Abstract
Background
Currently, there are no methods to identify patients with an increased risk of liver metastases to guide patient selection for liver-directed therapies. We tried to determine whether quantitative image features (radiomics) of the liver obtained from preoperative staging CT scans at the time of initial colon resection differ in patients that subsequently develop liver metastases, extrahepatic metastases, or demonstrate prolonged disease-free survival.
Methods
Patients who underwent resection of stage II/III colon cancer from 2004–2012 with available preoperative CT scans were included in this single-institution, retrospective case-control study. Patients were grouped by initial recurrence patterns: liver recurrence, extrahepatic recurrence, or no evidence of disease at 5 years. Radiomic features of the liver parenchyma extracted from CT images were compared across groups.
Results
The cohort consisted of 120 patients divided evenly between three recurrence groups, with an equal number of stage II and III patients in each group. After adjusting for multiple comparisons, 44 of 254 (17%) imaging features displayed different distributions across the three patient groups (p < 0.05), with the clearest distinction between those with liver recurrence and no evidence of disease. Increased heterogeneity in the liver parenchyma by radiomic analysis was protective of liver metastases.
Conclusions
CT radiomics is a promising tool to identify patients at high risk of developing liver metastases and is worthy of further investigation and validation.
Colorectal cancer (CRC) is the third most common cause of cancer death in the United States, and approximately 140,000 new cases are diagnosed annually.1 Staging for CRC includes cross-sectional imaging for evidence of distant metastases.2 In the absence of synchronous disseminated disease, current guidelines recommend patients undergo curative-intent resection of the primary tumor. After resection, patients with stage II or stage III CRC typically receive adjuvant chemotherapy, although this approach is debated for patients with stage II disease.3–6 Even with modern chemotherapy regimens, patients with stage II or III CRC develop metastases in 10–30% of cases.7–9 Predicting which patients will have distant recurrence remains challenging.10,11 In CRC, the liver is the most common, and often only, site of recurrence.12,13 Identification of patients at increased risk of liver-specific metastases would allow adjuvant therapies targeting the hepatic parenchyma. Indeed, preemptive hepatic regional chemotherapy has shown to be effective in a few randomized trials.14,15
Radiomics is an emerging field in which medical images are converted to mineable data by automated extraction of quantitative imaging features.16 Quantitative imaging features represent radiographic enhancement patterns at the pixel-level and may reveal information about the liver microenvironment associated with hepatic recurrence before visible evidence of disease. Quantitative image analysis has previously been applied to CRC, although only with respect to the primary colon tumor, liver metastases (CRLM) and surrounding hepatic parenchyma, and future liver remnant before hepatic resection.17–21 These prior studies often involved a very small sample size and limited, if any, evaluation of computed tomography (CT) images in patients without metastatic disease. Therefore, there is a paucity of studies that have thoroughly evaluated quantitative imaging features of the liver in a cohort of patients with stage II and III CRC and extensive clinical follow-up.
The purpose of this project was to determine whether quantitative imaging features of the liver obtained from preoperative staging CT scans at the time of colon resection are different between stage II/III CRC patients that develop liver metastases, extrahepatic metastases, or demonstrate prolonged disease-free survival (DFS). This work is a first step towards the development of radiomics for the prediction of hepatic-specific recurrence in patients with CRC.
