Computerized segmentation of whole-body bone scintigrams and its use in automated diagnostics

Summary

Bone scintigraphy or whole-body bone scan is one of the most common diagnostic procedures in nuclear medicine used in the last 25 years. Pathological conditions, technically poor image resolution and artefacts necessitate that algorithms use sufficient background knowledge of anatomy and spatial relations of bones in order to work satisfactorily. A robust knowledge based methodology for detecting reference points of the main skeletal regions that is simultaneously applied on anterior and posterior whole-body bone scintigrams is presented. Expert knowledge is represented as a set of parameterized rules which are used to support standard image-processing algorithms. Our study includes 467 consecutive, non-selected scintigrams, which is, to our knowledge the largest number of images ever used in such studies. Automatic analysis of whole-body bone scans using our segmentation algorithm gives more accurate and reliable results than previous studies. Obtained reference points are used for automatic segmentation of the skeleton, which is applied to automatic (machine learning) or manual (expert physicians) diagnostics. Preliminary experiments show that an expert system based on machine learning closely mimics the results of expert physicians.

Introduction

Whole-body scan or bone scintigraphy is a well known clinical routine investigation and one of the most frequent diagnostic procedures in nuclear medicine [1]. Indications for bone scintigraphy include benign and malignant diseases, infections, degenerative changes [2]. Bone scintigraphy has high sensitivity and the changes of the bone metabolism are seen earlier than changes in bone structure detected on radiographs [1].

The investigator’s role is to evaluate the image, which is of poor resolution due to the physical limitations of gamma camera. There are approximately 158 bones visible on both anterior and posterior whole-body scans [3]. Poor image resolution and the number of bones to inspect make the evaluation of images difficult. Some research on automating the process of counting the bone lesions has been done, but only a few studies attempted to automatically segment individual bones prior to the computerized evaluation of bone scans [4], [5], [6].

First attempts to automate scintigraphy in diagnostics for thyroid structure and function were made in 1973 [7]. Most of the research on automatic localization of bones has been done at the former Institute of medical information science at the University of Hildesheim in Germany from 1994 to 1996. The main contribution was made by Berning [5] and Bernauer [4] who developed semantic representation of the skeleton and evaluation of the images. Benneke [6] has realized their ideas in 1996.

Yin and Chiu [8] tried to find lesions using a fuzzy system. Their preprocessing of scintigrams includes rough segmentation of six parts with fixed ratios of the whole skeleton. Those parts are rigid and not specific enough to localize a specific bone. Their approach for locating abnormalities in bone scintigraphy is limited to point-like lesions with high uptake.

When dealing with lesion detection other authors like Noguchi [3] have been using intensity thresholding and manual lesion counting or manual bone ROI (region of interest) labelling. Those procedures are only sufficient for more obvious pathologies whereas new emerging pathological regions are overlooked.