Purpose

Neuroblastoma is the most frequent solid extracranial cancer in childhood. Its diagnosis and prognosis are based on the information provided by multiparametric magnetic resonance images. Our focus is to explore the utility of diffusion and perfusion changes in neuroblastoma as an early biomarker of diagnosis.

Material and methods

Multiple MR imaging real-word sequences, available within the H2020-PRIMAGE project, were used from 45 patients. Volumes-of-interest were calculated and transferred to DCE perfusion and apparent diffusion coefficient (ADC) maps. Histogram analysis and clustering-based unsupervised ML algorithms were used to determine the mean and standard deviation of initial area under the curve at 60 seconds (IAUC60) and ADC for automatic differentiation of neuroblastic tumors. The resolution of voxel intensity was estimated and the data was smeared accordingly to identify and remove the noise and low-quality voxels (defined as those with an uncertainty larger than 10%).

Results

Significant differences in mean ADC were found for neuroblastic tumors: 1.0 for ganglioneuroma, 0.82 for ganglioneuroblastoma, and 0.52 for neuroblastoma, with an uncertainty of 0.11%, 42% and 16%, respectively. This result improves tumor differentiation with respect to state-of-the-art voxel-by-voxel methodologies, which were found to be: 1.6 for ganglioneuroma, 1.7 for ganglioneuroblastoma, and 1.3 for neuroblastoma, with an uncertainty of 3.6%, 12% and 17%, respectively. Mean IAUC60 was found to have a value of 43 (and 14% uncertainty) for neuroblastoma, as opposed to a value of 17 (and 17% uncertainty) with state-of-the-art voxel-by-voxel methodologies.

Conclusion

The proposed novel technique to determine IAUC60 and ADC parameters manages to differentiate benign and malignant neuroblastic tumors.

Learning objectives

Learning objectives