Developing global image feature analysis models to predict cancer risk and prognosis.

Area of Science:

• Medical Imaging
• Machine Learning
• Oncology

Background:

• Personalized medicine requires accurate cancer risk and prognosis prediction.
• Conventional computer-aided detection (CAD) methods struggle with segmenting heterogeneous and fuzzy tumor regions.
• Existing CAD approaches often overlook global and background tissue characteristics.

Purpose of the Study:

• To investigate the feasibility of machine learning models trained on global image features for cancer risk and prognosis prediction.
• To develop novel computer-aided schemes for enhanced cancer detection and prediction.
• To evaluate the performance of global image features compared to traditional tumor-based features.

Main Methods:

• Developed and tested machine learning models using global image features.
• Utilized full-field digital mammography, magnetic resonance imaging, and computed tomography datasets for breast, lung, and ovarian cancers.
• Compared the performance of global feature models against conventional approaches.

Main Results:

• Machine learning models trained on global image features demonstrated higher predictive performance.
• Global image features provided complementary information to tumor-specific features for prognosis prediction.
• The proposed methods outperformed current clinical practice approaches.

Conclusions:

• Global image features can be effectively used alone or in combination with tumor-based features.
• These novel approaches hold potential for improving cancer risk and prognosis prediction.
• The findings support the development of advanced, case-based prediction models for precision medicine.