Share：
Share this content in WeChat
X
[Chinese] [PDF] 4 1
• Clinical Article •
Explainable machine learning model based on DKI, IVIM, and clinical features for preoperative prediction of lymphovascular invasion in rectal cancer
LIU Qingxiang  WU Shujian  YUAN Quan  FAN Lifang  ZHAI Jian 

DOI:10.12015/issn.1674-8034.2025.11.019.


[Abstract] Objective This study aimed to evaluate an explainable machine learning model for predicting lymphovascular invasion (LVI) in rectal cancer. The model was built using diffusion kurtosis imaging (DKI), intravoxel incoherent motion diffusion-weighted imaging (IVIM-DWI), and clinical data.Materials and Methods This retrospective study included 91 patients with pathologically confirmed rectal cancer. Patients were stratified into LVI-positive (+) and LVI-negative (-) groups according to histopathological findings. All patients underwent magnetic resonance imaging (MRI) examinations, including diffusion kurtosis imaging (DKI) and intravoxel incoherent motion diffusion-weighted imaging (IVIM-DWI). Six quantitative parameters were extracted from the scans: mean kurtosis (MK), mean diffusivity (MD), true diffusion coefficient (D), pseudo-diffusion coefficient (D*), perfusion fraction (f), and apparent diffusion coefficient (ADC). In addition, clinical variables such as serum tumor marker levels and pathological lymph node status were recorded. Finally, univariate followed by multivariate logistic regression analyses were performed to identify independent predictors of LVI. A combined predictive model was developed using logistic regression (LR). Model performance was evaluated using receiver operating characteristic (ROC) curves and the area under the curve (AUC), with five-fold cross-validation for internal validation. Feature contributions were visualized using SHapley Additive exPlanations (SHAP), and an online risk calculator was developed for individualized prediction.Results Univariate and multivariate regression analyses identified MK, D, f, and carcinoembryonic antigen (CEA) as independent predictors of LVI in rectal cancer. The LR-based combined model achieved an AUC of 0.887, with a mean AUC of 0.893 in five-fold cross-validation. SHAP analysis clearly illustrated the contribution of each feature to the prediction. The web-based risk calculator enabled real-time visualization of individualized risk estimates.Conclusions The explainable machine learning model based on DKI and IVIM-DWI quantitative parameters combined with clinical features effectively predicts LVI status in rectal cancer. This model not only demonstrates excellent predictive performance but also enhances clinical applicability and generalizability through transparent feature contribution and individualized risk assessment.
[Keywords] diffusion kurtosis imaging；intravoxel incoherent motion；rectal cancer；lymphovascular invasion；machine learning；magnetic resonance imaging

LIU Qingxiang1   WU Shujian1   YUAN Quan1   FAN Lifang2   ZHAI Jian1*  

1 Department of Radiology, the First Affiliated Hospital of Wannan Medical College (Yijishan Hospital), Wuhu 241001, China

2 Department of Medical Imaging, Wannan Medical College, Wuhu 241002, China

Corresponding author: ZHAI J, E-mail: yjszhaij@126.com

Conflicts of interest   None.

Received  2025-07-07
Accepted  2025-11-10
DOI: 10.12015/issn.1674-8034.2025.11.019
DOI:10.12015/issn.1674-8034.2025.11.019.

