Computational Analysis of Mris Predicts Osteosarcoma Chemoresponsiveness

Abstract

Aim: This study aimed to improve osteosarcoma chemoresponsiveness prediction by optimization of computational analysis of MRIs. Patients & methods: Our retrospective predictive model involved osteosarcoma patients with MRI scans performed before OsteoSa MAP neoadjuvant cytotoxic chemotherapy. Results: We found that several monofractal and multifractal algorithms were able to classify tumors according to their chemoresponsiveness. The predictive clues were defined as morphological complexity, homogeneity and fractality. The monofractal feature CV for Λ′(G) provided the best predictive association (area under the ROC curve = 0.88; p <0.001), followed by Y-axis intersection of the regression line for box fractal dimension, r² for FDM and tumor circularity. Conclusion: This is the first full-scale study to indicate that computational analysis of pretreatment MRIs could provide imaging biomarkers for the classification of osteosarcoma according to their chemoresponsiveness.

An Adapted One-dimensional Computational Approach for Irregular ROI Analysis Improves Osteosarcoma Classification

“…By the irregular ROI approach, this study indicated that 1-D Dh could extract useful predictive information only from tumour tissue, based on the result that analysis of entire images, without ROI, did not provide any signi cant predictive performance. This result was in line with the previous investigation showing the bene cial in uence of ROI delineation on the predictive performance of 2-D fractal analysis 10 .…”

“…2-D fractal analysis is also non-directional, but unlike 1-D Dh, it is not fully compatible with irregular ROI. This comparison indicated that the ideal compatibility of 1-D Dh with irregular ROI only offers a slight improvement in predictive performance over the previously reported 2-D box-counting 10 . There are no previous tumour morphology studies that could be directly compared to the tumour MRI analysis reported here.…”

“…1c). The predictive performance of the demographic, MRI and clinicopathological parameters in this patient group was previously reported in detail 10 . Only the tumour area reached signi cant predictive performance (AUC = 0.27, P < 0.05) to indicate that a larger area predicts higher chemoresistance 10 .…”

“…Heterogeneity of tumour microscopic or macroscopic morphology is increasingly exploited in the prediction of disease outcome or tumour chemoresistance 4,10 . However, tumour morphology obtained by MRI is still underexploited, possibly due to the analytical problem with irregular tumour ROIs.…”

“…Its treatment includes methotrexate [M] doxorubicin [A], cisplatin [P] (MAP) induction chemotherapy, followed by surgical tumour removal or amputation, depending on the local tumour invasiveness. The computational analysis of osteosarcoma MR images was reported only in two previous studies, both employing the standard box-counting 2-D analysis 8,10 . Magnetic resonance imaging is increasingly used because of its high contrast, spatial resolution, the possibility of speci c tissue enhancement and the ability to perform functional, diffusion and perfusion imaging 11 .…”