Converting dose-area product to effective dose in dental cone-beam computed tomography using organ-specific deep learning.

Abstract

Objective: To develop an accurate method for converting dose-area product (DAP) to patient dose for dental cone-beam computed tomography (CBCT) using deep learning.

Methods: 24,384 CBCT exposures of an adult phantom were simulated with PCXMC 2.0, using permutations of tube voltage, filtration, source-isocenter distance, beam width/height and isocenter position. Equivalent organ doses as well as DAP values were recorded. Next, using the aforementioned scan parameters as inputs, neural networks (NN) were trained using Keras for estimating the equivalent dose per DAP for each organ. Two methods were explored for positional input features: (1) 'Coordinate' mode, which uses the (continuous) XYZ-coordinates of the isocenter, and (2) 'AP/JAW' mode, which uses the (categorical) anteroposterior and craniocaudal position. Each network was trained, validated and tested using a 3/1/1 data split. Effective dose (ED) was calculated from the combination of NN outputs using ICRP 103 tissue weighting factors. The performance of the resulting NN models for estimating ED/DAP was compared with that of a multiple linear regression (MLR) model as well as direct conversion coefficients (CC).

Results: The mean absolute error (MAE) for organ dose/DAP on the test data ranged from 0.18% (bone surface) to 2.90% (oesophagus) in 'Coordinate' mode and from 2.74% (red bone-marrow) to 14.13% (brain) in 'AP/JAW' mode. The MAE for ED was 0.23% and 4.30%, respectively, for the two modes, vs. 5.70% for the MLR model and 20.19%-32.67% for the CCs.

Conclusion: NNs allow for an accurate estimation of patient dose based on DAP in dental CBCT.