[1]
SIEGEL R L, MILLER K D, GODING SAUER A, et al. Colorectal cancer statistics, 2020[J]. CA A Cancer J Clin, 2020, 70(3): 145-164. DOI: 10.3322/caac.21601.
[2]
WONG M C S, HUANG J J, LOK V, et al. Differences in incidence and mortality trends of colorectal cancer worldwide based on sex, age, and anatomic location[J/OL]. Clin Gastroenterol Hepatol, 2021, 19(5): 955-966.e61 [2025-05-30]. https://pubmed.ncbi.nlm.nih.gov/32088300/. DOI: 10.1016/j.cgh.2020.02.026.
[3]
GLYNNE-JONES R, WYRWICZ L, TIRET E, et al. Rectal cancer: ESMO Clinical Practice Guidelines for diagnosis, treatment and follow-up[J/OL]. Ann Oncol, 2017, 28(suppl_4): iv22-iv40 [2025-05-30]. https://pubmed.ncbi.nlm.nih.gov/28881920/. DOI: 10.1093/annonc/mdx224.
[4]
ZHOU M, BAO D Y, HUANG H Y, et al. Utilization of diffusion-weighted derived mathematical models to predict prognostic factors of resectable rectal cancer[J]. Abdom Radiol (NY), 2024, 49(9): 3282-3293. DOI: 10.1007/s00261-024-04239-2.
[5]
FILLI L, WURNIG M C, LUECHINGER R, et al. Whole-body intravoxel incoherent motion imaging[J]. Eur Radiol, 2015, 25(7): 2049-2058. DOI: 10.1007/s00330-014-3577-z.
[6]
YUAN J, GONG Z G, LIU K, et al. Correlation between diffusion kurtosis and intravoxel incoherent motion derived (IVIM) parameters and tumor tissue composition in rectal cancer: a pilot study[J]. Abdom Radiol (NY), 2022, 47(4): 1223-1231. DOI: 10.1007/s00261-022-03426-3.
[7]
YU J, DAI X, ZOU H H, et al. Diffusion kurtosis imaging in identifying the malignancy of lymph nodes during the primary staging of rectal cancer[J]. Colorectal Dis, 2018, 20(2): 116-125. DOI: 10.1111/codi.13835.
[8]
QIU L, LIU X L, LIU S R, et al. Role of quantitative intravoxel incoherent motion parameters in the preoperative diagnosis of nodal metastasis in patients with rectal carcinoma[J]. J Magn Reson Imaging, 2016, 44(4): 1031-1039. DOI: 10.1002/jmri.25250.
[9]
WONG C, LIU T, ZHANG C Y, et al. Preoperative detection of lymphovascular invasion in rectal cancer using intravoxel incoherent motion imaging based on radiomics[J]. Med Phys, 2024, 51(1): 179-191. DOI: 10.1002/mp.16821.
[10]
HUANG T, ZENG Y P, JIANG R J, et al. Interpretable MRI subregional radiomics-deep learning model for preoperative lymphovascular invasion prediction in rectal cancer: A dual-center study[J/OL]. J Imag Inform Med, 2025 [2025-05-30]. https://pubmed.ncbi.nlm.nih.gov/40646374/. DOI: 10.1007/s10278-025-01586-4.
[11]
SHI S M, SINGH A, MA J Q, et al. Development and validation of a multi-parametric MRI deep-learning model for preoperative lymphovascular invasion evaluation in rectal cancer[J]. Quant Imaging Med Surg, 2025, 15(1): 427-439. DOI: 10.21037/qims-24-789.
[12]
XU F R, HONG J W, WU X H. An integrative clinical and intra- and peritumoral MRI radiomics nomogram for the preoperative prediction of lymphovascular invasion in rectal cancer[J]. Acad Radiol, 2025, 32(7): 3989-4001. DOI: 10.1016/j.acra.2025.02.019.
[13]
ZHU L, PAN Z L, MA Q, et al. Diffusion kurtosis imaging study of rectal adenocarcinoma associated with histopathologic prognostic factors: preliminary findings[J]. Radiology, 2017, 284(1): 66-76. DOI: 10.1148/radiol.2016160094.
[14]
YIN H J, LIU W L, XUE Q, et al. The value of restriction spectrum imaging in predicting lymph node metastases in rectal cancer: a comparative study with diffusion-weighted imaging and diffusion kurtosis imaging[J/OL]. Insights Imaging, 2024, 15(1): 302 [2025-05-30]. https://pubmed.ncbi.nlm.nih.gov/39699826/. DOI: 10.1186/s13244-024-01852-z.
[15]
ZHANG L, JIN Z W, YANG F, et al. Added value of histogram analysis of intravoxel incoherent motion and diffusion kurtosis imaging for the evaluation of complete response to neoadjuvant therapy in locally advanced rectal cancer[J]. Eur Radiol, 2025, 35(3): 1669-1678. DOI: 10.1007/s00330-024-11081-z.
[16]
ZHOU M, CHEN M Y, LUO M F, et al. Pathological prognostic factors of rectal cancer based on diffusion-weighted imaging, intravoxel incoherent motion, and diffusion kurtosis imaging[J]. Eur Radiol, 2025, 35(2): 979-988. DOI: 10.1007/s00330-024-11025-7.
[17]
