Our institutional review board approved this retrospective radiographic study and waived the requirement for informed patient consent. The patient cohort in the current study partially overlapped with that in a previous publication [21]. A database of patients (n = 242) who underwent surgery for CLM between January 2008 and June 2014 was reviewed to identify patients who underwent preoperative MRI and curative surgery for CLM after preoperative chemotherapy. Figure 1 summarizes patient inclusion and exclusion criteria. Patients who had not undergone preoperative chemotherapy were excluded. The remaining patients had received preoperative (neoadjuvant) fluorouracil-based chemotherapy, mostly in combination with molecular-target drugs. Patients with a history of chemotherapy for various reasons before referral to our tertiary center were not excluded. Patients who were not evaluated by MRI were excluded. To exclude the influence of surgery on prognosis, patients who failed to undergo curative surgical resection (R2 resection) were also excluded. Patients with microscopically positive surgical margins (R1 resection) were included because several reports have found that R1 resection does not adversely affect survival if accompanied by effective perioperative chemotherapy [22, 23]. A patient with CLMs too small for evaluation of intratumoral fat on MR images was excluded. Ultimately, 59 patients (33 men, 26 women; median age, 62 years old; age range, 28–79 years old) with 452 CLMs were selected for analysis. All patients were subjected to preoperative MRI to determine the number and location of CLMs. The median interval from preoperative MRI to surgery was 18 days (range, 4–68 days). Of the 59 patients, 20 had also undergone MRI before chemotherapy (Fig. 1). The patients were followed up for CLMs by CT evaluation after every four to six cycles of chemotherapy and before hepatic resection. The median interval from preoperative CT to surgery was 13 days (range, 1–120 days). CT data obtained after chemo-therapy were not available in one patient. In five patients who were not evaluated for RECIST response, neither CT nor MRI data acquired within a month before chemotherapy were available. Preoperative serologic data of the patients (serum cholesterol, triglyceride, hemoglobin A1c, carcinoembryonic antigen [CEA], and carbohydrate antigen 19–9 [CA 19–9] levels) were also collected.

MRI studies for hepatic lesions were performed using 1.5-T or 3-T scanners. Dual-echo T1-weighted gradient-recalled echo MR images were acquired for all patients. Table 1 summarizes the scanners and scanning parameters; the variation in parameters was because of differences in scanners, system versions, and body size of patients. Subtraction images of opposed- and in-phase images were constructed for qualitative analysis of fat. Axial and coronal T2-weighted, fat-suppressed T2-weighted, and DW images were also acquired. Dynamic contrast-enhanced MRI was performed using Gd-EOB-DTPA or superparamagnetic iron oxide. All CT images were acquired using MDCT scanners with a tube voltage of 120 kV and reconstructed at a 5-mm slice thickness. Single-phase (late phase) or four-phase (unenhanced, arterial, portal, and late phases) contrast-enhanced images were acquired. Images were transferred to the PACS (Centricity, GE Healthcare) for analysis.

All images were interpreted by two radiologists (with 5 and 6 years of experience in abdominal imaging) blinded to the clinical and pathologic findings and each other's results. The interpretation of one reader was used only for assessment of inter-rater reliability to confirm the accuracy of evaluation. MR images were evaluated for the following features: number and maximum tumor diameter of CLMs; presence or absence of intratumoral fat on preoperative and pre-chemotherapeutic dual-echo gradient-recalled echo MR images before chemo-therapy (qualitative evaluation by subtraction of opposed-phase from in-phase images); proportion of fat-containing lesions (all CLMs or not); pattern of fat deposition in CLMs (diffuse or focal); and presence or absence of fatty liver. As a quantitative index of the amount of intratumoral fat, the fat signal fraction (FSF) of CLMs was defined by the following formula described in previous reports [24, 25]:

where SI_IP and SI_OP indicate the signal intensities of the lesion on in-phase and opposed-phase T1-weighted images, respectively. Signal intensities of CLMs containing fat were determined using an elliptic ROI, which was drawn as large as possible (minimum area, 12 mm²) to cover the densest region of fat deposition in CLMs in each patient. In case of focal or heterogeneous intratumoral fat deposition, the ROI was drawn focally to cover only the region with fat deposition. To determine which lesion had the highest FSF value, FSF was measured for all lesions having intratumoral fat in each patient. Figures 2 and 3 show the method of ROI placement in CLMs. Both SI_IP and SI_OP were measured three times, and the mean values were used for analysis. Tumor size of 7 mm or greater was defined by measurable lesions, because the greatest slice spacing and thickness were 7.8 and 6.0 mm, respectively. Additionally, CT images were evaluated for the following features: number and maximum tumor diameter of CLMs; presence or absence of intratumoral calcification; and morphologic response grade for evaluation of response to chemotherapy in CLMs [9]. Response to chemotherapy was determined in accordance with RECIST version 1.1 on the basis of enhanced CT or MRI findings [8].

All histologic analyses were performed by a board-certified pathologist (8 years of experience in gastrointestinal pathology) blinded to the clinical data. Primary colorectal tumors were classified into six histopathologic types in accordance with the Japanese classification of colorectal carcinoma [26]. Specimens of CLMs from patients who did not undergo chemotherapy between final MRI and hepatectomy (56 of 59 patients) were selected for microscopic examination. The lesions were classified—on the basis of tumor viability evaluated in H and E–stained sections—into groups separated by 5%, as described previously [27]. Special fat staining could not be performed because, in all of the specimens, lipids had been extracted by routine fixation procedures.

