Related Experiment Video
Updated: Jan 15, 2026
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
7.4K
Predictive model of Ki67 expression level in osteosarcoma based on weakly supervised segmentation and multi-type
1School of Electronics and Information Engineering, Tiangong University, Tianjin 300387, China; Tianjin Key Laboratory of Optoelectronic Detection Technology and Systems, Tianjin 300387, China.
Computer Methods and Programs in Biomedicine
|October 9, 2025
Summary
This study developed an AI model to predict osteosarcoma Ki67 levels from pathology images. The XGBoost+SVM model achieved high accuracy and sensitivity, offering a cost-effective alternative to traditional methods.
Area of Science:
- Computational pathology
- Artificial intelligence in oncology
- Biomarker quantification
Background:
- Osteosarcoma is a malignant bone tumor affecting children and adolescents.
- Ki67 protein expression indicates tumor proliferation and is crucial for prognosis.
- Traditional Ki67 assessment relies on immunohistochemistry, which can be time-consuming and costly.
Purpose of the Study:
- To develop an efficient, low-cost artificial intelligence (AI) model for predicting Ki67 expression levels in osteosarcoma.
- To analyze pathological images for automated Ki67 assessment.
- To reduce reliance on conventional immunohistochemistry methods.
Main Methods:
- Analysis of 73 H&E-stained osteosarcoma whole slide images (WSIs).
- Weakly supervised learning for tumor region segmentation and Hover-Net for nuclear feature extraction (215 features).
- Feature selection using LASSO, MI, RFE, WRST, and XGBoost, followed by integration with 8 machine learning classifiers (AdaBoost, BalancedRF, KNN, LightGBM, MLP, QDA, RF, SVM).
Main Results:
- The optimal hybrid model, XGBoost+SVM, was identified by combining 5 key features with 8 classifiers.
- The XGBoost+SVM model achieved high performance: accuracy (0.767), recall (0.872), F1-score (0.800), and ROC-AUC (0.884).
- The model demonstrated both high accuracy and sensitivity in predicting Ki67 levels.
Conclusions:
- The developed AI model offers an automated and reliable solution for osteosarcoma Ki67 assessment.
- This approach reduces dependence on traditional immunohistochemistry.
- The model shows significant potential for clinical translation in osteosarcoma management.
