R. Beets-Tan, Amsterdam, NL
The Netherlands Cancer Institute Department of RadiologyAuthor of 5 Presentations
SS 5.7 - CT texture analyses of colorectal liver metastases before and after thermal ablation can predict local tumour progression
Abstract
Purpose
To investigate whether CT texture analyses (CTTA) in patients with colorectal liver metastases (CRLM) are able to predict and identify local tumour progression (LTP) after thermal ablation.
Material and methods
We retrospectively included patients with up to 5 CRLM (size ≤3cm), who underwent RFA or MWA and had follow-up ≥6 months. An experienced reader manually delineated all the CRLM and ablation zones (AZ) on portal venous CT pre- and 4-8 weeks post-ablation. Texture parameters were extracted from the delineated volumes of interest (VOIs) with Pyradiomics, using different Laplacian of Gaussian (LoG) filters. Subsequently, texture parameters were compared between lesions with and without LTP, and between patients with new metastases and those who remained disease free.
Results
38/168 lesions (in 104 patients) developed LTP and 54 patients developed new metastases. Median follow-up time was 20 (range 6-134). Uniformity of CRLM pre-ablation was significantly higher in patients who developed new metastases than in those who remained disease free (mean 0.043 vs 0.039, p=0.029, unfiltered and mean 0.046 vs 0.042, p=0.066, LoG filter s0.5). Furthermore, mean grey-level intensity of the AZ (post-ablation) was significantly higher in lesions that developed LTP compared to those that did not (mean 1.18 vs 0.92, p=0.037, LoG filter σ0.5).
Conclusion
A higher uniformity of CRLM before ablation shows the potential for identifying patients at risk of developing new metastases. Additionally, AZ of lesions that develop LTP post-ablation shows a higher mean grey-level intensity. CTTA of both the CRLM (pre-ablation) and of the AZ (4-8 weeks after ablation) may, therefore, be predictive of the development of LTP and new metastases.
SS 7.1 - Machine learning-based analysis of CT radiomics model for the prediction of colorectal metachronous liver metastases
Abstract
Purpose
To develop and validate a CT-based radiomics model for the prediction of metachronous colorectal liver metastases (CRLM) at primary staging with a machine learning algorithm.
Material and methods
91 colorectal cancer (CRC) patients from three centers were included. Two groups were assessed: patients who had no evidence of CRLM at primary staging or during follow-up of ≥ 24 months (n=67) and patients who had no evidence of liver metastases at primary staging but developed metachronous CRLM < 24 months (n=24). After liver parenchyma segmentation (excluding vessels and benign liver lesions), 7,344 radiomics features were extracted from the primary staging CT per patient with PyRadiomics package. Patients were divided into a training set (n=54) and an independent validation set (n=37). Two predictive models were built based on only radiomics features and with the combination of clinical information, these models were compared regarding predictive value. Bayesian-optimized random forest with wrapper feature selection (5-fold cross-validation) was trained on the training set and optimized with sequential model-based optimization.
Results
The area under the curve (AUC) of the machine learning model based on CT radiomics feature performance in the training cohort was 0.88(95%CI: 0.85-0.92), and in the validation cohort it was 0.87(95%CI: 0.82-0.91) with sensitivity 0.88(95%CI 0.80-0.95) and specificity 0.72(95%CI 0.63-0.82). Adding clinical features to the radiomics model did not improve diagnostic performance.
Conclusion
A machine learning-based radiomics analysis of routine clinical CT imaging at diagnosis could provide valuable biomarkers to predict patients at risk for developing colorectal liver metastases.
SS 12.9 - Findings of sinusoidal obstruction syndrome on gadoxetic acid-enhanced MRI in patients with chemotherapy for colorectal liver metastases are poorly reproducible between radiologists
Abstract
Purpose
Neoadjuvant chemotherapy in patients with colorectal liver metastases (CRLM) may cause sinusoidal obstruction syndrome (SOS), potentially leading to increased morbidity after resection and decreased chemotherapy effect. Therefore, it is important to identify SOS. Studies have suggested to use gadoxetic acid-enhanced MRI (EOB-MRI) for the diagnosis of SOS. The purpose of our study was to assess the reproducibility of EOB-MRI to determine the presence and severity of SOS in patients treated with chemotherapy for CRLM.
Material and methods
32 patients treated with chemotherapy for CRLM (either oxaliplatin-based, oxaliplatin-based+bevacizumab, or non-oxaliplatin-based treatment), with available EOB-MRI scans in the hepatobiliary phase, were retrospectively included. The presence and severity of SOS were independently scored by three radiologists with varying experience (10-20 years of experience in EOB-MRI), who were blinded to the clinical data, using a 5-point scale (SOS score: 0=definitely not present to 4=definitely present). The interobserver agreement between readers was assessed with quadratic weighted kappa statistics.
Results
The mean time between chemotherapy completion and EOB-MRI scan was 1.1 (+/-1.4) months. Interobserver agreement was poor, with kappas ranging from 0.25 (95%CI 0.01-0.48) to 0.33 (95%CI 0.08-0.58). Furthermore, analysis of the SOS scores revealed that less experienced radiologists tended to score more equivocal scores (6.3% and 43.8% SOS score = 2), compared to a more experienced radiologist (3.1% SOS score=2).
Conclusion
The assessment of SOS on hepatobiliary phase EOB-MRI in patients treated with chemotherapy for CRLM shows a poor inter-observer agreement for the diagnosis of SOS. This questions the clinical value of hepatobiliary phase EOB-MRI to detect SOS.
SS 14.1 - Prognostic value of preoperative diffusion-weighted MRI in colorectal peritoneal carcinomatosis patients using leading prediction models
Abstract
Purpose
The purpose of the study is to compare the performance of diffusion-weighted (DW) MRI to predict complete cytoreductive surgery (CRS), disease-free survival and overall survival using 5 different prediction models for hyperthermic intraperitoneal chemotherapy (HIPEC) candidates with colorectal cancer.
Material and methods
Between February 2016 and October 2017, patients with colorectal peritoneal carcinomatosis considered for CRS/HIPEC who underwent DW-MRI for preoperative staging were included. DW-MRI images were evaluated, retrospectively, to determine the PCI. Relevant clinical parameters were obtained from patient files. Five scoring models were used; the Peritoneal Surface Disease Severity Score (PSDSS), Region Count, Simplified Peritoneal Cancer Index (SPCI), Peritoneal Cancer Index (PCI), and the Colorectal Peritoneal Metastases Prognostic Surgical Score (COMPASS). The performance of the prediction models was assessed by receiver operator characteristics (ROC) analysis and Cox-proportional analysis.
Results
Ninety-one patients were included. The mean age was 62.30 (±10.93) and 49% (45/92) of the patients were female. Fifty-three (58%) patients underwent successful CRS/HIPEC. For 38 patients, CRS/HIPEC was deemed infeasible, mostly due to extensive peritoneal carcinomatosis (23/38). ROC curve analysis for predicting a complete cytoreduction showed AUCs for PSDSS, region count, SPCI, PCI and COMPASS of 0.79, 0.81, 0.82, 0.85 and 0.86, respectively. PSDSS, region count, SPCI, PCI, and COMPASS showed significant hazard ratios for overall survival of 1.19, 1.69, 1.22, 1.12, and 1.02. None were found for disease-free survival.
Conclusion
DW-MRI seems to be able to select patients with colorectal peritoneal carcinomatosis eligible for CRS/HIPEC. All five predictive models are predictive of complete cytoreduction and overall survival, not of disease-free survival.