Click here to

Session: Radiography and Fluoroscopy [Return to Session]

Multiparameter Dense Neural Network (DNN) to Estimate Patient Eye-Lens Dose During Neuro-Interventional Procedures

J Collins1*, C Guo2, A Podgorsak3, S Rudin4, D Bednarek5, (1) University at Buffalo, Buffalo, NY, (2) ,Chicago, IL, (3) University at Buffalo, Buffalo, NY, (4) University at Buffalo (SUNY) School of Medicine, Buffalo, NY, (5) University at Buffalo, Buffalo, NY


TU-A-TRACK 2-6 (Tuesday, 7/27/2021) 10:30 AM - 11:30 AM [Eastern Time (GMT-4)]

Purpose: Fluoroscopically-guided neuro-interventional procedures can result in high patient radiation dose to the eye lens with associated deterministic risks such as cataract formation. Deep learning (DL) models were investigated to determine if the patient lens dose for given exposure conditions could be estimated in real-time hence better controlled.

Methods: Monte-Carlo simulations were done using NCICT computational phantoms to create a dataset of lens dose values for the right and left eyes to train the DL models. The dataset was obtained for a range of geometric and exposure parameters used in neuro-interventional procedures including entrance-field size, kVp, filter type, gantry angulation, patient head size, and patient x, y, z head position relative to the beam isocenter. The dose for each combination of parameters was expressed as lens dose per entrance air kerma (mGy/Gy). Data was split into training, validation and testing sets. Stacked models and median algorithms were implemented to create more robust models. The parameters to be input into the DNN during a procedure are obtained from our dose-tracking system (DTS) which is interfaced to the imaging system. Model performance was evaluated using the mean-absolute-percentage error (MAPE).

Results: Using the median and stacked algorithms both led to a MAPE of less than 10% for prediction of the testing set. The lens dose prediction time for a single-exposure projection was 8 ms and 6 ms for the median and stacked algorithms, respectively.

Conclusion: A DNN method is able to accurately predict the patient eye lens dose based on the geometric and exposure parameters that would be used in a fluoroscopic neuro-interventional procedure. This work shows that using a DNN is a viable option to be implemented in the DTS to predict patient eye-lens dose in real time and eliminate the need to have a large database of pre-calculated factors.

Funding Support, Disclosures, and Conflict of Interest: Authors receive research support from Canon (Toshiba) Medical Systems. The dose tracking system (DTS) software is licensed to Canon Medical Systems by the Office of Science, Technology Transfer and Economic Outreach of the University at Buffalo.



    Dose, Fluoroscopy, Radiation Risk


    IM- Radiation Dose and Risk: General (Most Aspects)

    Contact Email