Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model

Area of Science:

• Oncology
• Medical Imaging
• Machine Learning

Background:

• Machine learning (ML) integrates diverse features for enhanced classification and regression.
• Early detection of micrometastasis is crucial for accurate cancer staging, as standard imaging often misses these early signs.
• A site-specific rat model for breast cancer bone macrometastasis was utilized, with metastases developing exclusively in the right hind leg.

Purpose of the Study:

• To develop an ML algorithm for predicting breast cancer bone macrometastasis growth in a rat model.
• To enable detection of micrometastasis prior to standard imaging visibility.
• To identify key predictive features from early imaging data.

Main Methods:

• Extraction of features related to tissue vascularization (MRI) and glucose metabolism (PET/CT) from early imaging.
• Utilizing a model-averaged neural network (avNNet) to classify animals into metastatic or non-metastatic groups.
• Calculation of diagnostic parameters (accuracy, sensitivity, specificity, predictive values, likelihood ratios) and ROC curve development.

Main Results:

• The ML algorithm successfully predicted macrometastasis development.
• Identified critical features from MRI and PET/CT data indicative of future tumor growth.
• Demonstrated the potential for early detection of metastatic disease.

Conclusions:

• The developed ML algorithm facilitates early prediction of breast cancer bone metastases in a preclinical model.
• This protocol offers a flexible framework adaptable to various ML algorithms and features for diverse oncological applications.
• The approach can be extended to analyze other conditions like infections and inflammation.