Novel deep learning algorithm based MRI radiomics for predicting lymph node metastases in rectal cancer

Abstract

To explore the value of applying the MRI-based radiomic nomogram for predicting lymph node metastasis (LNM) in rectal cancer (RC). This retrospective analysis used data from 430 patients with RC from two medical centers. The patients were categorized into the LNM negative (LNM-) and LNM positive (LNM+) according to their surgical pathology results. We developed a physician model by selecting clinical independent predictors through physician assessments. Additionally, we developed deep learning radscore (DLRS) models by extracting deep features from multiparametric MRI (mpMRI) images. A nomogram model was constructed by combining the physician model and DLRS models. Among the patients, 192 (44.65%, 192/430) experienced LNM+. Six prediction models were developed, namely the physician model, three sequence models, the DLRS, and the nomogram. The physician model achieved AUC of the receiver operating characteristic (ROC) values of 0.78, 0.79, and 0.7, whereas the sequence models, DLRS model, and nomogram model achieved AUC values ranging from 0.83 to 0.99. The predictive performance of the DLRS and nomogram models was superior to that of the physician model. DLRS and nomogram models based on mpMRI provided higher accuracy in predicting LNM status in patients with RC than the other models.

Keywords: Deep learning; Lymph node metastasis; Magnetic resonance imaging; Rectal cancer.

Fig. 1

Enrollment of patients in this study.

Fig. 2

Flow chart of this study. First, the largest MR image of the RC lesion, obtained through the T2WI, DWI, and T1C sequence models, was selected. A rectangular region of interest (ROI) covering the lesion was chosen. Second, the ResNet-101 network was employed for feature extraction. A series of operations, including convolution, pooling, and residual connection, were performed to obtain the DL features of the fully connected layers. mRMR and Lasso regression were used to screen for highly collinear features. Third, the coefficients of each deep feature were calculated using logistic regression, completing the construction of the DL model. Subsequently, the physician model was integrated with the DLRS model to create a nomogram model. Forth, clinical application of models.

Fig. 3

Univariate and multivariate LNM analyses conducted by physicians. mrEMVI (OR = 2.21, p = 0.045) and mrN (OR = 2.83, p < 0.001) were identified as independent predictors that can be used to develop the physician model.

Fig. 4

The nomogram was constructed based on mrEMVI, mrN, and DLRS. Each variable corresponds to a specific point value, which is determined by projecting it onto the points axis. The total score is obtained by summing the individual point values of all variables. The linear predictor scale, ranging from − 14 to 10, is then used to convert the total score into a linear predictor value. Finally, by mapping this value onto the LNM probability axis, we obtain the corresponding probability of lymph node metastasis.

Fig. 5

The receiver operating characteristic (ROC) curve for different prediction models in training (a), internal (b) and external validation set (c).

Fig. 6

A DCA indicating that the net benefits of the nomogram and DLRS models are higher than those of the physician model.