Breast Cancer Induced Bone Osteolysis Prediction Using Temporal Variational Autoencoders

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Breast cancer commonly metastasizes to bone, leading to osteolytic lesions and skeletal-related events (SREs).
• Predicting the onset and progression of these bone lesions remains a significant clinical challenge.
• Current imaging techniques are limited in predicting future bone lesion development.

Purpose of the Study:

• To develop and evaluate a deep learning model for predicting bone osteolysis progression in murine breast cancer bone metastases.
• To improve the assessment of fracture risk and guide treatment strategies for breast cancer bone metastasis.

Main Methods:

• A temporal variational autoencoder (T-VAE) model was employed, combining variational autoencoders and long short-term memory networks.
• The model was trained on sequential micro-computed tomography (CT) images of murine tibiae with breast cancer bone metastases.
• It learns bone-cancer interactions from sequential images to predict future CT scans and lesion development.

Main Results:

• The T-VAE model demonstrated superior accuracy in predicting bone lesion progression compared to other deep learning models.
• The model achieved higher performance across various evaluation metrics in predicting osteolytic lesion emergence.
• Accurate prediction and visualization of osteolytic bone lesion progression were achieved.

Conclusions:

• A novel deep learning framework accurately predicts and visualizes osteolytic bone lesion progression.
• This framework can assist in planning and evaluating treatment strategies to prevent SREs in breast cancer patients.
• The model offers a valuable tool for managing breast cancer bone metastasis and mitigating associated complications.