An upfront patient selection strategy based on personalized data-driven computed tomography generation for deep inspiration breath-hold in breast radiotherapy

Background: Currently there is no widely used upfront selection method to determine whether patients are suitable for deep inspiration breath-hold (DIBH) in left-sided breast radiotherapy. Purpose: To establish an upfront patient selection strategy to improve the decision-making efficiency of DIBH and avoid extra computed tomography (CT) exposure to patients. Methods: A total of 174 patients who underwent both free-breathing (FB) and DIBH scans were enrolled. A general principal component analysis model for DIBH-CT synthesis was trained and consists of principal component feature vectors extracted from paired FB-CT and DIBH-CT in training set. The coefficients of the vectors were optimized to minimize the difference between synthetic CT and breath-hold scout image of each patient in test set, leading to personalized DIBH-CT synthesis. An upfront patient selection strategy was established based on cardiac dose in synthetic DIBH-CT plan. The performance of DIBH-CT synthesis was analyzed in terms of geometric and dosimetric consistency between synthetic and scanned DIBH-CTs. The accuracy of the patient selection strategy was evaluated. Time assumption of the patient selection workflow was analyzed. Results: Synthetic DIBH-CTs had average Dice similarity coefficients of 0.84 for the heart and 0.91 for the lungs compared with scanned DIBH-CTs. Synthetic DIBH-CT plans revealed an average mean heart dose reduction of 1.46 Gy, which was not significantly different from 1.51 Gy in scanned DIBH-CT plans (p = 0.878). The patient selection strategy yielded the correct benefit results with accuracy of 86.7 %. The average time assumption for patient selection was 11.9 ± 3.6 min. Conclusions: The proposed patient selection strategy can accurately identify patients benefiting from DIBH and provides a more efficient workflow for DIBH.