A hybrid deep learning model for forecasting lymphocyte depletion during radiation therapy

Area of Science:

• Oncology
• Radiotherapy
• Artificial Intelligence

Background:

• Severe depletion of absolute lymphocyte count (ALC) during radiation therapy (RT) is linked to poorer survival in solid tumors.
• Predicting RT-induced lymphocyte depletion is crucial for optimizing treatment plans.

Purpose of the Study:

• To develop a predictive model for radiation-induced lymphocyte depletion in esophageal cancer patients.
• To utilize patient characteristics and dosimetric features for early prediction of ALC trends during RT.

Main Methods:

• A hybrid deep learning model with a stacked structure, incorporating Long Short-Term Memory (LSTM) and neural networks.
• The model processes four feature categories through separate channels, integrating outputs for weekly ALC prediction.
• A discriminative kernel was employed to extract temporal features and assign weights to input sequences.

Main Results:

• The proposed model demonstrated superior performance compared to existing prediction methods.
• A 30% reduction in Mean Squared Error (MSE) for weekly ALC predictions using pre-treatment data was achieved.
• An extended model using first-week treatment data further reduced MSE by 70% compared to pre-treatment data.

Conclusions:

• The developed model effectively predicts radiation-induced lymphocyte depletion for RT planning.
• Predicting ALC enables physicians to assess lymphopenia risk and identify patients for treatment modification.
• This facilitates personalized RT strategies to mitigate adverse outcomes associated with lymphocyte depletion.