Multiparameter MRI-based model integrating radiomics and deep learning for preoperative staging of laryngeal squamous cell carcinoma

Abstract

The accurate preoperative staging of laryngeal squamous cell carcinoma (LSCC) provides valuable guidance for clinical decision-making. The objective of this study was to establish a multiparametric MRI model using radiomics and deep learning (DL) to preoperatively distinguish between Stages I-II and III-IV of LSCC. Data from 401 histologically confirmed LSCC patients were collected from two centers (training set: 213; internal test set: 91; external test set: 97). Radiomics features were extracted from the MRI images, and seven radiomics models based on single and combined sequences were developed via random forest (RF). A DL model was constructed via ResNet 18, where DL features were extracted from its final fully connected layer. These features were fused with crucial radiomics features to create a combined model. The performance of the models was assessed using the area under the receiver operating characteristic (ROC) curve (AUC) and compared with the radiologist performances. The predictive capability of the combined model for Progression-Free Survival (PFS) was evaluated via Kaplan-Meier survival analysis and the Harrell's Concordance Index (C-index). In the external test set, the combined model had an AUC of 0.877 (95% CI 0.807-0.946), outperforming the DL model (AUC: 0.811) and the optimal radiomics model (AUC: 0.835). The combined model significantly outperformed both the DL (p = 0.017) and the optimal radiomics models (p = 0.039), and the radiologists (both p < 0.050). Moreover, the combined model demonstrated great prognostic predictive value in patients with LSCC, achieving a C-index of 0.624 for PFS. This combined model enhances preoperative LSCC staging, aiding in making more informed clinical decisions.

Keywords: Cancer staging; Deep learning; Laryngeal squamous cell carcinoma; Multiparametric MRI; Radiomics.

Fig. 1

Flow chart of patient recruitment pathways.

Fig. 2

The study’s framework is structured as follows: (1) Multi-sequence MRI tumor segmentation; (2) Feature extraction and model Construction, where radiomics features were extracted from three-dimensional segmentation images to develop radiomics models. Deep learning networks were constructed using cropped two-dimensional regions of interest (ROIs), from which DL features were extracted. The crucial radiomics features were then fused with DL features to create a combined model; (3) Radiologist evaluation; and (4) Model performance assessment.

Fig. 3

The SHAP plot of the top 15 features in the combined model, with feature value indicating the importance of predictive features. SHAP value serve as a unified measure of the impact of a particular feature in the model. Red dots represent positive correlations, while blue dots represent negative correlations. DL, deep learning; glcm, gray-level co-occurrence matrix; gldm, gray level dependence matrix; glszm, gray-level size zone matrix.

Fig. 4

Performance comparison of three models and radiologists in the external test set. (a) Receiver Operating Characteristic (ROC) curves for the combined model, the Radiomics-ALL model, the Deep Learning (DL) model, and two radiologists. Numbers in parentheses indicate the respective areas under the ROC curves. (b) Calibration curve analysis for the combined model. The solid line demonstrates the performance of the combined model; the closer it is to the dashed diagonal line, the higher the consistency between predicted probabilities and observed probabilities.

Fig. 5

Grad-CAM Heatmaps for deep learning identification and focus in laryngeal squamous cell carcinoma. The red areas highlight the regions most critical to ResNet-18’s classification process, primarily located at tumor boundaries and within intratumoral regions. (a) Stage I–II; (b) Stage III–IV.

Fig. 6

Confusion matrices for three models and radiologists in the external test set. (a) Doctor 1, (b) Doctor 2, (c) The radiomics model, (d) The DL model, (e) The combined model.

Fig. 7

Kaplan–Meier survival curve of the combined model for predicting the progression-free survival in patients.