Preoperative Prediction of KRAS Mutation in Rectal Cancer Using a Combined T2-Weighted Imaging Radiomics and Volumetric Apparent Diffusion Coefficient Histogram Model

doi:10.21203/rs.3.rs-8283261/v1

Download PDF

Research Article

Preoperative Prediction of KRAS Mutation in Rectal Cancer Using a Combined T2-Weighted Imaging Radiomics and Volumetric Apparent Diffusion Coefficient Histogram Model

https://doi.org/10.21203/rs.3.rs-8283261/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Purpose

To evaluate a combined model incorporating T2-weighted imaging (T2WI)-based radiomics signature and apparent diffusion coefficient (ADC) histogram features for predicting kirsten rat sarcoma virus oncogene (KRAS) mutation status in rectal cancer patients.

Methods

220 patients with pathologically confirmed rectal adenocarcinoma from Center I (training dataset: n = 154; internal validation dataset: n = 66) and 61 from Center II (external validation dataset) were retrospectively included. A total of 851 radiomic features from T2WI and 20 ADC histogram features from diffusion-weighted imaging (DWI) were extracted. These two sets of features underwent separate feature selection and were then combined to construct a classification model for KRAS prediction. Model performance was evaluated using ROC curve analysis, and AUCs were compared using the DeLong test. P < 0.05 was considered statistically significant.

Results

Four T2WI radiomics features and two ADC histogram features were selected to construct the combined model, which achieved the highest performance with an AUC of 0.823 [95% confidence interval (CI): 0.701–0.931] in the internal validation dataset, outperforming the radiomics-only (AUC = 0.751 [0.623–0.873]) and ADC-only models (AUC = 0.702 [0.571–0.819]). In the external validation dataset, it maintained superior performance (AUC = 0.759 [0.625–0.870]) and significantly outperformed the radiomics-only (AUC = 0.668 [0.514–0.803], P < 0.05) and ADC-only models (AUC = 0.464 [0.298–0.626], P < 0.05).

Conclusion

The combined model demonstrated robust performance for predicting KRAS mutation status in rectal cancer and holds promise as a noninvasive adjunct to genetic testing in clinical settings.

Rectal cancer

KRAS mutation

Radiomics analysis

ADC histogram

Noninvasive prediction

Colorectal cancer (CRC) is one of the most common malignant tumors of the gastrointestinal tract worldwide, characterized by high incidence and mortality, and showing a trend of younger age, posing a serious threat to human health [1, 2].

Kirsten rat sarcoma virus oncogene (KRAS) is one of the most frequently mutated genes in CRC, with a mutation rate ranging from approximately 35% to 50%. As a proto-oncogene, mutations in KRAS lead to the activation of downstream signaling pathways, such as the RAS-RAF-MAPK cascade, mediated by the epidermal growth factor receptor (EGFR) [3, 4]. This activation promotes tumor cell proliferation, invasion, and metastasis, thereby contributing to tumor progression and the development of therapeutic resistance [5–7]. Previous studies reported that patients with KRAS mutations were less responsive to anti-EGFR monoclonal antibody therapies, and their treatment outcomes were generally poorer than those of patients with wild-type KRAS [8–10]. Therefore, accurate preoperative prediction of KRAS mutation status is of great significance for developing individualized treatment plans and implementing precision therapy [11]. Currently, genetic testing of tissue specimens is considered the gold standard for evaluating KRAS status. However, this approach has several limitations, including its invasive nature, time-consuming process, high cost, and inability to capture the full heterogeneity of the tumor. Moreover, it is not suitable for patients who are unable to undergo invasive procedures [12].