METHODS
Patients
This retrospective study was undertaken with approval from the Institutional Review Board with a waiver of informed consent. All patients who undergo colon or liver surgery at Memorial Sloan Kettering Cancer Center (New York, NY) are recorded in an institutional database. The institutional database was queried for all patients with resection of stage II or III colon cancer from 2004–2012. Ninety-day postoperative deaths and patients with another stage IV malignancy were excluded. Chart review was conducted for additional clinicopathologic variables and to determine the specific site and timing of recurrence. From this consecutive series, patients with a contrast-enhanced CT scan before the date of colon resection were selected. To minimize the impact of prior chemotherapy on quantitative image analysis of the liver, patients with neoadjuvant chemotherapy or chemotherapy administration within the past year were excluded. Similarly, due to variable treatment strategies, our initial query did not include patients with rectal cancer (distal 16 cm). The location of the primary tumor was grouped based on the location to the splenic flexure.22,23 Overall stage was confirmed and reported in accordance with current AJCC guidelines.2
Study Design
A case-control study design was implemented. Patients were grouped based on initial recurrence patterns: liver recurrence, extrahepatic recurrence, and no evidence of disease (NED). A patient was considered NED when there was a minimum of 5 years of DFS following colon resection. This time interval ensured that all patients considered NED had sufficient follow-up to ensure that recurrence at any point in their life was extremely unlikely.24 Liver-specific first recurrence was defined as patients with a diagnosis of CRLM as the only site of recurrence for at least 6 months from the date of diagnosis of CRLM. Patients with extrahepatic metastases were defined as those who developed stage IV disease at a location outside the liver and were without CRLM for a minimum of 6 months after the initial diagnosis of metastatic disease. Overall, three patients with initial extrahepatic metastases developed CRLM later in clinical follow-up beyond 6 months. The total number of patients was selected based on the maximum number available in the smallest group and stratified by stage. This process ensured an equal number of stage II and III patients within each group.
Quantitative Image Analysis
All patients had a preoperative, contrast-enhanced, portal venous phase CT scan within 2 months before the date of colon resection. CT images were transferred from the picture archiving and communication system (PACS) to the local workstation for image processing.
Standard image processing techniques were applied to extract liver parenchyma from the surrounding structures for quantitative analysis. The liver and vessels were semiautomatically segmented from the CT scans using Scout Liver (Analogic Corporation, Peabody, MA). The vessels were subtracted from the liver volume leaving only the parenchyma for image analysis. A set of 254 standard imaging features, representing heterogeneity, were extracted from the liver parenchyma region using gray-level co-occurrence matrices (GLCM), run-length matrices (RLM), local binary patterns (LBP), fractal dimension (FD), intensity histogram (IH), and angle co-occurrence matrices (ACMs) by a software developed by our group.25–32 These individual features represent heterogeneity in different ways. For example, GLCM measures the spatial distribution of pixels in an image, whereas ACM encodes the orientation of tissue patterns. The 254 features include 19 from each GLCM, Angle ACM1, and Angle ACM2; 11 RLM features; 5 intensity histogram-based features; 127 LBP-based features, and 54 FD-based features (Supplementary Material). Each feature was extracted from all slices of each CT image using MATLAB R2015a (The MathWorks, Inc., Natick, MA) and averaged to obtain a single value for the scan.
Statistical Analysis
Demographic information and clinicopathologic variables were described using counts and percentages for categorical variables and medians and ranges for continuous variables. Clinicopathologic factors between groups were compared with the chi-square test or Kruskal-Wallis test, as appropriate.
Initially, univariate comparisons tested if the distribution for each standardized imaging feature was the same across the three groups (liver recurrence, extrahepatic recurrence, and NED at 5 years). For these comparisons, Kruskal-Wallis tests were utilized to determine whether these three groups were statistically different with respect to a given feature. Wilcoxon rank-sum tests were used to assess pair-wise differences between groups (liver vs. NED, liver vs. extrahepatic disease, and NED vs. extrahepatic). In all analyses, significance was determined using a 5% alpha-level, after adjusting for multiple hypotheses using a bootstrap null distribution.33 Multidimensional scaling (MDS) was utilized to visualize differences between groups for selected imaging features. Finally, the Wilcoxon rank-sum test statistic and the area under the curve (AUC) provided measures of how accurately the distributions for each imaging feature were separated between any two groups. Data analysis was performed using R version 3.3.1.
