What type of analysis should i perform on my data

Question

I have two groups of clinical subjects i.e healthy (25 subjects) and patient group (25 subjects) ,for each subject i have volumetric data for several regions of brain (15 regions ).

I would like to analyze if there is any considerable volumetric atrophy in brain between healthy and patient group i.e. whether volume in specific region of brain decreases in patient group compared to healthy group

Hypothesis: Generally there would be volumetric atrophy in patient group compared to healthy.

Null hypothesis : There will be no difference between healthy and patient group .

I would like also perform post hoc analysis and compare both groups to see which specific region of brain shows more significance compared all other regions.

Any input on this would be greatly appreciated.

Answer 1

You can try many things, clearly there is no "only one solution" to do this.

You can to design a quick and dirty SVM to separate your healthy and patient data.

Depending on how it works (good or bad)

• If very good: you can try to add some dimensionality reduction to highlight the key factors to differentiate between healthy / patient.
• Else:
  • you can try to use different SVM Kernels (RBF, etc.)
  • A different model: KNN, Bayes, Random Forest, Decision Tree. If you try to compare different kind of models or different settings for a given type of model, do not forget to add a cross-validation step.

Finally, you should keep 30% of your dataset (30% healthy and 30% patient) aside from any analysis. Do not use these data any time for training, it will be useful to validate your model on "new data" and prevent overfitting. You should select these 30% data randomly.

Have a look to some machine learning tutorial (R or Python with ScikitLearn).