An automatic segmentation method for the measurement of the functional volume of oncological lesions on MR ADC maps

Human cancers frequently display intra-tumor phenotypic heterogeneity, whose nature can have profound implications both for tumor development and therapeutic outcomes. Some recent research efforts have been devoted to develop advanced image processing methods able to extract imaging descriptors characterizing such intra-tumor phenotypic heterogeneity. However, most methods need to accurately define the lesion volume in order to extract imaging descriptors. This work aims at assessing a novel segmentation method to measure the functional volume of lesions on MR ADC maps. The method was validated in advanced breast cancer patients addressed to Neoadjuvant Chemotherapy and surgical intervention, undergoing pre-treatment FDG-PET and multi-parametric MR studies. PET metabolic volume (MTV), SUVmean, SUVmax, and Total Lesion Glycolysis (TLG) of lesions were measured using an already validated segmentation algorithm [Gallivanone et al., J. Instr. 2016]. The MR functional volume of lesions segmented on the ADC map resulted directly correlated to PET MTV. We defined a new parameter characterizing the MR total diffusion of lesions, the Total Lesion Diffusion (TLD) that resulted directly correlated to PET TLG. Furthermore, we assessed an inverse correlation between SUVmax and ADCmin within the PET and MR functional volumes, respectively. Textural indexes were also evaluated. Correlations (p<0.05) were found among the textural image descriptors related to the spatial distribution of the signal extracted within the PET and MR functional volumes. In conclusion, our segmentation method is effective to define the functional volume of lesions on ADC maps.