1. Interpretable Machine Learning Model for Locoregional Relapse Prediction in Oropharyngeal Cancers
- Author
-
Marc Alfonsi, P. Boisselier, Enrique Chajon, Eliot Nicolas, Paul Giraud, Valentin Calugaru, Georges Noël, Michel Rives, X. Liem, Anita Burgun, Etienne Bardet, Philippe Giraud, Jean-Emmanuel Bibault, Magali Morelle, Pascal Pommier, Lionel Perrier, Institut Curie [Paris], Centre Léon Bérard [Lyon], Centre de Recherche des Cordeliers (CRC (UMR_S_1138 / U1138)), École Pratique des Hautes Études (EPHE), Université Paris sciences et lettres (PSL)-Université Paris sciences et lettres (PSL)-Institut National de la Santé et de la Recherche Médicale (INSERM)-Sorbonne Université (SU)-Université Paris Cité (UPCité), Hôpital Européen Georges Pompidou [APHP] (HEGP), Assistance publique - Hôpitaux de Paris (AP-HP) (AP-HP)-Hôpitaux Universitaires Paris Ouest - Hôpitaux Universitaires Île de France Ouest (HUPO), Institut de Recherche en Cancérologie de Montpellier (IRCM - U1194 Inserm - UM), CRLCC Val d'Aurelle - Paul Lamarque-Institut National de la Santé et de la Recherche Médicale (INSERM)-Université de Montpellier (UM), Institut Sainte Catherine [Avignon], Institut Claudius Regaud, CRLCC René Gauducheau, Laboratoire Traitement du Signal et de l'Image (LTSI), Université de Rennes (UR)-Institut National de la Santé et de la Recherche Médicale (INSERM), Centre Régional de Lutte contre le Cancer Oscar Lambret [Lille] (UNICANCER/Lille), Université de Lille-UNICANCER, Groupe d'analyse et de théorie économique (GATE Lyon Saint-Étienne), Centre National de la Recherche Scientifique (CNRS)-Université de Lyon-Université Jean Monnet [Saint-Étienne] (UJM)-Université Claude Bernard Lyon 1 (UCBL), Université de Lyon-Université Lumière - Lyon 2 (UL2)-École normale supérieure - Lyon (ENS Lyon), École pratique des hautes études (EPHE), Université Paris sciences et lettres (PSL)-Université Paris sciences et lettres (PSL)-Institut National de la Santé et de la Recherche Médicale (INSERM)-Sorbonne Université (SU)-Université de Paris (UP), Institut National de la Santé et de la Recherche Médicale (INSERM)-Université de Rennes 1 (UR1), Université de Rennes (UNIV-RENNES)-Université de Rennes (UNIV-RENNES), Université Lille Nord de France (COMUE)-UNICANCER, Groupe d'Analyse et de Théorie Economique Lyon - Saint-Etienne (GATE Lyon Saint-Étienne), and École normale supérieure de Lyon (ENS de Lyon)-Université Lumière - Lyon 2 (UL2)-Université Jean Monnet - Saint-Étienne (UJM)-Centre National de la Recherche Scientifique (CNRS)
- Subjects
Cancer Research ,medicine.medical_specialty ,Intraclass correlation ,oropharyngeal cancer ,medicine.medical_treatment ,Computed tomography ,[SDV.CAN]Life Sciences [q-bio]/Cancer ,lcsh:RC254-282 ,Article ,030218 nuclear medicine & medical imaging ,head and neck ,03 medical and health sciences ,0302 clinical medicine ,medicine ,Voxel volume ,medicine.diagnostic_test ,Receiver operating characteristic ,business.industry ,Predictive analytics ,lcsh:Neoplasms. Tumors. Oncology. Including cancer and carcinogens ,[SHS.ECO]Humanities and Social Sciences/Economics and Finance ,3. Good health ,Radiation therapy ,machine learning ,Oncology ,radiomics ,030220 oncology & carcinogenesis ,Cohort ,radiomics oropharyngeal cancer ,Radiology ,business ,Oropharyngeal Cancers ,XGBoost - 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&mdash, or predictive positive value&mdash, 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&mdash, positive predictive value&mdash, for relapse at 18 months on a different test cohort.
- Published
- 2020
- Full Text
- View/download PDF