The aim of this thesis was to determine whether machine learning can be
used to identify and interpret morphological differences between muscle fiber
cross-sections in extended and contracted muscles. For the purpose of the
study data was obtained from five mice specimens. Morphological features
of muscle fiber cross-sections suitable for this type of analysis were obtained
through machine and human segmentation of microscopic images of the entire
cross-section of the gastrocnemius muscle in both extended and contracted
states. After preprocessing the data, the performance of various model types
was evaluated using nested cross-validation, after which logistic regression
was selected for further analysis due to its high average balanced accuracy
and interpretability. Since the results supported the finding that the models
could identify morphometric differences between muscle fiber cross-sections in
the aforementioned states based on the analysed data, statistical significance
was then tested and confirmed by using a permutation test.
|
