A real-time tumor position prediction based multi-dimensional respiratory motion compensation puncture method.

Area of Science:

• Medical Imaging and Interventions
• Computational Biology and Bioinformatics
• Respiratory Medicine

Background:

• Accurate localization of lung tumors during respiratory motion is critical for effective percutaneous biopsy.
• Existing methods face challenges in real-time tracking and compensation for dynamic tumor movement.

Purpose of the Study:

• To develop and validate a real-time tumor position prediction-based method for multi-dimensional respiratory motion compensation during needle puncture.
• To enhance the accuracy and robustness of percutaneous lung biopsy procedures.

Main Methods:

• A hybrid model combining prediction and correlation was developed.
• A bidirectional Long Short-Term Memory neural network with attention (Bi-LSTM-ATT) predicted respiratory signals.
• A backpropagation neural network correlated signals with tumor positions for trajectory decomposition and optimal puncture window determination.

Main Results:

• The Bi-LSTM-ATT model achieved high accuracy in predicting respiratory signals (RMSE: 0.0482 mm, MAE: 0.0414 mm, R²: 97.90%).
• The correlation model localized lung tumors with a target registration error within 0.74 mm.
• The proposed puncture method demonstrated an average error of 1.2 mm (range: 1.00–1.32 mm).

Conclusions:

• The validated method exhibits high accuracy and robustness for real-time lung tumor tracking and compensation.
• This approach shows significant promise for improving the precision of percutaneous biopsy in the lung.