A newly published study in Frontiers in Oncology has shown that a deep learning-based hybrid model has the potential to be a valuable tool for the operative and noninvasive prediction of mitotic index (MI) in patients with gastrointestinal stromal tumors (GIST).

Deep learning techniques allow the development of neural networks that simulate the analysis and learning process of the human brain. Such algorithms have been widely used in segmentation, automatic recognition, and classification of different cancers, the investigators noted.

Read more about GIST diagnosis

Continue Reading

“In this study, we trained a convolutional neural network (CNN) classifier based on an integration of 2-dimensional (2D) multimodal magnetic resonance (MR) images and 3-dimensional (3D) shape-based radiomics features to perform preoperative prediction of mitotic index in GIST,” the authors wrote.

The study included141 patients who were newly diagnosed with GIST. The cohort had also undergone magnetic resonance imaging (MR) assessment from January 2013 to May 2022. The inclusion criteria for patients were as follows: 1) GIST was verified by postoperative pathology after radical treatment in their hospital; 2) MI was obtained through postoperative pathological examination; 3) preoperative MR examination including T2-weighted imaging and diffusion-weighted imaging sequences was made available.

The results suggest that the CNN integrating 2D tumor signal intensity, 3D tumor shape, patient age, and tumor size exhibited good predictive power in both the development and test sets. Moreover, GIST mitotic index proved to be an essential indicator of metastasis and prognosis and is independent of the tumor size and location. Additionally, significant differences in the age and maximum diameter between the high MI group and the low MI group were observed. 

In conclusion, we developed a deep learning-based model that used radiomics and clinical features to reliably predict the MI status in GIST based on conventional, unenhanced MR images. Our model is expected to serve as a practical tool for the noninvasive characterization of GIST to support personalized treatment plans, the authors concluded.

Gastrointestinal stromal tumors are the most common mesenchymal tumors of the digestive tract wall. The advancements in MR, being conventional imaging techniques, are limited by personalized human eye observation, providing limited information on the internal heterogeneity of tumors. Radiomics involves the high-throughput extraction of image information (containing shape, grayscale, and texture) from medical images and can help to transform image information to assess tumor grading and staging quantitatively. 

The research team indicated a few limitations in their study, including the small sample size, single-center-based study, and the manual segmentation of the tumors instead of automatic segmentation.


Yang L, Du D, Zheng T, et al. Deep learning and radiomics to predict the mitotic index of gastrointestinal stromal tumors based on multiparametric MRI. Front Oncol. Published online November 23, 2022. doi:10.3389/fonc.2022.948557.