Paul Giraud1, Philippe Giraud2, Eliot Nicolas2, Pierre Boisselier3, Marc Alfonsi4, Michel Rives5, Etienne Bardet6, Valentin Calugaru7, Georges Noel8, Enrique Chajon9, Pascal Pommier10, Magali Morelle11, Lionel Perrier10, Xavier Liem12, Anita Burgun1, Jean Emmanuel Bibault1,2. 1. INSERM UMR 1138 Lab, Centre de Recherche des Cordeliers, Medical Informatics Unit, Georges Pompidou European Hospital-20 rue Leblanc, 75015 Paris, France. 2. Radiation Oncology Department, Georges Pompidou European Hospital, 20 rue Leblanc, 75015 Paris, France. 3. Radiation Oncology Department, Montpellier Cancer Institute (ICM), 34090 Montpellier, France. 4. Radiation Oncology Department, Sainte Catherine Institute, 84918 Avignon, France. 5. Radiation Oncology Department, Claudius Regaud Institute, 31300 Toulouse, France. 6. Radiation Oncology Department, René Gauducheau Cancer Centre, 44800 Saint-Herblain, France. 7. Radiation Oncology Department, Curie Institut, 75005 Paris, France. 8. Institut de Cancerologie de Strasbourg, 17 rue Albert Calmette BP 23025, 67033 Strasbourg CEDEX, France. 9. Radiation Oncology Department, Eugène Marquis Cancer Centre, 35000 Rennes, France. 10. Radiation Oncology Department, Leon Bérard Cancer Centre, 69008 Lyon, France. 11. GATE L-SE UMR 5824, Lyon University, Léon Bérard Cancer Center, F-69008, Lyon, France. 12. Radiation Oncology Department, Oscar Lambret Cancer Centre, 59000 Lille, France.
Abstract
BACKGROUND: There is no evidence to support surgery or radiotherapy as the best treatment for resectable oropharyngeal cancers with a negative HPV status. Predictive algorithms may help to decide which strategy to choose, but they will only be accepted by caregivers and European authorities if they are interpretable. As a proof of concept, we developed a predictive and interpretable algorithm to predict locoregional relapse at 18 months for oropharyngeal cancers as a first step towards that goal. METHODS: The model was based on clinical and Pyradiomics features extracted from the dosimetric CT scan. Intraclass correlation was used to filter out features dependant on delineation. Correlated redundant features were also removed. An XGBoost model was cross-validated and optimised on the HN1 cohort (79 patients), and performances were assessed on the ART ORL cohort (45 patients). The Shapley Values were used to provide an overall and local explanation of the model. RESULTS: On the ART ORL cohort, the model trained on HN1 yielded a precision-or predictive positive value-of 0.92, a recall of 0.42, an area under the curve of the receiver operating characteristic of 0.68 and an accuracy of 0.64. The most contributory features were shape Voxel Volume, grey level size zone matrix Small Area Emphasis (glszmSAE), gldm Dependence Non Uniformity Normalized (gldmDNUN), Sex and Age. CONCLUSIONS: We developed an interpretable and generalizable model that could yield a good precision-positive predictive value-for relapse at 18 months on a different test cohort.
BACKGROUND: There is no evidence to support surgery or radiotherapy as the best treatment for resectable oropharyngeal cancers with a negative HPV status. Predictive algorithms may help to decide which strategy to choose, but they will only be accepted by caregivers and European authorities if they are interpretable. As a proof of concept, we developed a predictive and interpretable algorithm to predict locoregional relapse at 18 months for oropharyngeal cancers as a first step towards that goal. METHODS: The model was based on clinical and Pyradiomics features extracted from the dosimetric CT scan. Intraclass correlation was used to filter out features dependant on delineation. Correlated redundant features were also removed. An XGBoost model was cross-validated and optimised on the HN1 cohort (79 patients), and performances were assessed on the ART ORL cohort (45 patients). The Shapley Values were used to provide an overall and local explanation of the model. RESULTS: On the ART ORL cohort, the model trained on HN1 yielded a precision-or predictive positive value-of 0.92, a recall of 0.42, an area under the curve of the receiver operating characteristic of 0.68 and an accuracy of 0.64. The most contributory features were shape Voxel Volume, grey level size zone matrix Small Area Emphasis (glszmSAE), gldm Dependence Non Uniformity Normalized (gldmDNUN), Sex and Age. CONCLUSIONS: We developed an interpretable and generalizable model that could yield a good precision-positive predictive value-for relapse at 18 months on a different test cohort.
Entities:
Keywords:
XGBoost; head and neck; machine learning; oropharyngeal cancer; radiomics
Authors: Howard E Morgan; Kai Wang; Michael Dohopolski; Xiao Liang; Michael R Folkert; David J Sher; Jing Wang Journal: Quant Imaging Med Surg Date: 2021-12