Enhanced CT and MRI Focal Bone Tumor Classification with Machine Learning-based Stratification: A Multicenter Retrospective Study.

Journal: Radiology
PMID:

Abstract

Background Standardized bone tumor reporting is crucial for consistent, risk-aligned patient management. Current systems are based on expert consensus and/or lack multicenter validation. Purpose To evaluate a machine learning-based approach for differentiating between benign and malignant focal bone lesions and to propose a Bone Tumor Imaging Reporting and Data System (BTI-RADS) 2.0 for further risk stratification. Materials and Methods This retrospective multicenter trial included patients with solitary bone tumors undergoing radiography or CT and MRI at 10 centers from November 2009 to March 2022. Patients were divided into training and test datasets. Predefined radioclinical features were extracted. The training dataset was considered for bootstrapped χ feature selection, and extreme gradient boosting (XGBoost) classifiers were optimized using nested cross-validation. Continuous classifier outputs were thresholded to stratify patients into seven malignancy risk classes (BTI-RADS 2.0), and malignancy rates were assessed for the test set. XGBoost and human expert performances were compared using the Wilcoxon signed-rank significance test with a significance level of .05. Results In total, 1113 patients (mean age, 39 years ± 22 [SD]; 623 men) were included: 298 in the training and 815 in the test datasets. Twenty-seven of 80 (34%) multimodal features were selected based on χ analysis. Best classification performances were achieved by an XGBoost model trained on 27 features, with an F1 score of 0.81 (95% CI: 0.78, 0.84). This model performed slightly inferior to 28 experienced radiologists, who demonstrated an F1 score of 0.83 (95% CI: 0.80, 0.85; < .001). BTI-RADS 2.0 risk grades II-V were associated with malignancy rates of 0% (0 of 102; 95% CI: 0, 0), 8.3% (14 of 168; 95% CI: 4, 13), 45% (121 of 271; 95% CI: 39, 50), and 92% (252 of 274; 95% CI: 89, 95), respectively, identifying malignant lesions with a sensitivity of 96% (373 of 387; 95% CI: 94, 98). Conclusion A machine learning algorithm and risk stratification system achieved accurate and standardized bone tumor malignancy grading. Clinical trial registration no. NCT04884048 © RSNA, 2025 See also the editorial by Tordjman and Murphey in this issue.

