Abstrakt
Monitoring and tracking the size and the number of multiple sclerosis (MS) lesions is very important in clinical medicine to understand the course and estimate the progression of this demyelination disease. The lesions could be identified by the experts with brain magnetic resonance imaging (MRI) technology, especially the fluid attenuated inversion recovery sequence (FLAIR), which generates two-dimensional slices sampled from the three-dimensional space with specified slice thickness and increment values.
Not every MRI scan, however, could be contiguous nor overlapping due to many reasons, to prevent from a drastic increase in the overall duration of the scans. Particularly, it is very hard to stabilize a child for hours in the same position; therefore, the specialists keep the scan procedure as short as possible, by increasing the slice thickness and more importantly, reducing the number of slices which cause some consistent gaps emerging between the slices and leading to inconclusive results.
Given these facts, we propose a novel procedure to overcome this inadequacy by filling the gaps of incremental MRIs based on a Nakagami imaging and a content-based morphing method generating imaginary frames between the genuine MRI slices. Afterwards, the segmented images are reconstructed in three-dimensional space to estimate the lesion volumes for three consecutive scans of one patient.
The results are greatly encouraging that we calculated 95.72% as the mean average percentage accuracy (MAPA) with 92.17% dice score (DSC%); while a little sacrifice in DSC% down to 90.35% provided us a better MAPA of 96.44%; while without morphing, the MAPA was calculated using only the binary ground truth (GT) images as 85.97%. As an expert system, the automated framework we presented would be very beneficial for volume estimations in clinics as well as visualizing the lesions and tracking the progression of MS disease.