Deep Learning for Classification of Bone Lesions on Routine MRI

Abstract

Background Radiologists have difficulty distinguishing benign from malignant bone lesions because these lesions may have similar imaging appearances. The purpose of this study was to develop a deep learning algorithm that can differentiate benign and malignant bone lesions using routine magnetic resonance imaging (MRI) and patient demographics. Methods 1,060 histologically confirmed bone lesions with T1- and T2-weighted pre-operative MRI were retrospectively identified and included, with lesions from 4 institutions used for model development and internal validation, and data from a fifth institution used for external validation. Image-based models were generated using the EfficientNet-B0 architecture and a logistic regression model was trained using patient age, sex, and lesion location. A voting ensemble was created as the final model. The performance of the model was compared to classification performance by radiology experts. Findings The cohort had a mean age of 30±23 years and was 58.3% male, with 582 benign lesions and 478 malignant. Compared to a contrived expert committee result, the ensemble deep learning model achieved (ensemble vs. experts): similar accuracy (0·76 vs. 0·73, p=0·7), sensitivity (0·79 vs. 0·81, p=1·0) and specificity (0·75 vs. 0·66, p=0·48), with a ROC AUC of 0·82. On external testing, the model achieved ROC AUC of 0·79. Interpretation Deep learning can be used to distinguish benign and malignant bone lesions on par with experts. These findings could aid in the development of computer-aided diagnostic tools to reduce unnecessary referrals to specialized centers from community clinics and limit unnecessary biopsies. Funding This work was funded by a Radiological Society of North America Research Medical Student Grant (#RMS2013) and supported by the Amazon Web Services Diagnostic Development Initiative.

Malignant Bone Tumors Diagnosis Using Magnetic Resonance Imaging Based on Deep Learning Algorithms

“…In orthopedics, there are a few studies which use deep learning algorithms for classifying bone lesions. Compared with the method proposed by Feyisope R. Eweje et al [ 1 ] which requires the segmentation of the images before training, the proposed method requires no prior segmentation due to the fact that the classifier used for T1 and T2 images is a pretrained ResNet50, which classifies the entire image as benignant or malign. Even more so, the clinical model is a neural network classifier, with an input layer of six neurons, one hidden layer with three neurons, and one neuron as an output ( Figure 5 ).…”

“…The ResNet50 [ 24 ] architecture of the proposed system can be found in Table 2 . Feyisope R. et al [ 1 ] proposed a similar methodology for classifying bone tumors in which they have two models for predicting the output and a logistic regression for the clinical model. In their study, the MRI scans were manually segmented by a radiologist and the resulting images were used for training.…”

“…In their study, the MRI scans were manually segmented by a radiologist and the resulting images were used for training. Also, for the clinical model, Feyisope R. et al [ 1 ] used as input the following information: sex, age, and lesion location. The image classifier used in their study is a pretrained EfficientNet [ 31 ].…”

“…By using pretrained CNNs, the model is converging fast during training, and it can train on a small dataset. While other studies have proposed a manual segmentation of the images [ 1 ], in our study the images are not manually segmented but are sent to the training pipeline as a whole in order to predict the malignancy based on the entire image. In addition, the proposed clinical model has as inputs the values predicted by the two residual CNNs models and the bone location of the tumor.…”

Translatability Analysis of National Institutes of Health–Funded Biomedical Research That Applies Artificial Intelligence

“…As early as the 1980s, 1 it was understood that AI tools could eventually play a major role as expert consultants to physicians by using insights from data that may not be deemed actionable by human interpretation. From convolutional neural networks for imaging-based solid organ cancer screening 2 , 3 , 4 to natural language processing (NLP) to estimate the probability of diagnoses with data from the electronic health record, 5 , 6 , 7 the breadth of AI-powered technologies affecting our understanding of human health and health care delivery processes has rapidly expanded in recent years. 8 …”