Statistical analysis was performed using the EZR software (Saitama Medical Center, Jichi Medical University), a graphical user interface for R software (The R Foundation for Statistical Computing) [28]. Categoric variables were compared using Fisher exact test. Correlation between tumor FSF and percent pathologic tumor viability or tumor size was evaluated with Spearman rank correlation analysis. Overall survival (OS) and recurrence-free survival (RFS) were calculated from the date of hepatic resection using the Kaplan–Meier method and compared using the log-rank test, in which the cutoff value of FSF was determined to minimize the p value. Data from patients lost to follow-up were censored at the time of last follow-up. Predictive and prognostic factors were identified by multivariate analysis using the Cox proportional hazard model with stepwise backward selection and preceding backward elimination of variables identified as relatively significant (p < 0.15) on univariate analysis. Similarly, factors contributing to fat deposition in CLMs after preoperative chemo-therapy were also identified by multivariate logistic regression analysis. All p values were two-sided, and statistical significance was set at p < 0.05. For assessment of interrater reliability, Cohen coefficient kappa (κ) for categoric variables, Kendall coefficients of concordance (W) for ordered variables, and Ebel intraclass correlation coefficients (ICCs) for continuous variables were calculated.

Table 2 presents the demographic and clinicopathologic characteristics of the patients enrolled in this study (n = 59). Of the 59 patients, 12 (20%) had extrahepatic lesions (lungs; n = 6; distant lymph nodes, n = 6; spleen, n = 1), all of which were radically resected during or after hepatectomy. One patient had lung and splenic metastases. The median follow-up period was 36.6 months (range, 1.1–106.6 months). Twenty-five patients (42%) died during the study period, two (3%) within 90 days of surgery. Of the 59 patients, 44 (75%) developed tumor recurrence (local or metastatic).

Interrater reliability for each evaluation parameter was excellent: presence or absence of intratumoral fat, κ = 0.97; fatty liver, κ = 0.84; tumor calcification, κ = 0.96; RECIST response, W = 0.95; morphologic response on CT images, W = 0.91; FSF, ICC = 0.92; number of CLMs, ICC = 0.99; and largest tumor size, ICC = 0.94.

On MR images, intratumoral fat in CLMs was qualitatively detected in 32 of 59 (54%) patients after chemotherapy. In 20 patients who also underwent MRI before chemotherapy, intratumoral fat in CLMs was qualitatively detected in no patients (0%) before and nine patients (45%) after chemotherapy. In 32 patients with intratumoral fat in CLMs, the patterns of fat deposition were either focal (n = 9/32) or diffuse (n = 23). Thirteen patients exhibited fat deposition in all CLMs.

The results of Kaplan-Meier curve analysis indicated poorer OS among patients with tumor FSF values of 12% or more (log-rank test, p = 0.02; Fig. 4). This cutoff value was determined to minimize the p value. However, OS was not associated with qualitatively evaluated presence or absence of intratumoral fat (log-rank test, p = 0.34). Overall survival in patients with focal fat deposition in CLMs was lower than that in patients with diffuse or no fat deposition in CLMs (log-rank test, p = 0.048). The proportion of fat-containing lesions had no significant influence on OS (log-rank test, p = 0.64). The following preoperative factors were identified as independent predictors of unfavorable prognosis by multivariate analysis (Table S1, which can be viewed in the AJR electronic supplement to this article, available at www.ajronline.org): number of CLMs ≥ 5, FSF ≥ 12%, age ≥ 65 years, and RECIST 1.1 response indicating progressive or stable disease. With regard to RFS, the following preoperative factors were identified as independent predictors of unfavorable prognosis by multivariate analysis: number of CLMs ≥ 5 (HR, 5.25; 95% CI, 2.52–10.93; p < 0.0001); patient age ≥ 65 years (HR, 3.19; 95% CI, 1.57–6.49, p = 0.001); RECIST 1.1 response indicating progressive or stable disease (HR, 2.07; 95% CI, 1.04–4.12; p = 0.04); and morphologic response, group 3 (HR, 1.97; 95% CI, 1.01–3.86; p = 0.046). Neither qualitatively evaluated presence or absence of intratumoral fat nor FSF of 12% or more were significant predictors of poor RFS.

In the 56 patients (95% of the cohort) who did not undergo chemotherapy between final MRI and hepatectomy, FSF was found to be weakly and inversely correlated with percent tumor viability (Spearman rank correlation; ρ = −0.36; p = 0.007; Fig. 5A). Histologic findings revealed that necrosis, including infarctlike necrosis, occurred more frequently in lesions in the group with high FSF than in those in the group with low FSF (Fig. 6).

Although FSF was not correlated with the size of the tumor in which the FSF was measured (Spearman rank correlation, p = 0.06), it was weakly correlated with largest tumor size (Spearman rank correlation; ρ = 0.35; p = 0.006; Fig. 5B). In 49 patients (83%), the largest CLM lesion and the CLM lesion with the densest intratumoral fat deposition in which FSF was measured were the same. Multivariate analysis revealed that largest tumor size of 5 cm or more, tumor calcification, and history of cetuximab or panitumumab administration were predictive factors of tumor FSF being 12% or more (Table S2, which can be viewed in the AJR electronic supplement to this article, available at www.ajronline.org). Neither histologic type of the primary tumor nor serum CA19–9, CEA, total cholesterol, triglyceride, or hemoglobin A1c levels were significant predictive factors of tumor FSF ≥ 12%.