Magnetic resonance imaging (MRI) is the primary imaging modality for preoperative evaluation of rectal cancer and can be imaged in multiple sequences [13, 14]. Diffusion-weighted imaging (DWI) with quantitative apparent diffusion coefficient (ADC) values has been confirmed to noninvasively reflect biological abnormalities in tumors. However, conventional ADC values provide limited information of the histopathological characteristics of tumors. Volume-based ADC histogram analysis can reflect the overall heterogeneity of tumors by quantifying the diffusion distribution and variations across all voxels, thereby eliminating sampling bias and yielding more accurate results. It can provide more information on the histopathological features of tumors and has demonstrated favorable performance in distinguishing various histological subtypes of tumors [15–18]. In recent years, the rapidly evolving field of radiomics has demonstrated promising potential in predicting KRAS mutation status in rectal cancer through high-throughput feature extraction and quantitative analysis of MRI images [19, 20]. However, to the best of our knowledge, few studies have explored the integration of ADC histogram analysis and radiomics analysis based on conventional MRI for assessing KRAS mutation status in rectal cancer. Therefore, the aim of this study is to investigate the efficiency of a predictive model combining ADC histogram features and T2-weighted imaging (T2WI) radiomics features for the noninvasive prediction of KRAS mutation status in patients with rectal cancer.

2.1 Study population

The clinical, pathological, and MRI information of 320 patients with confirmed rectal cancer between August 2020 and December 2024 in Center I were reviewed. The inclusion criteria were as follows: 1) Diagnosed with rectal cancer, and a preoperative MRI examination was performed at Center I; 2) No prior treatment before MRI and surgery; 3) Pathological diagnosis of rectal cancer, pathological staging, and KRAS gene testing result were obtained after surgery. The exclusion criteria were as follows: 1) Patients who received radiotherapy, chemotherapy, immunotherapy, or other therapies before surgery; 2) Inadequate bowel preparation or unclear lesion visibility on MRI images; 3) Images with significant artifacts; 4) DWI without standard b values (0 and 1000 s/mm²). Finally, 220 patients were included in this study. Additionally, A set of 61 patients from Center II between April 2022 and April 2024 was used for external validation of the prediction model, following the same inclusion and exclusion criteria (Fig. 1). This retrospective study was approved by the local ethics committees. Informed consent was waived for all patients.

MRI was performed using a 3.0 T scanner (Philips, Netherlands) with a dedicated phased-array body coil. The patients were positioned in the supine position and instructed to breathe quietly. The MRI sequences and parameters are detailed in Table 1. Dynamic contrast-enhanced T1- weighted imaging (DCE-T1WI) was performed after administration of 0.2 ml/kg of body weight contrast agent (gadodiamide, 0.5 mmol/mL, Hokuriku)

Table 1
MR imaging protocol and parameters in patients with rectal cancer
Sequences/Parameters	Axial T1WI	dS Zoom T2WI	Sagittal T2WI	Coronal T2WI	Axial DWI	Axial DCE	Sagittal DCE	Coronal DCE
Center I
TR/TE (ms)	476/8.0	4875/105	3000/95	3496/100	6139/51	12/1.32	3.0/1.07	3.2/1.15
FOV (mm2)	220×220	120×120	240×240	300×300	220×220	220×220	220×220	400×350
Matrix	276×219	220×154	336×243	332×286	80×77	124×110	124×110	268×236
Gap (mm)	1	0	0.4	0.4	0.3	-2	-2	-2
NEX	1	3	1	1	2	1	1	1
b value (s/mm)	-	-	-	-	0, 1000	-	-	-
Center II
TR/TE (ms)	480/8.0	5340/105	4909/102	5817/102	2721/73.8	5.2/1.7	5.2/1.7	4.1/1.7
FOV (mm2)	380×380	200×200	240×240	220×220	380×380	380×380	320×320	400×400
Matrix	276×256	320×288	320×288	320×288	128×128	320×256	272×256	272×256
Gap (mm)	1	0	0.5	0.5	0.3	-2	-2	-2
NEX	0.7	4	2	2	1	0.7	0.7	0.7
b value (s/mm)	-	-	-	-	0, 1000	-	-	-

T1WI = T1-weighted imaging; T2WI = T2-weighted imaging; DCE = dynamic contrast enhancement; DWI = diffusion weighted imaging; TR/TE = time of repetition / time of echo; FOV = Field of view; NEX = number of excitation; AT = acquisition time.