RESULTS
Patient Characteristics
Overall, 981 patients met the inclusion criteria of undergoing resection of stage II or III colon cancer between 2004–2012. From this group, 588 patients (59.9%) had a portal venous phase pre-resection CT scan available in the local PACS and formed the initial cohort for patient selection. Within this cohort, 114 patients (19.4%) had a documented recurrence. In total, 80 patients with recurrence (n = 40 liver-only, n = 40 extrahepatic) based on the minimum number available in each group and stratified by stage to ensure equal inclusion of stage II (n = 40) and III (n = 40) tumors. The NED group (n = 40) consisted of a random sample of 40 patients (balanced between stage II [n = 20] and III [n = 20]) with 5-year DFS. Thus, the final study population for quantitative image analysis (n = 120) consisted of three groups (each n = 40)—liver recurrence, extrahepatic recurrence, and NED—with an equal number of patients with stage II and III tumors in each group.
Table 1 details the demographic and clinicopathologic variables of patients. The groups were designed to have an equal distribution of patients with stage II and III disease, and there were no statistically significant differences between groups with respect to gender, location of primary tumor, previous cancer, previous chemotherapy, grade, lymphovascular invasion, perineural invasion, or use of adjuvant chemotherapy. The only significant difference between groups was a younger median age in the NED group (p = 0.039).
Table 1.
Patient characteristics
| Liver recurrence | Extrahepatic recurrence | 5-year no evidence of disease (NED) | Total | ||
|---|---|---|---|---|---|
| (n = 40) | (n = 40) | (n = 40) | (n = 120) | p | |
| Gender | |||||
| Male | 18 (45%) | 17 (43%) | 19 (48%) | 54 (45%) | 0.904 |
| Female | 22 (55%) | 23 (58%) | 21 (53%) | 66 (55%) | |
| Median age at diagnosis, yr (range) | 65.5 (37.0 – 91.0) | 64.1 (37.0 – 90.0) | 59.0 (29.0 – 80.0) | 62.5 (29.0 – 91.0) | 0.039 |
| Median BMI, kg/m2 (range) | 26.3 (20.2 – 43.9) | 26.0 (18.7 – 37.7) | 28.6 (16.2 – 52.4) | 26.9 (16.2 – 52.4) | 0.247 |
| Location of primary | |||||
| Right | 17 (43%) | 19 (48%) | 23 (58%) | 57 (48%) | 0.393 |
| Left | 23 (58%) | 21 (53%) | 17 (43%) | 63 (53%) | |
| Previous cancer | |||||
| No | 33 (83%) | 28 (70%) | 31 (78%) | 92 (77%) | 0.413 |
| Yes | 7 (18%) | 12 (30%) | 9 (23%) | 28 (23%) | |
| Previous chemotherapy | |||||
| No | 38 (95%) | 38 (95%) | 37 (93%) | 113 (94%) | 0.859 |
| Yes | 2 (5%) | 2 (5%) | 3 (8%) | 7 (6%) | |
| Preoperative CEA, median (range) | 4.15 (0.80 – 201.00) | 5.00 (0.70 – 29.300) | 3.00 (0.60 – 19.50) | 3.90 (0.60 – 201.00) | 0.478 |
| Smoking status | |||||
| Never smoker | 23 (58%) | 22 (55%) | 23 (58%) | 68 (57%) | 0.801 |
| Former smoker | 15 (38%) | 14 (35%) | 12 (30%) | 41 (34%) | |
| Current smoker | 2 (5%) | 4 (10%) | 5 (13%) | 11 (9%) | |
| T stage | |||||
| T1 | 2 (5%) | 1 (3%) | 2 (5%) | 5 (4%) | 0.449 |
| T2 | 1 (3%) | 2 (5%) | 0 (0%) | 3 (3%) | |
| T3 | 30 (75%) | 25 (63%) | 32 (80%) | 87 (73%) | |
| T4 | 7 (18%) | 12 (30%) | 6 (15%) | 25 (21%) | |
| Lymph node stage | |||||
| No | 20 (50%) | 20 (50%) | 20 (50%) | 60 (50%) | 0.912 |
| N1 | 12 (30%) | 11 (28%) | 14 (35%) | 37 (31%) | |
| N2 | 8 (20%) | 9 (23%) | 6 (15%) | 23 (19%) | |
| TNM stage | |||||
