Development and Clinical Validation of a Protocol-Agnostic Machine Learning Platform for Automated Treatment Planning in External-Beam Radiotherapy

Abstract

Purpose: To develop and validate a protocol-agnostic machine learning platform ("Predictive Planning") for knowledge-based planning (KBP) in external-beam radiotherapy.Materials and Methods: 5334 retrospective photon-beam treatment plans (1145 Head and Neck (H&N), 1623 Thoracic, 781 Abdominal, 1785 Pelvis) from two institutions were used to develop and internally test general-purpose models for organ-at-risk (OAR) dose-volume histogram prediction.Models were deployed in a commercial system (Plan AI, Sun Nuclear Corporation), and 72 retrospective clinical plans (CP) (18 H&N, 20 Thoracic, 17 Abdominal, 17 Pelvis) were re-planned using treatment planning system optimization objectives predicted by the models ("Predictive Plans"; PP), without objective value changes.Dose metrics were compared for 25, 11, 8, and 7 OARs for H&N, Thoracic, Abdominal, and Pelvis plans, respectively, and for PTVs. Results:In the plan comparison study, there were no OAR dose metrics for which PP were statistically significantly greater than CP.For H&N, PP mean dose (Gy) was significantly lower for Brain (5.8 vs. 7.3, p=0.0003),Brainstem (9.6 vs. 13.8,p<0.0001),Glottis (26.9 vs. 31.9,p=0.0017),OpticChiasm (7.3 vs. 12.2, p<0.0001),Parotid_L/R (19.0/20.8 vs. 22.3/25.7,p<0.0001/<0.0001)and SpinalCord (10.1 vs. 15.7,p<0.0001).For Abdominal, PP mean dose was significantly lower for Heart (2.2 vs. 3.4, p=0.0039),Kidney_R (6.2 vs. 8.6, p=0.002),SpinalCanal (4.8 vs. 6.7,p=0.0004), and Stomach (9.1 vs. 10.9, p=0.002).For Pelvis, PP mean dose was significantly lower for Bladder (30.7 vs. 33.7,p<0.0001),Femur_Head_L/R (13.8/13.9 vs. 17.0/16.5,p=0.0026/0.0046),PenileBulb (15.8 vs. 21.9,p=0.0039), and Rectum (28.1 vs. 34.2,p=0.0002).Mean PTV coverage was significantly higher for H&N and Thoracic, and equivalent for Abdominal and Pelvis. Conclusions:Predictive Planning represents a shift in KBP from protocol-specific models trained with small, uniform datasets to protocol-agnostic, disease-site-specific models trained with large, heterogenous datasets.Plans generated using optimization objectives predicted by the models had equivalent or superior dosimetric outcomes.

Ähnliche Arbeiten