2.2 MRI analysis

2.3.1 ROI delineation

The ROI was manually delineated slice-by-slice along the tumor boundary on DWI images (b = 1000 s/mm²) and T2WI to generate the whole-tumor volume of interest (Fig. 2). T1WI and DCE-T1WI images were used as references to avoid areas of necrosis, cystic degeneration, and hemorrhage. Regions defined as necrotic or cystic were identified as having relatively low signal intensity on DWI (b = 1000 s/mm²), high signal intensity on T2WI, and no enhancement on DCE images. Regions with higher signal intensity than the tumor on T1WI and areas with no enhancement on DCE images were defined as hemorrhagic regions. The segmentation of the ROI was manually performed by a radiologist with 10 years of experience (Dr. A) and the results were reviewed by a radiologist with 20 years of experience (Dr. B). Both radiologists were blinded to the pathological and KRAS mutation results. Any discrepancies were resolved through discussion and consensus.

2.3.2 Feature extraction

All MRI images were normalized using Z-score normalization to reduce intensity variability across subjects. Radiomic feature extraction was conducted using the open-source package “PyRadiomics” (version 3.0.1; https://pyradiomics.readthedocs.io/en/latest/features.html). Totally 851 features including shape, first-order statistical, texture, and wavelet domain features were extracted in the tumor region in T2WI MRI images. Additionally, ADC histogram analysis was performed using Firevoxel software (version 314A; https://www.firevoxel.org/). Totally 20 features included tumor volume, minimum ADC (min ADC), maximum ADC (max ADC), mean ADC, ADC values at the 10th, 25th, 50th, 75th, and 90th percentiles, skewness, and kurtosis were extracted based on DWI sequence. Finally, features were standardized using Z-score normalization to ensure comparability prior to further analysis.

2.3.3 Feature Selection and Classification

For radiomic features, the least absolute shrinkage and selection operator (Lasso) algorithm was employed for feature selection. For ADC histogram features, those which showing significant differences between KRAS mutation status were retained. The selected features from both parts were then combined to form the final feature subset, and a support vector machine (SVM) classifier was used for model construction.

2.4 Statistical analysis

All statistical analyses were performed using Python (v3.11) with the SciPy and pandas libraries. The normality analysis of continuous data was performed by the Shapiro - Wilk test. Normal variables were expressed as mean ± standard deviation (SD) and compared between groups using the independent samples t-test. Non-normal variables expressed as median (interquartile range) and compared between groups using Mann-Whitney U test. Categorical variables were analyzed using either the chi-square test or Fisher’s exact test. The areas under the receiver operating characteristic curves (AUCs) were calculated to assess the efficiency of KRAS mutation predictive models, with the DeLong test used for the comparison between three models. P < 0.05 was considered statistically significant.

The internal dataset enrolled 220 patients including 127 patients with mutated KRAS (mean age, 65.22 years ± 12.79) and 93 patients with wild-type KRAS (mean age, 61.96 years ± 10.76). The external dataset enrolled 61 patients including 42 patients with mutated KRAS (mean age, 62.93 years ± 14.16) and 19 patients with wild-type KRAS (mean age, 60.21 years ± 11.96). Except for extramural vascular invasion (EMVI) (P = 0.018), no significant differences were observed in clinical and pathological characteristics including age, sex, CEA and CA199 levels, T stage, N stage, histological grade, and perineural invasion across the training, internal, and external validation datasets (P = 0.106–0.797). Additionally, none of the above clinicopathological characteristics was significantly different between the mutated and wild-type KRAS groups no matter which dataset (P = 0.055–1.000) (Table 2).

