Deep learning-based classification of primary bone tumors on radiographs: A preliminary study

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Primary bone tumors require accurate classification for effective treatment.
• Radiographs are a common initial imaging modality for bone lesions.
• Distinguishing between benign, intermediate, and malignant bone tumors is critical.

Purpose of the Study:

• To develop a deep learning (DL) model for classifying primary bone tumors using preoperative radiographs.
• To compare the diagnostic performance of the DL model against experienced radiologists.

Main Methods:

• A multi-institutional dataset of 1356 patients with primary bone tumors and radiographs was curated.
• A DL model was trained and validated for binary (benign/not-benign, malignant/not-malignant) and three-way (benign/intermediate/malignant) classification.
• Model performance was evaluated on an external test set and compared to five radiologists of varying experience levels.

Main Results:

• The DL model achieved an Area Under the Curve (AUC) of 0.894 (cross-validation) and 0.877 (external test) for benign vs. not-benign classification.
• For malignant vs. not-malignant, the model achieved AUCs of 0.907 (cross-validation) and 0.916 (external test).
• In three-way classification, the model's accuracy (73.4% on external test) was comparable to subspecialists and superior to junior radiologists.

Conclusions:

• Deep learning models can effectively classify primary bone tumors from conventional radiographs.
• The developed DL model demonstrates performance on par with subspecialist radiologists.
• This AI approach offers potential to assist in the diagnosis of primary bone tumors, especially outperforming junior clinicians.