The detection and significance of subtle changes in mixed-signal brain lesions by serial MRI scan matching and spatial normalization

The purpose of this work is to detect and assess the significance of subtle signal changes in mixed-signal lesions based on serial MRI scan matching. Pairs of serially acquired T1-weighted volume MR images from 20 normal controls and seven patients with epilepsy were matched and difference images obtained. The precision and consistency of the registration were evaluated. The Gaussian noise level in the difference images was determined automatically. A structured difference filter was then used to segment structured (changed) voxels from the Gaussian noise. In the controls, the structured difference images were normalized into Talairach space, resulting in a structured noise map. The significance of changes in patients was assessed by spatial normalization and comparison with the structured noise map. The precision and consistency of the co-registration were ≤0.06 mm with a registration success rate of 100%. The Gaussian noise level in the difference images was in the range 3.0–6.9. In the controls, an average of 1.6% of the brain voxels were classified as structured. Since-based registration resulted in a reduction of < 1 % in the amount of structure compared to linear interpolation. The structured noise map in controls showed high noise density in areas affected by image artefacts. We show examples of significant changes found in lesions which had been reported as unchanged on visual inspection. A novel quantitative approach has been presented for the detection and quantification of subtle signal changes in lesions. This method is of potential clinical value in the non-invasive characterization of signal change and biological behaviour of neoplastic lesions.