Table 2
Demographic and Clinical Characteristics of Patients with Rectal Cancer in the Internal Training, Internal Validation, and External Validation Cohorts
Mutation status	Training dataset	Internal validation	External validation	p-value
Mutation status	Mutated (n = 78) Wild-type (n = 74)	Mutated (n = 49) Wild-type (n = 19)	Mutated (n = 42) Wild-type (n = 19)	p-value
Age, mean ± SD, years	64.01 ± 11.34	64.21 ± 12.94	62.41 ± 13.44	0.640
Gender (%)				0.530
Male	103 (67.8%)	44 (67.8%)	45 (73.8%)
Female	49 (32.2%)	24 (35.3%)	16 (26.2%)
Tumor differentiation (%)				0.618
Moderate	109 (71.7%)	51 (75.0%)	41 (67.2%)
Poor	43 (28.3%)	17 (25.0%)	20 (32.8%)
CEA (%)				0.106
normal ≤ 5	94 (61.8%)	48 (70.6%)	32 (52.5%)
abnormal>5	58 (38.2%)	20 (29.4%)	29 (47.5%)
CA199 (%)				0.771
normal ≤ 20	122 (80.3%)	52 (76.5%)	47 (77.0%)
abnormal > 20	30 (19.7%)	16 (23.5%)	14 (23.0%)
pT stage (%)				0.510
	T1 9 (5.9%) T2 46 (30.3%) T3 97 (63.8%)	T1 6 (8.8%) T2 19 (27.9%) T3 43 (63.2%)	T1 0 (0%) T2 3 (4.9%) T3 58 (95.1%)
pN stage (%)				0.797
	N0 91 (59.9%) N1 52 (34.2%) N2 9 (5.9%)	N0 44 (64.7%) N1 18 (26.5%) N2 6 (8.8%)	N0 37 (60.7%) N1 20 (32.8%) N2 4 (6.6%)
pEMVI (%)				0.018^*
	86 (56.6%)	40 (58.8%)	47 (77.0%)
	66 (43.4%)	28 (41.2%)	14 (23.0%)
pNeural invasion (%)				0.751
Negative	88 (57.9%)	38 (55.9%)	38 (62.3%)
Positive	64 (42.1%)	30 (44.1%)	23 (37.7%)

CEA: Carcinoembryonic Antigen;CA199༚Carbohydrate Antigen 19 − 9༛EMVI༚Extramural Vascular Invasion.

After feature selection, a final feature subset comprising four radiomic features and two ADC histogram features was established for model construction. The selected radiomic features included original shape Surface Volume Ratio (SVR), original glcm Correlation (GLCM-Corr), wavelet-LHH gldm Small Dependence High Gray Level Emphasis (GLDM-SDHGLE), and wavelet-LLL glszm Zone Entropy (GLSZM-ZE). SVR quantifies the relationship between tumor surface area and volume, reflecting the compactness or irregularity of tumor shape. GLCM-Corr is a second-order texture feature derived from the gray-level co-occurrence matrix, measuring the linear dependency of gray-level intensities between neighboring voxels. GLDM-SDHGLE describes the emphasis of high gray-level values associated with small spatial dependencies after wavelet transformation, reflecting fine-scale intensity variations. GLSZM-ZE calculates the entropy of gray-level size zone distributions, which characterizes the randomness and complexity of homogeneous regions within the tumor. The two ADC histogram features were Skewness and Kurtosis. Skewness measures the asymmetry of the ADC value distribution; higher skewness indicates a right-shifted distribution with more voxels showing low diffusivity. Kurtosis quantifies the peakedness of the distribution, with lower values suggesting broader dispersion of ADC values and increased variation in tissue diffusion characteristics. The differences of each feature between the two genotypic groups are shown in Fig. 3.

The combined model (ADC histogram and radiomic features) for predicting KRAS mutation achieved an AUC of 0.823, an accuracy of 0.765, a sensitivity of 0.737, and a specificity of 0.776 in the internal test set. In the external test set, it yielded an AUC of 0.759, an accuracy of 0.645, a sensitivity of 0.850, and a specificity of 0.548. Overall, the combined model outperformed both the radiomics-only model and the ADC histogram-only model in both test datasets, with detailed results shown in Table 3.

