Julie Dufau1, Amine Bouhamama2, Benjamin Leporq3, Lison Malaureille4, Olivier Beuf3, François Gouin5, Franck Pilleul2, Perrine Marec-Berard6. 1. Institut d'hématologie et d'oncologie pédiatrique, 1, place Professeur Joseph-Renaut, 69008 Lyon, France. Electronic address: julyduf@hotmail.com. 2. Université de Lyon, CREATIS (CNRS UMR 5220, Inserm U1206, INSA-Lyon, UJM Saint-Étienne, UCB Lyon1), 69621 Villeurbanne, France; Centre de lutte contre le cancer Léon Bérard, département de radiologie, 28, rue Laennec, 69008 Lyon, France. 3. Université de Lyon, CREATIS (CNRS UMR 5220, Inserm U1206, INSA-Lyon, UJM Saint-Étienne, UCB Lyon1), 69621 Villeurbanne, France. 4. Centre de lutte contre le cancer Léon Bérard, département de radiologie, 28, rue Laennec, 69008 Lyon, France. 5. Centre de lutte contre le cancer Léon Bérard, département de chirurgie, 28, rue Laennec, 69008 Lyon, France; Université de Nantes, faculté de médecine, laboratoire de physiopathologie de la résorption osseuse et thérapie des tumeurs osseuses primitives, Inserm UI957, rue Gaston Veil, 44000 Nantes, France. 6. Institut d'hématologie et d'oncologie pédiatrique, 1, place Professeur Joseph-Renaut, 69008 Lyon, France.
Abstract
INTRODUCTION: Osteosarcoma is the most common malignant bone tumor before 25 years of age. Response to neoadjuvant chemotherapy determines continuation of treatment and is also a powerful prognostic factor. There are currently no reliable ways to evaluate it early. The aim is to develop a method to predict the chemotherapy response using radiomics from pre-treatment MRI. METHODS: Clinical characteristics and MRI of patients treated for local or metastatic osteosarcoma were collected retrospectively in the Rhône-Alpes region, from 2007 to 2016. On initial MRI exams, each tumor was segmented by expert radiologist and 87 radiomic features were extracted automatically. Univariate analysis was performed to assess each feature's association with histological response following neoadjuvante chemotherapy. To distinguish good histological responder from poor, we built predictive models based on support vector machines. Their classification performance was assessed with the area under operating characteristic curve receiver (AUROC) from test data. RESULTS: The analysis focused on the MRIs of 69 patients, 55.1% (38/69) of whom were good histological responders. The model obtained by support vector machines from initial MRI radiomic data had an AUROC of 0.98, a sensitivity of 100% (IC 95% [100%-100%]) and specificity of 86% (IC 95% [59.7%-111%]). DISCUSSION: Radiomic based on MRI data would predict the chemotherapy response before treatment initiation, in patients treated for osteosarcoma.
INTRODUCTION:Osteosarcoma is the most common malignant bone tumor before 25 years of age. Response to neoadjuvant chemotherapy determines continuation of treatment and is also a powerful prognostic factor. There are currently no reliable ways to evaluate it early. The aim is to develop a method to predict the chemotherapy response using radiomics from pre-treatment MRI. METHODS: Clinical characteristics and MRI of patients treated for local or metastatic osteosarcoma were collected retrospectively in the Rhône-Alpes region, from 2007 to 2016. On initial MRI exams, each tumor was segmented by expert radiologist and 87 radiomic features were extracted automatically. Univariate analysis was performed to assess each feature's association with histological response following neoadjuvante chemotherapy. To distinguish good histological responder from poor, we built predictive models based on support vector machines. Their classification performance was assessed with the area under operating characteristic curve receiver (AUROC) from test data. RESULTS: The analysis focused on the MRIs of 69 patients, 55.1% (38/69) of whom were good histological responders. The model obtained by support vector machines from initial MRI radiomic data had an AUROC of 0.98, a sensitivity of 100% (IC 95% [100%-100%]) and specificity of 86% (IC 95% [59.7%-111%]). DISCUSSION: Radiomic based on MRI data would predict the chemotherapy response before treatment initiation, in patients treated for osteosarcoma.