Abstract #2046

What is the best method for robust statistical inference on connectomic graph metrics?

Mark Drakesmith^1,2, David Linden², Anthony S David³, and Derek K Jones^1,2

¹CUBRIC, Cardiff University, Cardiff, United Kingdom, ²Neuroscience and Mental health Research Institute, Cardiff University, Cardiff, United Kingdom, ³Institute of Psychiatry, Psychology and Neurosceince, Kings College London, London, United Kingdom

Connectomic network analyses, while powerful, suffer from high instability, which is problematic for robust statistical inference. The area under the curve (AUC) across thresholds is a common approach, but lacks robustness to this instability. A superior approach is multi-threshold permutation correction (MTPC), but this is computationally expensive. Smoothed AUCs (smAUCs) are less costly and theoretically can achieve the same level of sensitivity as MTPC. smAUC was tested and compared with MTPC in a virtual patient-control comparison. Results show that smAUC sensitivity is not consistently comparable to MTPC and that exhaustive searching across the threshold space is required for robust inference.

This abstract and the presentation materials are available to members only; a login is required.

Join Here