetic group.