Validating an image segmentation program devised for staging lymphoma

Abstract

Hybrid positron emission tomography–computed tomography (PET–CT) imaging systems are an important tool for assessing the progression of lymphoma. PET–CT systems offer the ability to quantitatively assess lymphocytic bone involvement throughout the body. There is no standard methodology for staging lymphoma patients using PET–CT images. Automatic image segmentation algorithms could offer medical specialists a means to evaluate bone involvement from PET–CT images in a consistent manner. To devise and validate an image segmentation program that may assist staging lymphoma by determining the degree of bone involvement based from PET–CT studies. A custom-made program was developed to segment regions-of-interest from images by utilising an enhanced fuzzy clustering technique that incorporates spatial information. The program was subsequently tested on digital and physical phantoms using four different performance metrics before being employed to extract the bony regions of clinical PET–CT images acquired from 248 patients staged for lymphoma. The algorithm was satisfactorily able to delineate regions-of-interest within all phantoms. When applied to the clinical PET–CT images, the algorithm was capable of accurately segmenting bony regions in less than half of the subjects (n = 103). The performance of the algorithm was adversely affected by the presence of oral contrast, metal implants and the poor image quality afforded by low dose CT images in general. Significant changes are necessary before the algorithm can be employed clinically in an unsupervised fashion. However, with further work performed, the algorithm could potentially prove useful for medical specialists staging lymphoma in the future.