FUSCO R, SANSONE M, GRANATA V, et al. Diffusion and perfusion MR parameters to assess preoperative short-course radiotherapy response in locally advanced rectal cancer: a comparative explorative study among Standardized Index of Shape by DCE-MRI, intravoxel incoherent motion- and diffusion kurtosis imaging-derived parameters[J]. Abdom Radiol (NY), 2019, 44(11): 3683-3700. DOI: 10.1007/s00261-018-1801-z.
[18]
CUI Y F, YANG X T, DU X S, et al. Whole-tumour diffusion kurtosis MR imaging histogram analysis of rectal adenocarcinoma: Correlation with clinical pathologic prognostic factors[J]. Eur Radiol, 2018, 28(4): 1485-1494. DOI: 10.1007/s00330-017-5094-3.
[19]
YAMADA I, YOSHINO N, HIKISHIMA K, et al. Oral carcinoma: Clinical evaluation using diffusion kurtosis imaging and its correlation with histopathologic findings[J]. Magn Reson Imaging, 2018, 51: 69-78. DOI: 10.1016/j.mri.2018.04.014.
[20]
WANG C, YU J, LU M, et al. Diagnostic efficiency of diffusion sequences and a clinical nomogram for detecting lymph node metastases from rectal cancer[J]. Acad Radiol, 2022, 29(9): 1287-1295. DOI: 10.1016/j.acra.2021.10.009.
[21]
LU B L, YANG X Y, XIAO X J, et al. Intravoxel incoherent motion diffusion-weighted imaging of primary rectal carcinoma: correlation with histopathology[J]. Med Sci Monit, 2018, 24: 2429-2436. DOI: 10.12659/msm.908574.
[22]
HE R, SONG G S, FU J Y, et al. Histogram analysis based on intravoxel incoherent motion diffusion-weighted imaging for determining the perineural invasion status of rectal cancer[J]. Quant Imaging Med Surg, 2024, 14(8): 5358-5372. DOI: 10.21037/qims-23-1614.
[23]
TAHER H J, ARIAN A, AGHILI M, et al. Role of intravoxel incoherent motion and dynamic-contrast-enhanced perfusion in response prediction of patients with colorectal cancer after neoadjuvant chemotherapy[J]. Asian Pac J Cancer Prev, 2024, 25(9): 3029-3037. DOI: 10.31557/APJCP.2024.25.9.3029.
[24]
YANG A, LIN L B, XU H, et al. Combination of intravoxel incoherent motion histogram parameters and clinical characteristics for predicting response to neoadjuvant chemoradiation in patients with locally advanced rectal cancer[J]. Abdom Radiol (NY), 2025, 50(4): 1505-1515. DOI: 10.1007/s00261-024-04629-6.
[25]
YU X P, WEN L, HOU J, et al. Discrimination between metastatic and nonmetastatic mesorectal lymph nodes in rectal cancer using intravoxel incoherent motion diffusion-weighted magnetic resonance imaging[J]. Acad Radiol, 2016, 23(4): 479-485. DOI: 10.1016/j.acra.2015.12.013.
[26]
PAKU M, UEMURA M, KITAKAZE M, et al. Clinical significance of preoperative and postoperative serum CEA and carbohydrate antigen 19-9 levels in patients undergoing curative resection of locally recurrent rectal cancer[J]. Dis Colon Rectum, 2023, 66(3): 392-400. DOI: 10.1097/DCR.0000000000002655.
[27]
HAO C J, SUI Y B, LI J, et al. The clinical value of the combined detection of enhanced CT, MRI, CEA, and CA199 in the diagnosis of rectal cancer[J/OL]. J Oncol, 2021, 2021: 8585371 [2025-05-30]. https://pubmed.ncbi.nlm.nih.gov/34335762/. DOI: 10.1155/2021/8585371.
[28]
KAWADA M, TAKAOKA M, FUJIWARA H, et al. A case of CA19-9-producing esophagogastric junction cancer with ovarian juvenile granulosa cell tumor and rectal cancer[J]. Nihon Shokakibyo Gakkai Zasshi, 2022, 119(8): 750-760. DOI: 10.11405/nisshoshi.119.750.
[29]
SHAN J J, GU B X, SHI L M, et al. Prognostic value of CEA and CA19-9 in patients with local advanced rectal cancer receiving neoadjuvant chemoradiotherapy, radical surgery and postoperative chemotherapy[J]. Transl Cancer Res, 2021, 10(1): 88-98. DOI: 10.21037/tcr-20-2269.
[30]
SUN R Z, ZENG Y M, FAN Y Y, et al. A new post-operative prognostic System Combining CEA and CA199 for locally advanced rectal cancer patients undergoing neoadjuvant chemoradiotherapy followed by total mesorectal excision[J/OL]. Asian J Surg, 2023, 46(7): 2819 [2025-05-30]. https://pubmed.ncbi.nlm.nih.gov/36822944/. DOI: 10.1016/j.asjsur.2023.01.061.

PREV Predicting microvascular invasion in hepatocellular carcinoma using Delta radiomics model based on Gd-EOB-DTPA enhanced MRI
NEXT Application research of cellular microstructural parameters based on magnetic resonance in predicting lymph node metastasis and tumor deposit of rectal cancer
  


LINK


Tel & Fax: +8610-67113815    E-mail: editor@cjmri.cn