Table 3
Performance Comparison of KRAS Mutation Classification
Model	Accuracy	Sensitivity	Specificity	AUC
Internal test set
Combined	0.765 [0.662, 0.853]	0.737 [0.500, 0.933]	0.776 [0.644, 0.889]	0.823 [0.701, 0.931]
Radiomics	0.750 [0.588, 0.809]	0.579 [0.429, 0.850]	0.816 [0.592, 0.837]	0.751 [0.623, 0.873]
ADC histogram	0.529 [0.412, 0.647]	0.947 [0.833, 1.000]	0.367 [0.235, 0.500]	0.702 [0.571, 0.819]
External test set
Combined	0.645 [0.516, 0.758]	0.850 [0.609, 0.960]	0.548 [0.419, 0.711]	0.759 [0.625, 0.870]
Radiomics	0.613 [0.452, 0.694]	0.500 [0.280, 0.708]	0.667 [0.475, 0.766]	0.668^* [0.514, 0.803]
ADC histogram	0.323 [0.323, 0.565]	0.750 [0.286, 0.714]	0.119 [0.278, 0.575]	0.464^* [0.298, 0.626]

KRAS: Kirsten Rat Sarcoma Viral Oncogene Homolog, ADC: apparent diffusion coefficient, AUC: area under the ROC curve. Statistical tests for AUC comparisons were based on Delong test, data with * superscript indicates statistically significant (P < 0.05). Values in square brackets represent the 95% confidence intervals, obtained via bootstrap resampling (1000 iterations).

The comparison of the ROC curves for each model is presented in Fig. 4. The DeLong test showed no statistically significant difference in AUC between the combined model and the radiomics-only model (0.823 vs. 0.751, P = 0.069) or the ADC histogram-only model (0.823 vs. 0.702, P = 0.103) in the internal test set. However, in the external test set, the combined model demonstrated significantly better performance than both comparison models (0.759 vs. 0.668, P = 0.022; 0.759 vs. 0.464, P = 0.003).

This study proposed a method for predicting KRAS mutation status by combining T2WI radiomics analysis and ADC histogram analysis. The proposed model demonstrated superior and robust performance in both internal and external test datasets, outperforming traditional models based solely on radiomics or histogram analysis. The model has the potential to provide a noninvasive alternative for assessing KRAS mutation status and help identify suitable candidates for targeted therapy in rectal cancer.

The T2WI sequence provides detailed information on tumor structure and morphology, while ADC maps reflect the diffusivity of water molecules within tissues, which can indirectly indicate cellular density and intratumoral heterogeneity [21]. These two imaging modalities offer complementary biological information, and their combination enhances the ability to discriminate KRAS mutation status. In our study, the integrated model outperformed either single-modality model in terms of AUC, especially in external validation. Meng et al. [22] reported an AUC of 0.651 for predicting KRAS mutations using a multiparametric MRI-based radiomics model in rectal cancer. Oh JE et al. [23] used a decision tree model based on three texture features and achieved an accuracy of 81.7% for identifying KRAS mutation. Another CT-based radiomics study reported an AUC of 0.869 for detecting KRAS/NRAS/BRAF mutations in colorectal cancer in the training cohort but lacked external validation [24]. Compared with above studies, our model achieved consistent performance across both internal and external validation cohorts, with notable improvement in the external test set, suggesting its potential value as a noninvasive tool for KRAS mutation prediction in rectal cancer.

