Multi-tracer PET image fusion using fuzzy theory: A feasibility study

Positron emission tomography (PET) scan analysis has been proved to be a suitable tool for patient monitoring and early assessment of tumour response in oncology. Currently, 2'-deoxy-2'-[
¹⁸
F]-fluoro-D-glucose (FDG) is the most widely used tracer in clinical practice. However, imaging tumor's glucose consumption, with the FDG alone may not be sufficient to assess the therapy response [1]. Adding the measure of other features of cancer metabolism like proliferation, hypoxia and apoptosis may improve tumor response assessment. 3'-[
¹⁸
F]-fluoro-3'-deoxythymidine (FLT) is a proliferation tracer providing earlier and more specific information in patient monitoring than FDG [2]. Hypoxia imaging using the 18-Fluoromisonidazole (FMISO) tracer can identify tumours resistant to therapy, especially radiotherapy [3]. Fusing all available measurements obtained with these different tracers could be valuable to thoroughly and potentially more accurately assess tumor response. The images corresponding to each tracer acquisition demonstrate various sensitivity and specificity depending on the tumour tracer uptake and are therefore not directly comparable. The aim of our method is to create an automated fused 3D map from multiple scans with different tracers. The first step consists in putting different images in a common mathematical framework using the fuzzy logic. First of all, the fuzzy-C means (FCM) algorithm will be apply to each tracer scan leading to the building of fuzzy membership functions (FMF). Then, fuzzy maps representing each tracer distribution were derived from the FMF. The third step of the process is to fuse these fuzzy maps using an operator managing their complementarities and their redundancy. The fused map should reflect each tracer distribution and offer workable information in one image underlying their differences and similarities. This study aims to demonstrate the feasibility of such an approach by comparing various fusion methodologies on simulated datasets of multi-tracer images of tumours.