A validation framework for probabilistic maps using Heschl's gyrus as a model.

Area of Science:

• Neuroimaging
• Neuroanatomy
• Medical Image Analysis

Background:

• Probabilistic maps are crucial for anatomical labeling and data analysis in functional neuroimaging.
• Accurate localization of structures like Heschl's gyrus (HG) in new individuals is vital but challenging.
• The accuracy of probabilistic maps depends on subject variability, template choice, and registration methods.

Purpose of the Study:

• To investigate the impact of various registration techniques on the accuracy of probabilistic maps for Heschl's gyrus (HG).
• To compare the performance of five different registration methods against a previously published affine registration method.

Main Methods:

• Probabilistic maps of Heschl's gyrus (HG) were generated using five registration techniques: implicit reference-based (IRG), DARTEL, BSpline-based, HAMMER, and unified segmentation (SPM5).
• A previously published affine registration-based map was included for comparison.
• Accuracy was assessed using sensitivity, specificity, and positive predictive value (PPV) within a leave-one-out cross-validation framework.

Main Results:

• Implicit reference-based (IRG) and DARTEL registration techniques outperformed other methods in generating accurate probabilistic maps of HG.
• All tested registration techniques demonstrated high positive predictive values (PPV).
• Sensitivity rates were generally low across all methods, suggesting conservative localization estimates.

Conclusions:

• IRG and DARTEL are superior registration methods for creating accurate probabilistic maps of Heschl's gyrus (HG).
• Probabilistic maps offer reliable but conservative estimations of HG location and extent in new subjects.
• Further refinement of registration techniques is needed to improve sensitivity in probabilistic mapping.