In this study, four radiomic features were identified to be significantly associated with KRAS gene mutations in rectal cancer: SVR, GLCM-Correlation, GLDM-SDHGLE, and GLSZM-ZE. SVR is a shape feature that reflects the ratio of tumor surface area to volume; a decreased SVR value indicates a tumor shape that is closer to spherical. Previous studies have suggested that tumors with a larger axial-to-longitudinal dimension ratio are more likely to carry KRAS mutations [19], which was consistent with our findings. High-order texture features can quantify intratumoral heterogeneity that is not readily discernible to the naked eye. The texture feature GLCM-Correlation captures the linear dependency of gray-level values between neighboring voxels. In our study, this feature showed a significant difference in the KRAS mutation group. This finding aligned with the results of Mo et al. [25], who demonstrated that various GLCM-based features were strongly associated with KRAS mutations, reflecting increased disruption of internal tumor structure and enhanced tissue heterogeneity. The GLSZM-ZE measures the complexity of gray-level and structural distributions within the tumor. Higher ZE values indicate a more disordered distribution of gray-level zones, suggesting greater intratumoral heterogeneity. In our analysis, the KRAS mutation group exhibited significantly elevated ZE values, indicating more complex tissue architecture and increased microscopic heterogeneity. This result was consistent with findings reported by Shin et al. [26], who also observed elevated GLSZM-related features in tumors with KRAS mutations. Skewness in the ADC histogram reflects the asymmetry of the ADC value distribution. An increased skewness indicates a right-skewed distribution, which may suggest the presence of more diffusion-restricted regions within the tumor and higher intratumoral heterogeneity. Kurtosis reflects the peakedness or concentration of the ADC distribution; a higher kurtosis value implies greater cellular density and tighter tissue organization [27–29]. Jo et al. [30] reported that patients with KRAS-mutated rectal cancer exhibited significantly elevated ADC skewness, indicating more asymmetric diffusion characteristics. Our findings were consistent with above studies.

This study has several limitations. First, the sample size was relatively limited and data were collected from only two centers. Future research should involve larger, multicenter cohorts with varying MRI scanners and acquisition protocols to further validate and optimize the robustness and generalizability of the proposed model. Second, this study focused solely on predicting KRAS mutation status, without investigating other relevant genetic alterations, which warrants further exploration. Third, although external validation was conducted, the model still exhibited some performance degradation on the external dataset. This highlights the need for improved normalization strategies and the potential value of exploring deep learning-based classification algorithms in future studies.

In conclusion, the combined predictive model based on T2WI radiomics features and ADC histogram features showed good performance in predicting KRAS mutation status in patients with rectal cancer, and may be helpful for clinical assessment of KRAS status as a complementary approach to genetic testing.

Author Contribution

L.Q: Conceptualization, Methodology, Data collection, Data analysis, Writing original draft；Z.R: Data processing, Model construction, Statistical analysis；Y.Z: Data collection, Data analysis；Z.J: Conceptual design, Methodological supervision, Data extraction, Critical revision of the manuscript.All authors have read and approved the final version of the manuscript.

