Validation of a Knowledge-Based Automated Planning System in Cervical Cancer As a Clinical Trial Quality System


      The aim of this project was to train and validate an automated knowledge-based planning (KBP) system for the phase 3 portion of INTERTECC, an ongoing international trial of IMRT in cervical cancer (NCT01554397).


      Eighty-four patients with stage IB-IVA cervical cancer, planned according to INTERTECC guidelines (prescription: 45–50.4 Gy), from two institutions were used. These plans are based on a single PTV and two primary OARs: pelvic bone marrow (PBM) and small bowel (SB). Rectum, bladder, and sigmoid were also included as avoidance OARs but were not prioritized over PTV, PBM, or SB. Initial knowledge-based DVH estimation models for these OARs were trained using all 84 plans. To obtain the final KBP models, we identified 30 high-quality original plans for PBM by comparing original DVHs for V10 and V20 to the initial model-estimated DVHs. Similarly, we selected another 30 plans (8 in common with PBM set) for SB model training by comparing V40 and V45. We developed the final INTERTECC automated planning module using fixed optimization objectives for the PTV and OAR objectives based on the final DVH estimation models. All plans were normalized with PTV V100 to 95%. After generating automated plans for all 84 patients, we compared the resultant PTV and OAR DVHs between automated and delivered plans. Using established NTCP models (PMIDs: 22516388, 20400238) we compared the probability of grade 2+ gastrointestinal (GI) toxicity and estimated white blood cell count (WBC) nadirs for delivered versus KBP plans. Paired t-tests were used to test differences in DVH metrics and NTCP.


      The mean (IQR) time to generate a fixed-field IMRT plan with KBP was 6.85 (2.58) minutes. In the 84-patient sample, KBP plans outperformed the delivered plans according to specified plan quality metrics (Table 1). The mean NTCP for GI toxicity (based on SB V45) was lower for KBP versus delivered plans (49.2% vs 52.4%, P < .001). The estimated mean WBC nadir (based on V20) was higher (2.74 vs 2.56, P < .001) with KBP versus delivered plans, indicating lowered probability of hematologic toxicity with KBP.
      Oral Scientific Abstracts 88; Table 1.Comparison of IMRT Plans
      PTV (Gy)D1%49.3 2.149.5 2.2.01
      D99%44.6 2.144.4 2.8.21
      PBM (%)V10 Gy84.2 4.585.6 4.7<.001
      V20 Gy66.4 5.269.5 4.9<.001
      SB (cc)V40 Gy272 127295 140<.001
      V45 Gy159 94.2178 99.9<.001
      Rectum (%)V30 Gy90.3 7.089.9 12.5.69
      V45 Gy47.5 20.448.8 20.7.25
      Dmax(Gy)47.6 2.048.7 2.4<.01
      Bladder (%)V45 Gy46.3 16.151.2 17.1<.01
      Dmax(Gy)47.9 2.248.9 2.4<.01
      Sigmoid (Gy)Dmax49.1 2.448.4 2.3<.01


      With both dramatic efficiency gains and improved normal tissue sparing, the final KBP automated planning module was validated as both a clinical trial quality control system and a valuable tool for high-quality clinical planning in cervical cancer.


      Commenting Guidelines

      To submit a comment for a journal article, please use the space above and note the following:

      • We will review submitted comments as soon as possible, striving for within two business days.
      • This forum is intended for constructive dialogue. Comments that are commercial or promotional in nature, pertain to specific medical cases, are not relevant to the article for which they have been submitted, or are otherwise inappropriate will not be posted.
      • We require that commenters identify themselves with names and affiliations.
      • Comments must be in compliance with our Terms & Conditions.
      • Comments are not peer-reviewed.