Automatic method for tumor segmentation from 3-points dynamic PET acquisitions

Abstract

In this paper a novel technique to segment tumor voxels in dynamic positron emission tomography (PET) scans is proposed. An innovative anomaly detection tool tailored for 3-points dynamic PET scans is designed. The algorithm allows the identification of tumoral cells in dynamic FDG-PET scans thanks to their peculiar anaerobic metabolism experienced over time. The proposed tool is preliminarily tested on a small dataset showing promising performance as compared to the state of the art in terms of both accuracy and classification errors.

Graph Laplacian for image anomaly detection

“…PET images carry information about cells metabolism and are therefore suitable for this task; however, PET segmentation is still an open problem mainly because of limited image resolution and strong presence of acquisition noise [69]. In [10,63,65], we successfully explored the use of RXD to identify the anomalous behavior of cancer cells over time in sequences of three FDG-PET images acquired over a time span of one hour. A quick visual overview of this setup is shown in Figure 7.…”

“…To do this, we build a 4D matrix I, having the three spatial dimensions as first three dimensions, and time as fourth dimension. Being acquired at different times, with the subject assuming slightly different positions, it is worth recalling that the images need to be aligned using registration algorithms as detailed in [65]. The resulting matrix I will then have size 144 × 144 × 45 × 3.…”

“…One of the most used and widely validated techniques for anomaly detection is known as Reed-Xiaoli Detector, often called RX Detector for short [56], which is the most known example of covariance-based anomaly detectors. This class of detectors has found wide adoption in many domains, from hyperspectral [49] to medical images [65]; however, methods of this type suffer from crucial drawbacks, most noticeably the need for covariance estimation and inversion. Many situations exist where the drawbacks of these state-of-the-art anomaly detectors lead to poor and unreliable results [67].…”

Global and local anomaly detectors for tumor segmentation in dynamic pet acquisitions

“…Registration of DS1 and DS2 with respect to ES is therefore required. The registration parameters have been selected following common practice in the literature and detailed explanation about the procedure can be found in [4]. We will refer to the two registered images as DS1' and DS2'; their voxels can be considered aligned to those of ES.…”

“…We have already presented some early results on the topic in [4], where the study was limited to a single global anomaly detection algorithm. Using 3 PET images acquired at different times, the approach presented aims at recognizing tumoral voxels by their anomalous behavior over time.…”