Keller DS, Berho M, Perez RO, Wexner SD, Chand M. The multidisciplinary management of rectal cancer. Nat Rev Gastroenterol Hepatol. 2020; 17(7):414–429. https://doi.org/10.1038/s41575-020-0275-y.
Sung H, Ferlay J, Siegel RL, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin. 2021; 71(3):209–249. https://doi.org/10.3322/caac.21660.
Cefalì M, Epistolio S, Palmarocchi MC, Frattini M, De Dosso S. Research progress on KRAS mutations in colorectal cancer. Journal of Cancer Metastasis and Treatment. 2021. https://doi.org/10.20517/2394-4722.2021.61.
Zhu G, Pei L, Xia H, Tang Q, Bi F. Role of oncogenic KRAS in the prognosis, diagnosis and treatment of colorectal cancer. Mol Cancer. 2021; 20(1):143. https://doi.org/10.1186/s12943-021-01441-4.
Asawa P, Bakalov V, Kancharla P, et al. The prognostic value of KRAS mutation in locally advanced rectal cancer. Int J Colorectal Dis. 2022; 37(5):1199–1207. https://doi.org/10.1007/s00384-022-04167-x.
Zheng S, You Z, Guo G, Lin Z, Wang S, Yang G. Effect of KRAS mutation status on clinicopathological characteristics and overall survival in patients with rectal cancer. BMC Gastroenterol. 2025; 25(1):37. https://doi.org/10.1186/s12876-025-03615-6.
Aljehani MA, Bien J, Lee JSH, Fisher GA, Jr., Lin AY. KRAS Sequence Variation as Prognostic Marker in Patients With Young- vs Late-Onset Colorectal Cancer. JAMA Netw Open. 2023; 6(11):e2345801. https://doi.org/10.1001/jamanetworkopen.2023.45801.
Peng J, Lv J, Peng J. KRAS mutation is predictive for poor prognosis in rectal cancer patients with neoadjuvant chemoradiotherapy: a systemic review and meta-analysis. Int J Colorectal Dis. 2021; 36(8):1781–1790. https://doi.org/10.1007/s00384-021-03911-z.
Park JS, Lee SS, Choi YS. The effects of KRAS Mutations on the Prognosis of Rectal Cancer Following Neoadjuvant Chemoradiotherapy and Surgery. Asian Pac J Cancer Prev. 2024; 25(7):2337–2342. https://doi.org/10.31557/APJCP.2024.25.7.2337.
Cremolini C, Antoniotti C, Lonardi S, et al. Activity and Safety of Cetuximab Plus Modified FOLFOXIRI Followed by Maintenance With Cetuximab or Bevacizumab for RAS and BRAF Wild-type Metastatic Colorectal Cancer: A Randomized Phase 2 Clinical Trial. JAMA Oncol. 2018; 4(4):529–536. https://doi.org/10.1001/jamaoncol.2017.5591.
Dai D, Wang Y, Zhu L, Jin H, Wang X. Prognostic value of KRAS mutation status in colorectal cancer patients: a population-based competing risk analysis. PeerJ. 2020; 8:e9149. https://doi.org/10.7717/peerj.9149.
Effendi-Ys R. Colonoscopy, Biomarkers, and Targeted Therapy in Colorectal Cancer. Acta Med Indones. 2022; 54(3):476–486.
Li Z, Qin Y, Liao X, et al. Comparison of clinical, radiomics, deep learning, and fusion models for predicting early recurrence in locally advanced rectal cancer based on multiparametric MRI: a multicenter study. Eur J Radiol. 2025; 189:112173. https://doi.org/10.1016/j.ejrad.2025.112173.
Ning X, Yang D, Ao W, et al. A novel MRI-based radiomics for preoperative prediction of lymphovascular invasion in rectal cancer. Abdom Radiol (NY). 2025; 50(8):3377–3390. https://doi.org/10.1007/s00261-025-04800-7.
Kyriazi S, Collins DJ, Messiou C, et al. Metastatic ovarian and primary peritoneal cancer: assessing chemotherapy response with diffusion-weighted MR imaging–value of histogram analysis of apparent diffusion coefficients. Radiology. 2011; 261(1):182–192. https://doi.org/10.1148/radiol.11110230.
Yin JD, Song LR, Lu HC, Zheng X. Prediction of different stages of rectal cancer: Texture analysis based on diffusion-weighted images and apparent diffusion coefficient maps. World J Gastroenterol. 2020; 26(17):2082–2096. https://doi.org/10.3748/wjg.v26.i17.2082.
Šarošković M, Vuković M, Stojanoski S, et al. Significance of apparent diffusion coefficient in diagnosis of rectal carcinoma. Front Oncol. 2024; 14:1464183. https://doi.org/10.3389/fonc.2024.1464183.
Ma X, Shen M, He Y, et al. The role of volumetric ADC histogram analysis in preoperatively evaluating the tumour subtype and grade of endometrial cancer. Eur J Radiol. 2021; 140:109745. https://doi.org/10.1016/j.ejrad.2021.109745.
Cui Y, Liu H, Ren J, et al. Development and validation of a MRI-based radiomics signature for prediction of KRAS mutation in rectal cancer. Eur Radiol. 2020; 30(4):1948–1958. https://doi.org/10.1007/s00330-019-06572-3.
Jia LL, Zhao JX, Zhao LP, Tian JH, Huang G. Current status and quality of radiomic studies for predicting KRAS mutations in colorectal cancer patients: A systematic review and meta–analysis. Eur J Radiol. 2023; 158:110640. https://doi.org/10.1016/j.ejrad.2022.110640.
Ma X, Ren X, Shen M, et al. Volumetric ADC histogram analysis for preoperative evaluation of LVSI status in stage I endometrioid adenocarcinoma. Eur Radiol. 2022; 32(1):460–469. https://doi.org/10.1007/s00330-021-07996-6.
Meng X, Xia W, Xie P, et al. Preoperative radiomic signature based on multiparametric magnetic resonance imaging for noninvasive evaluation of biological characteristics in rectal cancer. Eur Radiol. 2019; 29(6):3200–3209. https://doi.org/10.1007/s00330-018-5763-x.
Oh JE, Kim MJ, Lee J, et al. Magnetic Resonance-Based Texture Analysis Differentiating KRAS Mutation Status in Rectal Cancer. Cancer Res Treat. 2020; 52(1):51–59. https://doi.org/10.4143/crt.2019.050.
Yang L, Dong D, Fang M, et al. Can CT-based radiomics signature predict KRAS/NRAS/BRAF mutations in colorectal cancer? Eur Radiol. 2018; 28(5):2058–2067. https://doi.org/10.1007/s00330-017-5258-0.
Hu T, Gong J, Sun Y, et al. Magnetic resonance imaging-based radiomics analysis for prediction of treatment response to neoadjuvant chemoradiotherapy and clinical outcome in patients with locally advanced rectal cancer: A large multicentric and validated study. MedComm (2020). 2024; 5(7):e609. https://doi.org/10.1002/mco2.609.
Shin J, Seo N, Baek SE, et al. MRI Radiomics Model Predicts Pathologic Complete Response of Rectal Cancer Following Chemoradiotherapy. Radiology. 2022; 303(2):351–358. https://doi.org/10.1148/radiol.211986.
De Robertis R, Tomaiuolo L, Pasquazzo F, et al. Correlation between ADC Histogram-Derived Metrics and the Time to Metastases in Resectable Pancreatic Adenocarcinoma. Cancers (Basel). 2022; 14(24). https://doi.org/10.3390/cancers14246050.
Jiménez de Los Santos ME, Reyes-Pérez JA, Domínguez Osorio V, et al. Whole lesion histogram analysis of apparent diffusion coefficient predicts therapy response in locally advanced rectal cancer. World J Gastroenterol. 2022; 28(23):2609–2624. https://doi.org/10.3748/wjg.v28.i23.2609.
Peng Y, Tang H, Meng X, et al. Histological grades of rectal cancer: whole-volume histogram analysis of apparent diffusion coefficient based on reduced field-of-view diffusion-weighted imaging. Quant Imaging Med Surg. 2020; 10(1):243–256. https://doi.org/10.21037/qims.2019.12.32.
Jo SJ, Kim SH, Park SJ, Lee Y, Son JH. Association between Texture Analysis Parameters and Molecular Biologic KRAS Mutation in Non-Mucinous Rectal Cancer. Taehan Yongsang Uihakhoe Chi. 2021; 82(2):406–416. https://doi.org/10.3348/jksr.2022.0004.

