Evaluation of the efficacy of automated machine learning enhanced planning system and a comparative analysis with manual planning system

• Published in: Journal of cancer research and therapeutics Vol. 21; no. 3; pp. 593 - 601
• Main Authors: Jadhav, Anand, Gupte, Ajinkya, Rasal, Sachin, Awate, Omkar, Dandekar, Prasad Raj
• Format: Journal Article
• Language: English
• Published: India Wolters Kluwer - Medknow 01.04.2025; Medknow Publications and Media Pvt. Ltd; Medknow Publications & Media Pvt. Ltd
• Edition: 2
• Subjects: Algorithms; Automation; Cancer; Conformity; Humans; Machine Learning; Male; Oncology, Experimental; Organs at Risk - radiation effects; Original Article; Patient outcomes; Planning; Prostate cancer; Prostatic Neoplasms - pathology; Prostatic Neoplasms - radiotherapy; Radiation therapy; Radiotherapy; Radiotherapy Dosage; Radiotherapy Planning, Computer-Assisted - methods; Radiotherapy, Intensity-Modulated - methods; Retrospective Studies; Technology application; India; ethos; intelligent optimization engine; Automated Plan generation; prostate cancer; eclipse
• ISSN: 0973-1482; 1998-4138; 1998-4138
• DOI: 10.4103/jcrt.jcrt_1373_24

ABSTRACT Introduction: The aim of radiotherapy treatment is to deliver a high dose of radiation precisely to the target volume while minimizing exposure to the surrounding organs at risk. This approach maximizes the likelihood of tumor control and reduces the risk of adverse side effects. Treatment planning systems (TPS) are crucial in achieving this goal. However, the manual planning process is time-consuming, resource-intensive, and subject to variability based on the skill and experience of individual planners. Automated planning aims to reduce inter-plan variation and planning duration while maintaining or improving plan quality. Varian Medical Systems introduced the Ethos platform, an automated planning and delivery system utilizing an Intelligent Optimization Engine (IOE). This study evaluates the efficacy of automated plan generation using the Varian Ethos IOE for prostate cancer treatment, compared with plans generated using the Eclipse TPS with the anisotropic analytical algorithm (AAA). Materials and Methods: Fifteen retrospective patients diagnosed with prostate cancer, treated with a dose of 60 Gy in 20 fractions to the prostate, were included. Treatment approved Eclipse plans were recalculated and reoptimized with the same objective function, and then exported to the Ethos TPS. The Ethos TPS generates a total of five plans-7-, 9-, and 12-field IMRT plans, and 2- and 3-arc VMAT plans, respectively, maintaining fixed beam geometry. Two additional plans were also generated on Ethos: one maintaining identical parameters from Eclipse for calculation purposes, and a second involving re-optimization. The primary objective was to assess the number of prespecified dose constraints met, while the secondary objective was to compare dosimetric parameters, such as target coverage, dose conformity, dose homogeneity, and OAR sparing between the Ethos and Eclipse plans. Results: There was no statistically significant difference between the Eclipse plan and the Ethos-generated plans in meeting the prespecified criteria. For PTV coverage, mean values for V95 > 95% were achieved across all plans. The mean values for V105 < 5% were well below the threshold, indicating minimal hotspots. The conformity index (CI) was close to 1, and the homogeneity index (HI) was close to 0 across all plans, indicating good dose distribution and uniformity. OAR sparing for the urinary bladder, rectum, and penile bulb was within acceptable limits, meeting dose constraints in all plans. Monitor unit (MU) values were higher for Ethos plans compared to Eclipse but remained within clinically acceptable ranges. Conclusion: The Ethos TPS, using its IOE, demonstrated the capability to generate high-quality radiotherapy plans for prostate cancer that are comparable to those produced by the Eclipse TPS. This suggests that the automated planning system can effectively reduce planning time and resource consumption while maintaining plan quality, thus supporting its potential clinical implementation.