Texture Analysis in the Assessment of Rectal Cancer: Comparison of T2WI and Diffusion-Weighted Imaging

Texture analysis (TA) techniques derived from T2-weighted imaging (T2WI) and apparent diffusion coefficient (ADC) maps of rectal cancer can both achieve good diagnosis performance. This study was to compare TA from T2WI and ADC maps between different pathological T and N stages to confirm which TA analysis is better in diagnosis performance. 146 patients were enrolled in this study. Tumor TA was performed on every patient's T2WI and ADC maps, respectively; then, skewness, kurtosis, uniformity, entropy, energy, inertia, and correlation were calculated. Our results demonstrated that those significant different parameters derived from T2WI had better diagnostic performance than those from ADC maps in differentiating pT3b-4 and pN1-2 stage tumors. In particular, the energy derived from T2WI was an optimal parameter for diagnostic efficiency. High-resolution T2WI plays a key point in the local stage of rectal cancer; thus, TA derived from T2WI may be a more useful tool to aid radiologists and surgeons in selecting treatment.


Introduction
The incidence and mortality of colorectal cancer in China have maintained an upward trend, which ranks third and fifth among all malignant tumors. Among them, rectal cancer accounts for half of colorectal cancer in China. [1].
Pathological T and N stages of rectal cancer were important prognostic factors affecting patients [2]. The patients of rectal cancer with lymph node metastasis require neoadjuvant chemotherapy. pT1-2 stage patients do not require neoadjuvant chemotherapy due to a low recurrence rate; conversely, pT3-4 stage patients need neoadjuvant chemotherapy [3,4]. The prognosis of pT3 stage rectal cancer varies greatly based on extramural depth (EMD) of tumor invasion [5]. The 5 mm cutoff of EMD has been determined to be the most discriminating and simple to use in clinical practice regardless of differences in overall survival and local recurrence [5][6][7]. Therefore, an accurate preoperative evaluation is essential because individualized treatment measures are required for different local staging patients.
Diffusion-weighted imaging (DWI) could improve contrast between the lesion and normal tissue, which could improve the accuracy of rectal cancer detection. Compared with T2WI alone, T2WI combined with DWI has the best accuracy and specificity for the diagnosis of rectal cancer; moreover, that can improve accuracy of rectal cancer staging [8,9].
Texture analysis (TA), as a new image postprocessing technology, can quantitatively describe tissue heterogeneity in medical images. About rectal cancer, there are many TA studies involved on T2WI and ADC maps. The study of Yang et al. [10] reported that T2WI histogram parameters could differentiate the positive and negative lymph node. TA derived from T2WI can be used for prediction of the rectal cancer T stage. [11]. In addition, our previous research results showed that TA from ADC maps could predict the local stage of rectal cancer [12].
In the present study, we investigated to compare TA from T2WI and ADC maps between different pathological T and N stages and to evaluate diagnostic performance to further confirm which TA analysis is better in diagnosis performance.

Materials and Methods
2.1. Patients. This retrospective study was approved by the Institutional Review Board of Changshu Hospital of Soochow University. As this study has retrospective nature, requirements for written informed consent were waived. Inclusion criteria were (1) biopsy-proven nonmucinous adenocarcinoma and (2) the complete pathological T and N stage report. Exclusion criteria were (1) motion artifacts and magnetic sensitive artifacts affecting image quality and (2) treatment with preoperative chemotherapy or radiotherapy.
Finally, 146 patients were enrolled in the final study between October 2016 and October 2020.
2.2. MRI. All MRI examinations were performed on 3.0 T scanner (Intera Achieva TX, Philips Medical System) using a phased-array surface coil.
MR sequences included sagittal, oblique axial (perpendicular to the tumor axis), and oblique coronal (parallel to     (Figures 1 and 2). Then, texture features based on T2WI and ADC maps were calculated automatically. The seven texture features that we chose included skewness, kurtosis, and uniformity (first-order statistics) and entropy, energy, inertia, and correlation (secondorder statistics).

Histopathologic Analysis.
A pathological T (pT) and N (pN) staging report was performed based on the eighth edition of the American Joint Committee on Cancer (AJCC 8th) [13]. The histopathological report of local T and N staging was referred to as the gold standard in this study.  First, the Kolmogorov-Smirnov test was used; data were presented as mean ± standard deviation or median ± interquartile range. To assess the difference of texture parameters between pT1-3a vs. pT3b-4 stage and pN0 vs. pN1-2 stage from the T2WI and ADC maps, the t-test or Mann-Whitney U test was performed. Subsequent receiver operating characteristic (ROC) curve analysis was performed with those significant different parameters from the T2WI and ADC maps. The correlations between those significant different parameters from the T2WI and ADC maps and pathological T and N stages were used by the Spearman analysis. Intraclass correlation coefficients (ICCs) were calculated for interobserver agreements. SPSS and MedCalc software was used for statistical analysis. The significant level was set as p value ≤ 0.05.  Table 2. According to the design of this study, 70 patients were grouped as pT1-3a and 76 patients were grouped as pT3b-4. 91 patients were grouped as pN0, and 55 patients were grouped as pN1-2.