No competing interests reported.

Download PDF

Editorial decision: Revision requested
06 Jan, 2026
Reviews received at journal
06 Jan, 2026
Reviewers agreed at journal
06 Jan, 2026
Reviewers agreed at journal
06 Jan, 2026
Reviewers agreed at journal
05 Jan, 2026
Reviewers agreed at journal
04 Jan, 2026
Reviewers agreed at journal
04 Jan, 2026
Reviews received at journal
21 Dec, 2025
Reviewers agreed at journal
11 Dec, 2025
Reviewers invited by journal
08 Dec, 2025
Editor assigned by journal
08 Dec, 2025
Submission checks completed at journal
08 Dec, 2025
First submitted to journal
04 Dec, 2025

You are reading this latest preprint version

Preoperative Prediction of KRAS Mutation in Rectal Cancer Using a Combined T2-Weighted Imaging Radiomics and Volumetric Apparent Diffusion Coefficient Histogram Model

Status:

Version 1

Abstract

Purpose

Methods

Results

Conclusion

Figures

1. Introduction

2. Material and Methods

2.1 Study population

2.2 MRI analysis

2.3.1 ROI delineation

2.3.2 Feature extraction

2.3.3 Feature Selection and Classification

2.4 Statistical analysis

3. Results

4. Discussion

Declarations

Author Contribution

References

Additional Declarations

Status:

Version 1