Authors

  • Astrée Lemore
    CHRU de Nancy Pôle Imagerie, Service d'imagerie Guilloz, Nancy, Lorraine, France.
  • Nora Vogt
    IADI U1254, Inserm, Université de Lorraine, 54511 Nancy, France.
  • Julien Oster
  • Edouard Germain
    Guilloz Imaging Department, University of Lorraine, Central Hospital, University Hospital Center of Nancy, Nancy, France.
  • Marc Fauvel
    CIC-IT, INSERM 1433, Université de Lorraine, CHRU Nancy, Nancy, France.
  • Romain Gillet
    Guilloz Imaging Department, University of Lorraine, Central Hospital, University Hospital Center of Nancy, Nancy, France.
  • François Sirveaux
    Service de Chirurgie Traumatologique et Orthopédique, Centre Chirurgical Emile Gallé, Orthopedics Nancy, Lorraine, France.
  • Béatrice Marie
    Department of Pathology, Nancy University Hospital Nancy, Lorraine, France.
  • Nicolas Sans
    CHU Purpan, Service central d'imagerie médicale Place Baylac Toulouse, Haute-Garonne, France.
  • Marie Faruch
    CHU Purpan, Service central d'imagerie médicale Place Baylac Toulouse, Haute-Garonne, France.
  • Franck Lapègue
    CHU Purpan, Service central d'imagerie médicale Place Baylac Toulouse, Haute-Garonne, France.
  • François Lafourcade
    CHU Purpan, Service central d'imagerie médicale Place Baylac Toulouse, Haute-Garonne, France.
  • Sammy Badr
    Univ. Lille, CHU Lille, Marrow Adiposity and Bone Laboratory (MABlab) ULR 4490, Department of Rheumatology, Lille, France.
  • Anne Cotten
    Department of Musculoskeletal Imaging, Lille University Hospital, 59000 Lille, France; Lille University School of Medicine, 59000 Lille, France; Collège des Enseignants en Radiologie de France (CERF), 75013 Paris, France; Société Française de Radiologie, 75013 Paris, France.
  • Fadila Mihoubi Bouvier
    Hôpital Cochin, Service de radiologie ostéo-articulaire, Paris, Île-de-France, France.
  • Sisi Yang
    Department of Radiology, Hôpital Cochin, APHP, 75014 Paris, France; Université Paris Cité, 75006 Paris, France.
  • Jean-Luc Drapé
    Service de Radiologie B, Groupe Hospitalier Cochin, AP-HP Centre, Université de Paris, Paris, France.
  • Maxime Pastor
    IMAGINE UR UM 103, Montpellier University, Department of Medical Imaging, Nîmes University Hospital, 30029 Nîmes, France.
  • Yann Thouvenin
    Osteoarticular Medical Imaging Section, Department of Medical Imaging Montpellier, University Hospital Centre of Montpellier, Herault, France.
  • Marie Pierre Baron
    Osteoarticular Medical Imaging Section, Department of Medical Imaging Montpellier, University Hospital Centre of Montpellier, Herault, France.
  • Catherine Cyteval
    Osteoarticular Medical Imaging Section, Department of Medical Imaging Montpellier, University Hospital Centre of Montpellier, Herault, France.
  • David Fadli
    Department of Radiology, Institut Bergonie, 33000, Bordeaux, France.
  • Claire Fournier
    Department of Musculoskeletal Imaging, University Hospital Centre Bordeaux Pellegrin Hospital Group, Bordeaux, Aquitaine, France.
  • Olivier Hauger
    Department of Musculoskeletal Imaging, University Hospital Centre Bordeaux Pellegrin Hospital Group, Bordeaux, Aquitaine, France.
  • Mariem Ben Haj Amor
    Centre Oscar Lambret Departement d'Imagerie Médicale, Lille, Hauts-de-France, France.
  • Nicolas Stacoffe
    Department of Radiology, Groupement Hospitalier Sud, Hospice Civils de Lyon, Pierre Bénite, Rhône, France.
  • Sophie Daubié
    Hospices Civils de Lyon, Service d'Imagerie Médicale, Centre Hospitalier Lyon Sud, 69310, Pierre Bénite Cedex, France.
  • Jean-Baptiste Noel
    Department of Interventional Radiology, Centre Léon Bérard, Lyon, Rhône-Alpes, France.
  • Jean-Baptiste Pialat
    INSA-Lyon, CREATIS UMR5220, Université Claude Bernard Lyon 1, Villeurbanne, France.
  • Stéphane Cherix
    Department of Orthopedic Surgery, Lausanne University Hospital, Lausanne, Switzerland.
  • Fabio Zanchi
    Department of Radiology, CHUV, Lausanne University Hospital, and University of Lausanne, Lausanne, Switzerland.
  • Patrick Omoumi
    Department of Radiology, Lausanne University Hospital and University of Lausanne, Switzerland, Lausanne, Switzerland.
  • Alain Blum
    Guilloz Imaging Department, University of Lorraine, Central Hospital, University Hospital Center of Nancy, Nancy, France.
  • Gabriela Hossu
    INSERM U1254, IADI, Université de Lorraine, 54511 Vandoeuvre-les-Nancy, France; CIC-IT, CHRU Nancy, Université de Lorraine, 54000 Nancy, France.
  • Pedro A Gondim Teixeira
    CHRU de Nancy Pôle Imagerie, Service d'imagerie Guilloz, 29 avenue du Maréchal de Lattre de Tassigny, Nancy, Lorraine, France 54035; IADI, INSERM U1254, Université de Lorraine, Nancy, France.