Evaluation of deep learning based dose prediction in head and neck cancer patients using two different types of input contours

Purpose: This study evaluates deep learning (DL) based dose predictionmethods in head and neck cancer (HNC) patients using two types of inputcontours. Materials and methods: Seventy-five HNC patients undergoing two-stepvolumetric-modulated arc therapy were included. Dose prediction was per-formed using the AIVOT prototype (AiRato.Inc, Sendai, Japan), a commercialsoftware with an HD U-net-based dose distribution prediction system. Modelswere developed for the initial plan (46 Gy/23Fr) and boost plan (24 Gy/12Fr),trained with 65 cases and tested with 10 cases. The 8-channel model usedone target (PTV) and seven organs at risk (OARs),while the 10-channel modeladded two dummy contours (PTV ring and spinal cord PRV). Predicted anddeliverable doses, obtained through dose mimicking on another radiation treat-ment planning system, were evaluated using dose-volume indices for PTV andOARs. Results: For the initial plan, both models achieved approximately 2% predic-tion accuracy for the target dose and maintained accuracy within 3.2 Gy forOARs. The 10-channel model outperformed the 8-channel model for certaindose indices. For the boost plan, both models exhibited prediction accuraciesof approximately 2% for the target dose and 1 Gy for OARs. The 10-channelmodel showed significantly closer predictions to the ground truth for D50% andDmean. Deliverable plans based on prediction doses showed little significantdifference compared to the ground truth, especially for the boost plan. Conclusion: DL-based dose prediction using the AIVOT prototype software inHNC patients yielded promising results.While additional contours may enhanceprediction accuracy, their impact on dose mimicking is relatively small.