Prediction of breast dose in chest CT examinations using adaptive neuro-fuzzy inference system (ANFIS)

In chest computed tomography (CT), the breasts located within the scan range receive a substantial radiation dose. Due to the risk of breast-related carcinogenesis, analyzing the breast dose for justification of CT examinations seems necessary. The main goal of this study is to overcome the limitations of conventional dosimetry methods, such as thermoluminescent dosimeters (TLDs) by introducing the adaptive neuro-fuzzy inference system (ANFIS) approach. In this study, the breast dose of 50 adult female patients who underwent chest CT examinations was measured directly by TLDs. Then, the ANFIS model was developed with four inputs including dose length product (DLP), volumetric CT dose index (CTDIvol), total mAs, and size-specific dose estimate (SSDE), and one output (TLD dose). Additionally, multiple linear regression (MLR) as a traditional prediction model was used for linear modeling and its results were compared with the ANFIS. The TLD reader results showed that the breast dose value was 12.37 ± 2.46 mGy. Performance indices of the ANFIS model, including root mean square error (RMSE) and correlation coefficient (R), were calculated at 0.172 and 0.93 for the testing dataset, respectively. Also, the ANFIS model had superior performance in predicting the breast dose than the MLR model (R = 0.805). This study demonstrates that the proposed ANFIS model is efficient for patient dose prediction in CT scans. Therefore, intelligence models such as ANFIS are suggested to estimate and optimize patient dose in CT examinations.