Related Experiment Video
Updated: Jul 26, 2025
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
6.9K
Preoperatively Predicting Ki67 Expression in Pituitary Adenomas Using Deep Segmentation Network and Radiomics
Hongxia Li1, Zhiling Liu2, Fuyan Li3
1Department of Radiology, The Second Hospital of Shandong University, Jinan 250033, China (H.L.).
Academic Radiology
|June 17, 2023
Summary
Radiomics and deep learning accurately predict the Ki67 proliferation index in pituitary adenomas (PAs) using MRI. This approach shows clinical value in assessing tumor aggressiveness and recurrence risk.
Area of Science:
- Radiology
- Oncology
- Artificial Intelligence
Background:
- The Ki67 proliferation index is crucial for pituitary adenoma (PA) behavior and recurrence prediction.
- Radiomics and deep learning are emerging tools for pituitary tumor analysis.
Purpose of the Study:
- To assess the feasibility of predicting the Ki67 proliferation index in PAs.
- To utilize deep segmentation networks and radiomics on multiparameter MRI for Ki67 prediction.
Main Methods:
- Trained a cfVB-Net autosegmentation model, evaluating its Dice Similarity Coefficient (DSC).
- Classified 1214 patients into high (HG) and low (LG) Ki67 expression groups.
- Developed classification models using radiomics features and a nomogram incorporating clinical factors, imaging features, and Radscores.
Main Results:
- The cfVB-Net model achieved good segmentation performance (DSC: 0.723-0.930).
- Optimal radiomic features were identified for differentiating HG from LG across T1WI, T2WI, and contrast-enhanced T1WI (CE T1WI).
- A combined CE T1WI and T1WI model showed high predictive accuracy (AUC: 0.927 training, 0.831 validation, 0.825 testing); age, Hardy' grade, and Radscores predicted high Ki67 expression.
Conclusions:
- Deep segmentation networks and radiomics analysis on multiparameter MRI are effective for predicting Ki67 expression in PAs.
- The combined approach demonstrates significant clinical application value for PA management.
