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

Abstract

Background: To develop a deep learning model to classify primary bone tumors from preoperative radiographs and compare performance with radiologists. Methods: A total of 1356 patients (2899 images) with histologically confirmed primary bone tumors and preoperative radiographs were identified from five institutions' pathology databases. Manual cropping was performed by radiologists to label the lesions. Binary discriminatory capacity (benign versus not-benign and malignant versus not-malignant) and three-way classification (benign versus intermediate versus malignant) performance of our model were evaluated. The generalizability of our model was investigated on data from external test set. Final model performance was compared with interpretation from five radiologists of varying level of experience using the Permutations tests. Findings: For benign vs. not benign, model achieved area under curve (AUC) of 0894 and 0877 on cross-validation and external testing, respectively. For malignant vs. not malignant, model achieved AUC of 0907 and 0916 on cross-validation and external testing, respectively. For three-way classification, model achieved 721% accuracy vs. 746% and 721% for the two subspecialists on cross-validation (p = 003 and p = 052, respectively). On external testing, model achieved 734% accuracy vs. 693%, 734%, 731%, 679%, and 634% for the two subspecialists and three junior radiologists (p = 014, p = 089, p = 093, p = 002, p < 001 for radiologists 1À5, respectively). Interpretation: Deep learning can classify primary bone tumors using conventional radiographs in a multiinstitutional dataset with similar accuracy compared to subspecialists, and better performance than junior radiologists.

Multitask Deep Learning for Segmentation and Classification of Primary Bone Tumors on Radiographs

“…Fifth, the image resolution of the radiographs as input to the model was fixed at 800 3 1200 pixels because of constraints from pretraining, thus possibly reducing the original resolution or fidelity of the radiograph. Nevertheless, the resolution used by the model was higher than the input resolution used by previous DL models (12,14,28). Sixth, the data set did not include any radiographs of bone metastases, a common bone lesion.…”

Qualitative Histopathological Classification of Primary Bone Tumors Using Deep Learning: A Pilot Study

“…Bao et al (29) have incorporated various features from radiographic observations and demographic information to build a naïve Bayesian-based model for ranking and classifying a wide range of bone tumor diagnoses. Yu et al (30) have established a DL algorithm to classify bone tumors in terms of aggressiveness on plain radiographs, finding the model has the ROC curve AUC of 0.877 for binary classification (benign vs. non-benign) and the CKS of 0.560 for ternary classification on testing dataset. However, the radiological information is relatively limited for AI models to train and learn because only several radiographic images can be obtained from one patient diagnosed with the bone tumor.…”

The Lodwick classification for grading growth rate of lytic bone tumors: a decision tree approach

“…Furthermore, they seem well-suited for evaluation of rare diseases like bone tumors, since extensive knowledge regarding bone tumor diagnosis cannot be taken for granted among general radiologists. Initial results of modern AI applications in the field of bone tumor diagnosis are promising [20,21]. He and colleagues reported on a deep learning algorithm that is able to automatically classify bone tumors (benign versus malignant) as good as experienced readers [21].…”

“…Initial results of modern AI applications in the field of bone tumor diagnosis are promising [20,21]. He and colleagues reported on a deep learning algorithm that is able to automatically classify bone tumors (benign versus malignant) as good as experienced readers [21]. It needs to be highlighted that the naïve Bayes classifiers used by Lodwick and colleagues [13], Kahn and colleagues [14], and Do and colleagues [5] are easy probabilistic machine learning (a branch of AI [22]) algorithms that process predefined, manually extracted features.…”