| Stage II | 20 (50%) | 20 (50%) | 20 (50%) | 60 (50%) | 1.000 |
| Stage III | 20 (50%) | 20 (50%) | 20 (50%) | 60 (50%) | |
| Grade | |||||
| Moderately differentiated | 34 (85%) | 31 (78%) | 30 (75%) | 95 (79%) | 0.518 |
| Poorly differentiated | 6 (15%) | 9 (23%) | 10 (25%) | 25 (21%) | |
| Lymphovascular invasion | |||||
| No | 18 (45%) | 19 (48%) | 20 (50%) | 57 (48%) | 0.905 |
| Yes | 22 (55%) | 21 (53%) | 20 (50%) | 63 (53%) | |
| Perineural invasion | |||||
| No | 28 (70%) | 24 (60%) | 33 (83%) | 85 (71%) | 0.177 |
| Yes | 12 (30%) | 15 (38%) | 7 (18%) | 34 (28%) | |
| Unknown | 0 (0%) | 1 (3%) | 0 (0%) | 1 (1%) | |
| Adjuvant chemotherapy | |||||
| No | 19 (48%) | 16 (40%) | 12 (30%) | 47 (39%) | 0.274 |
| Yes | 21 (53%) | 24 (60%) | 28 (70%) | 73 (61%) |
Comparison of Quantitative Imaging Features between Recurrence Groups
After adjusting for multiple comparisons, 44 of 254 (17%) standardized imaging features displayed significantly different distributions across the three patient groups (Supplementary Table 1). Figure 1 illustrates the number of significant imaging features for each pair-wise comparison, grouped by feature type. After adjustment for multiple comparisons, 2, 45, and 8 standardized imaging features were significantly different between the liver versus extrahepatic, liver versus NED, and extrahepatic versus NED group comparisons, respectively. The majority of significant imaging features were the FD feature-type, with overlap across comparisons (Fig. 1).
Fig. 1.
Number and type of significant imaging features at 5% alpha-level using Wilcoxon rank sum tests with an adjustment for multiple comparisons
The 45 significant features from the liver versus NED group comparison were visualized using an MDS plot (Fig. 2). The relative distance between 2D points in the MDS plot indicates that the liver versus NED data points center near the origin. This pattern suggests the two study groups can be distinguished using this set of image features. Figure 3 provides a representative example of the visual differences in the liver parenchyma rendered with pixel detail at the same window level (bottom right) for the three study groups.
Fig. 2.
Multidimensional scaling plot of the 45 significant features for the liver recurrence and NED group comparison
Fig. 3.
Representative images of pixel-level heterogeneity differences in the three study groups, rendered with the same CT window
Imaging Features that Distinguish Recurrence Groups
For each of the three comparisons, we calculated the AUC statistic, an indicator of how well a given feature discriminated between the two groups, for each of the 254 features. The deviation of AUC from 0.5, its null value, was plotted so that a value of 0 indicates no discrimination, and AUC values were ordered from worst to best to improve visualization (Supplementary Fig. 1). We found that the imaging feature that most strongly distinguished the liver versus NED groups was IH skewness, a feature that captures the asymmetry of the intensity histogram. Figure 4 shows a plot of the IH skewness values for all patients and demonstrates the spread of values in the liver recurrence versus NED groups. Higher skewness is apparent for the NED patients suggests that increased heterogeneity is protective of liver metastases.
Fig. 4.
