Key Findings:
- Feasibility Study
- Evaluation Study
-
- Unfiltered Texture Analysis
-
- Non-Contrast CT
- Contrast-Enhanced CT
- Filtered Texture Analysis
-
- Non-Contrast CT
- Contrast-Enhanced CT
3. Biological Correlates For Liver Texture
1. Feasibility Study
Tables 2, 3 and 4 indicates the regression values along with their significance for the unfiltered and filtered texture analysis for mean grey-level intensity, entropy and uniformity respectively against HPI and patient survival. The feasibility study demonstrated that unfiltered texture demonstrated poor correlations against hepatic perfusion index (HPI) and patient survival. Predominantly coarse liver texture quantified as entropy during portal phase correlated with both HPI (r = - 0.50, p = 0.0074, see Figure 4) and patient survival (r = 0.49, p = 0.0095) – Table 3 whereas mean grey-level intensity correlated with only HPI – Table 2 and uniformity correlated with only patient survival – Table 4.
2. Evaluation Study
In the evaluation study, based on the PET data and CT images patients were divided into three diagnostic groups
Group A: patients with no evidence of tumour (n = 15)
Group B: patients with extra-hepatic tumour only (n = 9)
Group C: patients with hepatic metastases (n = 8)
Table 5 summarizes the CT perfusion measurements for the three groups of patients. No difference in any of the blood flow parameters were found between the three diagnostic groups. However, hepatic arterial and portal perfusion tended to be higher while hepatic perfusion index (HPI) tended to be lower in patients with hepatic- and extra- hepatic metastases.
Unfiltered Texture Analysis
Non-Contrast CT & Contrast-Enhanced CT:
Table 6 summarizes the unfiltered quantification parameters for the three groups of patients for unenhanced, arterial and portal phase hepatic CT.
Mean grey-level intensity (absolute liver attenuation) values were not significantly different for the three groups of patients in unenhanced, arterial or portal CT images, although these values tended to be higher in patients with hepatic metastases in all three phases (Figure 5A).
In the arterial phase, hepatic entropy was significantly lower (p = 0.034, Figure 5B) hepatic uniformity was significantly higher (p = 0.034, Figure 5C) in apparently disease-free liver from patients with hepatic metastases as compared to those with no metastases.
A similar trend was found for entropy and uniformity in the other two phases but not reaching statistical significance. No difference in any parameter was found between patients with extra-hepatic tumour and the other two groups in any phase of enhancement.
Filtered Texture Analysis
Non-Contrast CT:
The texture parameters for the filtered non-contrast enhanced CT images for all the three diagnostic groups are summarized in Table 7. There was no difference between groups A (no tumour) and B (extra-hepatic tumour only) for any texture parameter.
For coarse and medium texture images, there was a trend towards higher alues for mean grey-level intensity and entropy in group C (liver metastases) as compared to groups A and B, reaching statistical significance for coarse texture images (p < 0.05).
Greater separation of the patient groups was achieved by using ratios of texture values. In particular, fine to medium texture ratios were most significant in differentiating the different diagnostic groups (see Table 7).
For comparison between groups A and C, the most significant difference was obtained for the ratio of fine to medium texture using the texture parameter entropy (p = 0.0257, Figure 6A) whilst the difference in this ratio for mean grey-level intensity was less significant (p = 0.049, Figure 6B).
For groups B and C, the most significant difference was obtained using the fine to medium texture ratio for the texture parameter uniformity (p = 0.0143, Figure 6C). Entropy also discriminated these two groups, with the highest significance obtained using the fine to medium texture ratio (p = 0.03).
For fine to medium texture ratios, an entropy value of less than 1.26 identified patients with liver metastases (group C) from the rest of the patients (Sensitivity = 87.5%, Specificity = 71%, Odds ratio = 17, 95% Confidence Interval = 1.75 – 165 with p < 0.015).
Contrast-Enhanced CT:
The texture parameter for filtered portal-phase contrast enhanced CT images for all the three diagnostic groups are summarized in Table 8.
Liver texture was significantly different in patients with extra-hepatic metastases (Group B) compared to patients with no tumour (Group A) as indicated by higher intensity values on normalized coarse texture (2.0 / 2.5, p < 0.04) and normalized medium texture images (1.5 / 2.5, 0.04 < p < 0.05).
For patients with a normal liver appearance on conventional CT, a normalized coarse liver texture value of greater than 1.13 indicated an increased likelihood of extra-hepatic metastases than for patients with lower texture values with a sensitivity of 62.5% and specificity of 100% (p = 0.0035). Kaplan-Meier survival curves for patients with normal hepatic appearances on conventional CT separated by a normalized coarse texture were significantly different (p = 0.0074, Figure 7). Reduced survival was found for patients with liver texture values above 1.13.
3. Biological Correlates For Liver Texture
Furthermore, two related biological correlates for liver texture on portal phase CT in colorectal cancer patients without hepatic metastases were identified as hepatic blood flow and glucose metabolism. The hepatic phosphorylation fraction index (HPFI) of glucose which is derived from the ratio of Standardized Uptake Value (SUV) of glucose in the liver and Total Hepatic Perfusion (THP) – combination of arterial (HAP) and portal perfusion (HPP) was identified as the most possible biological correlate for normalized mean coarse texture (r = -0.59, p = 0.0062, see Figure 8). This texture parameter also correlated inversely with hepatic glucose utilization (SUV: r = -0.587, p = 0.007) and positively with hepatic blood flow (THP: r = 0.512, p = 0.021 and HPP: r = 0.451, p = 0.046). A statistically significant positive correlation was also observed for normalized coarse uniformity parameter (HPFI: r = 0.552, p = 0.012 and SUV: r = 0.468, p = 0.038).