Texture Parameters in
Differentiating pT1-3a and pT3b-4 Stages. On T2WI, the pT3b-4 stage tumor demonstrated significantly higher kurtosis and entropy and lower energy than the pT1-3a stage tumor (all p < 0:05, Table 3). On the ADC map, the pT3b-4 stage tumor demonstrated significantly higher skewness, kurtosis, and entropy than the pT1-3a stage tumor (all p < 0:05). No significant differences were found in other TA parameters between different T stages. The results of ICCs are shown in Table 3.

Texture Parameters in Differentiating pN0 and pN1-2
Stages. On T2WI, the pN1-2 stage tumor demonstrated significantly higher skewness and entropy and lower energy than the pN0 stage tumor (all p < 0:05, Table 4). On the ADC map, the pN1-2 stage tumor demonstrated significantly higher skewness and kurtosis than the pN0 stage tumor (all p < 0:05). No significant differences were found in other TA parameters between different N stages. The results of ICCs are shown in Table 4.

Discussion
The present study showed that kurtosis, energy, and entropy from T2WI and skewness, kurtosis, and entropy from ADC maps could significantly differentiate pT1-3a and pT3b-4 stages. In addition, we found that skewness, energy, and entropy from T2WI and skewness and kurtosis from ADC maps could significantly differentiate pN0 and pN1-2 stages. Among the aforementioned significant different parameters, those parameters derived from T2WI had higher AUC differentiating pT3b-4 and pN1-2 stage tumors than those derived from ADC maps. Furthermore, those parameters derived from T2WI showed a better correlation with pathological T and N stages than those derived from ADC maps. The results indicate that the diagnostic performance of TA parameters derived from T2WI was better than that derived from ADC maps.
High-resolution T2WI plays a key point in the local stage of rectal cancer. In recent years, some studies applied TA derived from T2WI in order to improve the preoperative stage and predict prognostic factors of rectal cancer [10,11,14]. In our study, the pT3b-4 stage tumor demonstrated higher kurtosis and entropy and lower energy than the pT1-3a stage tumor. In addition, the pN1-2 stage tumor demonstrated higher skewness and entropy and lower energy than the pN0 stage tumor. Theoretically, higher entropy and lower energy reflect higher heterogeneity of the lesion [15]. This may interpret our results. Yang et al. [10] have reported that the lymph node-positive group had a significant lower energy and higher entropy than the lymph node-negative group.
DWI could improve contrast between the lesion and normal tissue. It has been widely used in preoperative imag-ing of rectal cancer because T2WI combined with DWI can improve the accuracy of rectal cancer staging [8]. Our previous research and other studies have shown that the ADC value estimated from DWI was related to tumor aggressiveness [16][17][18]. TA parameters derived from ADC maps were more sensitive and reliable markers for the accurate operative stage and evaluate the efficacy to treatment in rectal cancer [12,19]. In our study, the pT3b-4 stage tumor demonstrated higher skewness, kurtosis, and entropy than the pT1-3a stage tumor. In addition, the pN1-2 stage tumor demonstrated higher skewness and kurtosis than the pN0 stage tumor. Higher value of skewness and kurtosis reflects more complexity and heterogeneity in tumors. Our result concluded that skewness and kurtosis derived from ADC maps as imaging biomarkers might be differentiating the positive lymph node of rectal cancer. ROC curves showed that significant different parameters derived from T2WI had higher AUC differentiating pT3b-4  5 Computational and Mathematical Methods in Medicine and pN1-2 stage tumors than those derived from ADC maps. Energy was shown to be the best performance in predicting pT3b-4 and pN1-2 stage tumors among TA parameters with yielding AUC of 79.3% and 71.3%, respectively. Furthermore, the Spearman correlation analysis showed that energy derived from T2WI had the better correlation with pathological T and N stages than parameters derived from the ADC map. Therefore, our results indicated that the diagnostic performance of TA parameters derived from T2WI was better than that derived from ADC maps. The reason may be that high-resolution T2WI can reflect much more details in the image of the lesion. In addition, T2WI is the preferred sequence for the preoperative stage of rectal cancer. Thus, TA derived from T2WI may be a more useful tool to aid radiologists and surgeons in selecting treatment.
There are some limitations. First, rectal cancer often grows infiltrating along the bowel wall. Lesions and normal bowel walls are often unclear; then, the measurement cannot accurately delineate the entire tumor. So, the measurement in our study was performed on a single slice of the largest tumor slice. Second, we evaluated seven TA parameters in our study because these parameters have been proven by previous studies to be related to the heterogeneity of the disease. No other more parameters were included for the group study. Third, we only included pathological T and N stage as the grouping standard. More prognostic factors should be included in the later period for further research.
In conclusion, our study showed that the diagnostic performance of TA parameters derived from T2WI was better than that derived from ADC maps. In particular, the energy derived from T2WI was the optimal parameter for diagnostic efficiency. Thus, TA derived from T2WI may be a more useful tool to aid radiologists and surgeons in selecting treatment.

Data Availability
The data used to support the findings of this study are available from the corresponding author upon request.