Box-plot of skewness values for all CT slices for each individual patient. Positive skewness in general represents increased heterogeneity. The lower and upper box boundaries (in blue) correspond to the first and third quartiles. The line within the box represents the median. The red ‘+’ signs represents skewness values outside the box region. Higher skewness is apparent for the NED patients
Identification of Imaging Feature Set for Construction of a Prediction Model
Identification and standardization of a feature set would allow construction of a prediction model for recurrence. We considered an absolute centered AUC value of 0.2 (or an AUC value <0.3 or >0.7) as a threshold for selecting a set of imaging features. Using this cutoff for the liver versus NED comparison, 206 imaging features were dropped (48 remaining); the 45 significant features identified in the liver versus NED comparison (Fig. 1) were among these. In the liver versus extrahepatic and extrahepatic versus NED comparisons, 251 and 238 imaging features were dropped, respectively.
We used the ordered AUC values and identified the top twenty imaging features that strongly distinguish the study groups for each pairwise comparison (indices 235–254). Fourteen imaging features overlapped across the “20 best features” for the three pairwise comparisons (FD1_1, FD1_4, FD1_6, FD1_7, FD1_9, FD1_10, FD1_25, FD1_26, FD1_27, FD1_28, FD1_29, FD1_30, FD1_31, and IH_Skewness), and three of these features were significantly different in all three comparisons (IH_Skewness, FD1_6, and FD1_26). Skewness of the intensity histogram (IH_Skewness) showed the largest difference between the liver versus NED groups based on the AUC statistic. The liver and extrahepatic groups display similar distributions for IH Skewness values, spanning from −2 to 1, whereas the values for the NED group are densely concentrated within the range of 0 to 1 (Fig. 5, left panel). For FD1_6 (Fig. 5, middle panel), the liver recurrence patients cluster at lower, negative values near −1, whereas the values for the NED group display large variability, ranging from −1 to 3. Finally, the values of FD1_26 (Fig. 5, right panel) display a large range and variability for the liver and extrahepatic recurrence groups and conversely, a narrow (positive) range of values for the NED group.
Fig. 5.
Plot of IH skewness, FD1–6, and FD1–26, the most predictive feature for the three pairwise comparisons (Liver vs NED, Liver vs. Extra, and Extra vs. NED)
DISCUSSION
We hypothesize that CT radiomics can be used to at the time of initial colon resection to identify differences in hepatic parenchyma in patients that go on to develop liver metastases. Using a series of patients with stage II or III colon cancer, grouped based on recurrence patterns, we identified standardized imaging features extracted from pre-resection CT images—a time point when cross-sectional imaging had no identifiable hepatic metastases—that displayed significantly different distributions between patients that later developed liver recurrence, extrahepatic recurrence, or were free of disease at 5 years. We found the clearest distinction between imaging features for patients with liver recurrence versus those with NED. The imaging features identified here are promising predictors of hepatic metastases that may be used to develop a multivariate model in a larger dataset. This study demonstrates the potential of quantitative image analysis to identify and predict patients at risk of hepatic recurrence before metastases are visible.
Certain imaging features (Fig. 1) could clearly distinguish the liver recurrence and NED groups, especially the FD-type features and skewness of the intensity histogram. FD-type features are just one of many radiomic features, each of which weighs an aspect of heterogeneity differently. In general, the liver parenchyma in CT images of patients NED at 5 years was more heterogeneous than patients who developed CRLM. Because positive skewness in the distribution of the intensity histogram broadly represents increased heterogeneity, higher skewness for NED patients suggests that increased heterogeneity could be related to underlying differences in the hepatic parenchyma protective of liver metastases.34 This observation is consistent with our group’s previous evaluation of radiomic features of the future liver remnant and survival in patients undergoing hepatic resection of CRLM, in which improved survival was associated with a heterogeneous future liver remnant.19 The underlying biology for this observation remains unclear, but accumulating evidence suggests that heterogeneous liver parenchyma is associated with a reduced likelihood of developing metastases.
