JMIR Publications

ABSTRACT

Background:

No studies have reported the use of deep learning radiomics to predict T staging in gastric cancer via integrating radiomics and deep learning.

Objective:

To develop a computed tomography (CT)-based model for the automatic prediction of the T stage of gastric cancer (GC) via radiomics and deep learning.

Methods:

A total of 771 GC patients from 3 centers were retrospectively enrolled and divided into training, validation, and testing cohorts. GC patients were classified into mild (stage T1 and 2), moderate (stage T3) and severe (stage T4) groups. Three predictive models based on the labelled CT images were constructed by using the radiomics features (Radiomics model), deep features (Deep learning model) and the combination of both (Hybrid model).

Results:

The overall classification accuracy of the radiomics model was 64.3% in the internal testing dataset. The deep learning model and hybrid model showed better performance than the radiomics model, with overall classification accuracies of 75.7% (p = 0.037) and 81.4% (p = 0.001), respectively. On the subtasks of binary classification of tumor severity, the AUCs of the radiomics model, deep learning model and hybrid model were 0.875, 0.866 and 0.886 in the internal testing dataset and 0.820, 0.818 and 0.972 in the external testing dataset for differentiating mild (stage T1~2) from nonmild (stage T3~4) patients, while yielding 0.815, 0.892 and 0.894 in the internal testing dataset and 0.685, 0.808 and 0.897 in the external testing dataset for differentiating nonsevere (stage T1~3) from severe (stage T4) patients, respectively.

Conclusions:

The hybrid model integrating radiomics features and deep features shows favorable performance in diagnosing the pathological stage of gastric cancer. Clinical Trial: None