Evaluation of Subcutaneous and Intermuscular Adipose Tissues by Application of Pattern Recognition and Neural Networks to Ultrasonic Data: a Model Study

In medical diagnostics, there are strong reasons to distinguish between subcutaneous adipose tissue (SAT) and intermuscular adipose tissue (IMAT), as infiltration of IMAT into muscle causes health complications with aging (sarcopenia), diabetes, cardiovascular disease, and obesity. Since such assessments are performed using stationary and labor-intensive imaging modalities (MRI, CT), the development of portable devices based on ultrasound measurements, could aid proactive medicine and expand screening capabilities. The aim of this model study was to demonstrate the feasibility of differentially assessing SAT and IMAT by extracting evaluation criteria from propagating ultrasound signals. A set of 25 phantoms, using gelatin gel as the muscle matrix and oil for the SAT and IMAT compartments, formed a network with gradual changes in SAT and IMAT ranging from zero to 50%. Ultrasound signals were recorded at frequencies of 0.8 and 2.2 MHz, and assessment criteria were used, including ultrasound velocity and intensity derivatives. The intersection of decision rules based on evaluation criteria generated domains of possible recognition solutions. In parallel, using the same data partitions for training and test sets, artificial neural network (ANN/LSTM) analysis was applied. Both approaches demonstrated diagnostically acceptable SAT and IMAT resolution, opening up prospects for the ultrasound method.