This is the largest study to date to look at quantitative imaging features of the at-risk hepatic parenchyma in patients with CRC and no visible metastases. Preoperative imaging of patients with stage II and III CRC that develop documented hepatic recurrence is a large undertaking that requires a considerable number of patients and follow-up. Previous studies have been limited by small sample sizes and unclear inclusion criteria. An additional strength of our study was the use of two comparison groups (extrahepatic recurrence and 5 years NED) with matched clinical characteristics. Rao et al. previously evaluated quantitative imaging features of the liver in 29 patients with CRC, only 4 of whom had metachronous CRLM within 18 months.20 The sample size in this study was too small to draw any conclusions regarding predictive quantitative imaging features of the liver in patients who do or do not develop recurrence. In a separate study, Ganeshan et al. reported increased heterogeneity in patients without hepatic metastases (n = 15) compared with those with CRLM (n = 8), which is consistent with our findings; however, their study utilized noncontrast CT scans and thus direct comparisons cannot be made with our findings.21
Our results, which will require validation, represent an application of radiomics that may help clinicians predict recurrence and treat patients at high risk of developing CRLM. With this in mind, our study evaluated the imaging features that could be prioritized for creation of a multivariate model. Survival and outcomes were incorporated into the study design, but a future prospective evaluation could collect specific imaging features and associated recurrence patterns. The scope and expense of that study would be significant but could combine clinicopathologic characteristics used in nomograms with radiomic and genomic information.10 The potential exists for development of a robust and accurate predictive model.
This study has several limitations. First, it is a retrospective study at a large academic medical center. Thus, there are inherent biases, and the results may not be applicable to all institutions. We purposefully selected patients based on their documented survival and outcome, but they may not represent all patients with liver recurrence, extrahepatic recurrence, or those with 5 years NED. It is likely that there is even more variation within these populations. However, this study is our best attempt to analyze quantitative imaging features of patients at the time of colon resection, and we used a carefully curated dataset specifically for this purpose. Patients with stage I CRC may develop recurrence, but it is a rare event. This dataset does not address those outliers. Furthermore, potential imaging confounders, such as recent chemotherapy, were intentionally excluded. In statistical terms, a portion of the extrahepatic group may develop CRLM later in their clinical course. Therefore, this group is potentially not a pure cohort, and it may have reduced our power to detect true differences between disease groups. Molecular testing for KRAS and BRAF mutations was not routinely performed for all patients with stage II and III CRC during the study period. A prospective study that incorporates molecular testing would allow further investigation into the correlation between radiomics and genomics. This study demonstrates the potential utilization of radiomics as a predictive tool for hepatic recurrence after colon resection and is a first step in the development of individualized prognosis and selection of therapies. An accurate prediction model for patient’s at increased risk of hepatic recurrence would help select patients for regional liver-directed therapy or even adjuvant systemic chemotherapy in patients that would not typically have received that treatment.
In summary, we identified radiomic features of the liver extracted from CT scans obtained at the time of initial colon resection that were discriminatory for patients who had subsequent hepatic recurrence versus those with NED at 5 years. CT radiomics is a promising tool to identify patients at high risk of developing CRLM; given the clinical implications of a predictive tool for hepatic metastasis, further investigation and validation in a larger cohort are warranted.
Supplementary Material
SYNOPSIS.
In this retrospective study, we found radiomic features extracted from preoperative CT scans at the time of colon resection varied between patients with subsequent hepatic recurrence versus those with no evidence of disease. CT radiomics is a promising tool to identify patients at high risk of developing liver metastases.
ACKNOWLEDGMENTS
This study was supported in part by NIH/NCI Cancer Center Support Grant P30 CA008748 and the Society for Memorial Sloan Kettering.
Footnotes
Publisher's Disclaimer: This Author Accepted Manuscript is a PDF file of an unedited peer-reviewed manuscript that has been accepted for publication but has not been copyedited or corrected. The official version of record that is published in the journal is kept up to date and so may therefore differ from this version.
