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
Methods and materials
Clinical and radiological data from 55 consecutive patients with iCCs who had undergone TARE between 2010 and 2018 were retrospectively reviewed.
All patients underwent quadri-phase contrast-enhanced CT (CECT) imaging within 30 days prior to treatment.
Images obtained in the late arterial, venous and delayed phases were used for the texture analysis. For each patient, a Volume of Interest (VOI) of the whole target lesion for each contrast phase was obtained by means of semi-automated segmentation, manually refined to exclude surrounding non-tumoural areas by two radiologists with 5 (A.B.) and 14 (C.M.) years of experience in hepatic imaging. The feature extraction was carried out using the LifeX software [8]. After TARE, the patients were regularly evaluated at 1 and 3 months, and at 3-months intervals.
Tumour response was assessed using tumour size criteria (Response Evaluation Criteria In Solid Tumours, RECIST 1.1) [9]. The objective response (OR) of the target lesion represented the primary outcome measure and was defined as the sum of complete response (CR) and partial response (PR).
Overall survival (OS) was defined as the time elapsed from TARE to death. Progression-free survival (PFS) was defined as the time elapsed from TARE to evidence of tumour progression, either intra or extrahepatic, or death.
The categorical variables were reported as number of cases and percentages and were compared using Fisher’s exact test when necessary. The continuous variables were reported as medians and interquartile ranges (IQRs: 25th and 75th percentiles), and differences between the subgroups were compared using the Mann-Whitney test. The survival rates were calculated using the Kaplan–Meier estimator. The radiomic features showed high correlation to each other at Spearman rank analysis. The Least Absolute Shrinkage and Selection Operator (LASSO) method was applied to logistic regression in order to select the most useful prognostic imaging features for the best OR of the target lesion after TARE.
A radiomic score was calculated for each patient as the linear combination of selected features weighted by their respective coefficients. For eventual clinical usefulness, the number of texture features was reduced to a minimum by entering only variables with a p“radiomic signature” was finally tested together with the clinical variables for predicting PFS and OS by means of simple and multivariable Cox regression analyses.