A grid overlay framework for analysis of medical images and its application to the measurement of stroke lesions

Objectives To create and evaluate an interactive software tool for measuring imaging data in situations where hand-drawn region-of-interest measurements are unfeasible, for example, when the structure of interest is patchy with ill-defined boundaries.

Methods An interactive grid overlay software tool was implemented that enabled coding of voxels dependent on their imaging appearance with a series of user-defined classes. The Grid Analysis Tool (GAT) was designed to automatically extract quantitative imaging data, grouping the results by tissue class. Inter- and intra-observer reproducibility was evaluated by six observers of various backgrounds in a study of acute stroke patients.

Results The software tool enabled a more detailed classification of the stroke lesion than would be possible with a region-of-interest approach. However, inter-observer coefficients of variation (CVs) were relatively high, reaching 70% in "possibly abnormal" tissue and around 15-20% in normal appearing tissues, while intra-observer CVs were no more than 13% in "possibly abnormal" tissue and generally less than 1% in normal-appearing tissues.

Conclusions The grid-overlay method overcomes some of the limitations of conventional Region Of Interest (ROI) approaches, providing a viable alternative for segmenting patchy lesions with ill-defined boundaries, but care is required to ensure acceptable reproducibility if the method is applied by multiple observers.