Disclosures: None
REFERENCES
- 1.Siegel R, Desantis C, Jemal A. Colorectal cancer statistics, 2014. CA Cancer J Clin. 2014;64(2):104–17. [DOI] [PubMed] [Google Scholar]
- 2.Edge SB, American Joint Committee on Cancer. AJCC cancer staging manual, 7th edn. New York: Springer; 2010. [DOI] [PubMed] [Google Scholar]
- 3.Andre T, Boni C, Mounedji-Boudiaf L, et al. Oxaliplatin, fluorouracil, and leucovorin as adjuvant treatment for colon cancer. N Engl J Med. 2004;350(23):2343–51. [DOI] [PubMed] [Google Scholar]
- 4.Benson AB 3rd, Schrag D, Somerfield MR, et al. American Society of Clinical Oncology recommendations on adjuvant chemotherapy for stage II colon cancer. J Clin Oncol. 2004;22(16):3408–19. [DOI] [PubMed] [Google Scholar]
- 5.Casadaban L, Rauscher G, Aklilu M, Villenes D, Freels S, Maker AV. Adjuvant chemotherapy is associated with improved survival in patients with stage II colon cancer. Cancer. 2016;122(21):3277–87. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.O’Connor ES, Greenblatt DY, LoConte NK, et al. Adjuvant chemotherapy for stage II colon cancer with poor prognostic features. J Clin Oncol. 2011;29(25):3381–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Manfredi S, Bouvier AM, Lepage C, Hatem C, Dancourt V, Faivre J. Incidence and patterns of recurrence after resection for cure of colonic cancer in a well defined population. Br J Surg. 2006;93(9):1115–22. [DOI] [PubMed] [Google Scholar]
- 8.Pugh SA, Shinkins B, Fuller A, Mellor J, Mant D, Primrose JN. Site and stage of colorectal cancer influence the likelihood and distribution of disease recurrence and postrecurrence survival: data From the FACS Randomized Controlled Trial. Ann Surg. 2015. [DOI] [PubMed] [Google Scholar]
- 9.Shah MA, Renfro LA, Allegra CJ, et al. Impact of patient factors on recurrence risk and time dependency of oxaliplatin benefit in patients with colon cancer: analysis from modern-era adjuvant studies in the Adjuvant Colon Cancer End Points (ACCENT) Database. J Clin Oncol. 2016;34(8):843–53. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Weiser MR, Landmann RG, Kattan MW, et al. Individualized prediction of colon cancer recurrence using a nomogram. J Clin Oncol. 2008;26(3):380–5. [DOI] [PubMed] [Google Scholar]
- 11.Gill S, Loprinzi C, Kennecke H, et al. Prognostic web-based models for stage II and III colon cancer: A population and clinical trials-based validation of numeracy and adjuvant! online. Cancer. 2011;117(18):4155–65. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Amri R, Bordeianou LG, Sylla P, Berger DL. Variations in Metastasis Site by Primary Location in Colon Cancer. J Gastrointest Surg. 2015;19(8):1522–7. [DOI] [PubMed] [Google Scholar]
- 13.Manfredi S, Lepage C, Hatem C, Coatmeur O, Faivre J, Bouvier AM. Epidemiology and management of liver metastases from colorectal cancer. Ann Surg. 2006;244(2):254–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Sadahiro S, Suzuki T, Ishikawa K, et al. Prophylactic hepatic arterial infusion chemotherapy for the prevention of liver metastasis in patients with colon carcinoma: a randomized control trial. Cancer. 2004;100(3):590–7. [DOI] [PubMed] [Google Scholar]
- 15.Xu J, Zhong Y, Weixin N, et al. Preoperative hepatic and regional arterial chemotherapy in the prevention of liver metastasis after colorectal cancer surgery. Ann Surg. 2007;245(4):583–90. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Lambin P, Rios-Velazquez E, Leijenaar R, et al. Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer. 2012;48(4):441–6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Ng F, Ganeshan B, Kozarski R, Miles KA, Goh V. Assessment of primary colorectal cancer heterogeneity by using whole-tumor texture analysis: contrast-enhanced CT texture as a biomarker of 5-year survival. Radiology. 2013;266(1):177–184. [DOI] [PubMed] [Google Scholar]
