301. Prostate cancer Radiomics using multiparametric MR imaging: An exploratory study. (December 2018)

• Record Type: Journal Article
• Authors: Barucci, A.; Bastiani, P.; Carpi, R.; Fondelli, S.; Giannetti, A.; Olmastroni, M.; Pini, R.; Ratto, F.; Rucco, M.; Zatelli, G.; Esposito, M.
• Is Part Of: Physica medica. Volume 56(2018)Supplement 2
• Page Start: 246
• Publication Date: 2018-12
• Subjects: Medical physics -- Periodicals; Biophysics -- Periodicals; Biophysics -- Periodicals; Imagerie médicale -- Périodiques; Radiothérapie -- Périodiques; Rayons X -- Sécurité -- Mesures -- Périodiques; Physique -- Périodiques; Médecine -- Périodiques; 610.153
• DOI: 10.1016/j.ejmp.2018.04.310
• Languages: English
• ISSNs: 1120-1797

Abstract

Abstract : Purpose: In this exploratory study we investigated the potentiality of radiomic features extracted from multi-parametric magnetic resonance imaging (mp-MRI), as a tool to quantify tumour phenotypic characteristics in Prostate Cancer (PCa)[1] . Methods and materials: A retrospective cohort of about 20 PCa patients who underwent a 1.5 Tesla mp-MRI was considered for this study. For each patient diffusion maps were generated using conventional imaging sequences in mp-MRI (DWI). Then radiomic features (about 800) were extracted from two ROIs (healthy and tumour tissue respectively) segmented by radiologists, by means of research softwares (3D Slicer –www.slicer.org ; and software developed by authors in Matlab). Subsequently a subset of robust features was selected based on the capability to discriminate healthy tissue from tumour. Results: Cluster analysis was then performed using this subset of radiomic features, the resulting dendrogram showing two bigger cluster (healthy tissue and tumour) with many subclusters. This result shows the radiomic features potentiality to discriminate healthy tissue from tumour, at the same time opening at the possibility to quantify tumour phenotypic characteristics exploring subclusters. Conclusion: Radiomic features derived from mp-MRI diffusion maps and evaluated using research softwares have shown the potentiality to distinguish healthy from tumour regions on prostate cancer lesions. Cluster analysis supports this results. However some critical issues emerged related to the number of patients used in the study, which deserve to be investigated more deeply. As future activities we plan to use topology and geometry based methods as robust approaches for improving features selections. On top of the new feature space we will train Machine Learning algorithms (e.g. Artificial Neural Network) for automatically classifying ROIs.