Reduction of Amyloid Plaque FP Detections in MR Images of the APP Transgenic Mouse Brain using Unsupervised SVM

Gheorghe Iordanescu^1,2, Palamadai Venkatasubramanian^1,2, Alice Wyrwicz^1,3

We present a novel method for amyloid plaques FP reduction in MR images. FP sources like vessels and brain region borders are modeled by multiscale features computed based on the square matrix of second-order partial derivatives and its eigenvalues. Our approach is novel since it does not require supervised training, and we did not follow the common approach of computing specific functions describing the sheetness or lineness of a catchment basin. Instead, we use the SVM flexibility of computing non-linear classification functions that can be used to detect FPs of specific shape. Our results show that our unsupervised algorithm is flexible and can be extended to reduce FP for plaque detection in MR images of AD mouse models, making our method suitable for the analysis of individual plaques and plaque distribution within different brain structures.