- 18.Lubner MG, Stabo N, Lubner SJ, et al. CT textural analysis of hepatic metastatic colorectal cancer: pre-treatment tumor heterogeneity correlates with pathology and clinical outcomes. Abdom Imag. 2015;40(7):2331–7. [DOI] [PubMed] [Google Scholar]
- 19.Simpson AL, Doussot A, Creasy JM, et al. Computed tomography image texture: a noninvasive prognostic marker of hepatic recurrence after hepatectomy for metastatic colorectal cancer. Ann Surg Oncol. 2017. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Rao SX, Lambregts DM, Schnerr RS, et al. Whole-liver CT texture analysis in colorectal cancer: Does the presence of liver metastases affect the texture of the remaining liver? United Eur Gastroenterol J. 2014;2(6):530–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Ganeshan B, Miles KA, Young RC, Chatwin CR. Texture analysis in non-contrast enhanced CT: impact of malignancy on texture in apparently disease-free areas of the liver. Eur J Radiol. 2009;70(1):101–10. [DOI] [PubMed] [Google Scholar]
- 22.Creasy JM, Sadot E, Koerkamp BG, et al. Right versus left: Impact of primary location on survival and cure in patients undergoing hepatic resection for metastatic colon cancer. J Clin Oncol. 2017;35(4_suppl):664–4. [Google Scholar]
- 23.Benedix F, Kube R, Meyer F, Schmidt U, Gastinger I, Lippert H. Comparison of 17,641 patients with right- and left-sided colon cancer: differences in epidemiology, perioperative course, histology, and survival. Dis Colon Rectum. 2010;53(1):57–64. [DOI] [PubMed] [Google Scholar]
- 24.Tan MC, Butte JM, Gonen M, et al. Prognostic significance of early recurrence: a conditional survival analysis in patients with resected colorectal liver metastasis. HPB. 2013;15(10):803–13. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Haralick RM, Shanmuga K, Dinstein I. Textural Features for Image Classification. Ieee T Syst Man Cyb. 1973;Smc3(6):610–621. [Google Scholar]
- 26.Tang XO. Texture information in run-length matrices. Ieee T Image Process. 1998;7(11):1602–9. [DOI] [PubMed] [Google Scholar]
- 27.Pietikainen M, Zhao GY, Hadid A, Ahonen T. Local Binary Patterns for Still Images. Comput Imaging Vis. 2011;40:13–47. [Google Scholar]
- 28.Mehta R, Egiazarian KO. Rotated Local Binary Pattern (RLBP)-Rotation Invariant Texture Descriptor. Paper presented at: ICPRAM2013
- 29.Buczkowski S, Kyriacos S, Nekka F, Cartilier L. The modified box-counting method: Analysis of some characteristic parameters. Pattern Recogn. 1998;31(4):411–8. [Google Scholar]
- 30.Al-Kadi OS, Watson D. Texture analysis of aggressive and nonaggressive lung tumor CE CT images. Ieee T Bio-Med Eng. 2008;55(7):1822–30. [DOI] [PubMed] [Google Scholar]
- 31.Chakraborty J, Rangayyan RM, Banik S, Mukhopadhyay S, Desautels JEL. Statistical measures of orientation of texture for the detection of architectural distortion in prior mammograms of interval-cancer. J Electron Imaging. 2012;21(3). [Google Scholar]
- 32.Zheng J, Chakraborty J, Chapman WC, et al. Preoperative prediction of microvascular invasion in hepatocellular carcinoma using quantitative image analysis. J Am Coll Surg. 2017. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Pollard KS, Dudoit S, van der Laan MJ. Multiple testing procedures. 2007:1–30. [Google Scholar]
- 34.Davnall F, Yip CS, Ljungqvist G, et al. Assessment of tumor heterogeneity: an emerging imaging tool for clinical practice? Insights Imaging. 2012;3(6):573–89. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.