Keywords:
Artificial Intelligence and Machine Learning, Artificial Intelligence, Interventional vascular, Liver, CT, Embolisation, Radiation therapy / Oncology, Multidisciplinary cancer care, Retrospective, Diagnostic or prognostic study, Multicentre study
Authors:
A. Bruno1, C. Mosconi1, A. cucchetti1, A. Cappelli1, I. Bargellini2, G. Peta1, F. Modestino1, R. Cioni2, R. Golfieri1; 1Bologna/IT, 2Pisa/IT
DOI:
10.26044/ecr2020/C-06273
Conclusion
The arterial phase was observed to be the best acquisition phase for providing information regarding the “sensitivity” of the tumour to TARE. In particular, the results in the present study suggest that iCCs with a higher uptake of iodine contrast in the arterial phase (higher mean histogram values) and with more homogeneous distribution (lower kurtosis, contrast, dissimilarity and higher homogeneity and correlation) represent hypervascular tumours having fewer acellular and necrotic areas. Such hypervascular tumours are evidently more suitable for intra-arterial radiation therapy than hypovascular tumours.
In the venous phase, the OR cases still had a higher uptake of iodine contrast (namely, higher histogram mean). In the late phase, it was observed that textural features could still play a role as a prognostic response factor. This role was highlighted by the LASSO regression which showed that higher skewness (coeff: 0.1316) and GLCM Dissimilarity (coeff: 1.9183), correlated with the likelihood of achieving an OR, together with lower kurtosis (coeff: -0.0122). These findings could easily be explained by juxtaposing the observation that the histogram mean was lower in non-OR cases, and, showed a more homogeneous washout in the delayed phase but with a higher prevalence of the darkest pixels. Conversely, the OR cases were more likely to retain more iodine contrast during the delayed phase because of their previous higher enhancement.
Considering all these aspects, it was not surprising to find the present “radiomic signature” as a determinant of OR (AUC: 0.896) and PFS (p=0.008). It should be added that the correlation between the present “radiomic signature” and OS was not so strong as that observed for PFS. This could be the consequence of the small sample size, but also that other, more impactful characteristics mitigated the role of Radiomics. The findings that CA19-9, the history of previous surgery or embolo-therapies as well as tumour burden were related to OS were in keeping with the previous literature [7,10-13] and supported the fact that the forecast of the course of iCCs after TARE is likely the result of a complex interrelationship between the tumour and the host.
The present study is burdened by the common limitations of retrospective studies. This led to a heterogeneous patient population, depending on the practice pattern of the referring oncologist. Second, the relatively small number of patients did not allow for identification of all the potential prognostic factors which might influence patient outcomes.
In conclusion, the present findings supported the application of texture analysis in identifying the iCC patients who would best respond and benefit from radioembolisation. Hypervascular tumours with a more homogeneous uptake of iodine contrast in the arterial phase were those most likely to be effectively treated by TARE and, of particular importance, this hypervascularisation could be quantified by Radiomics.