Dataset Information

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 pre-operative 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 0•894 and 0•877 on cross-validation and external testing, respectively. For malignant vs. not malignant, model achieved AUC of 0•907 and 0•916 on cross-validation and external testing, respectively. For three-way classification, model achieved 72•1% accuracy vs. 74•6% and 72•1% for the two subspecialists on cross-validation (p = 0•03 and p = 0•52, respectively). On external testing, model achieved 73•4% accuracy vs. 69•3%, 73•4%, 73•1%, 67•9%, and 63•4% for the two subspecialists and three junior radiologists (p = 0•14, p = 0•89, p = 0•93, p = 0•02, p < 0•01 for radiologists 1-5, respectively).

Interpretation

Deep learning can classify primary bone tumors using conventional radiographs in a multi-institutional dataset with similar accuracy compared to subspecialists, and better performance than junior radiologists.

Funding

The project described was supported by RSNA Research & Education Foundation, through grant number RSCH2004 to Harrison X. Bai.

SUBMITTER: He Y

PROVIDER: S-EPMC7689511 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:BackgroundMost primary bone tumors are often found in the bone around the knee joint. However, the detection of primary bone tumors on radiographs can be challenging for the inexperienced or junior radiologist. This study aimed to develop a deep learning (DL) model for the detection of primary bone tumors around the knee joint on radiographs.MethodsFrom four tertiary referral centers, we recruited 687 patients diagnosed with bone tumors (including osteosarcoma, chondrosarcoma, giant cell tumor of bone, bone cyst, enchondroma, fibrous dysplasia, etc.; 417 males, 270 females; mean age 22.8±13.2 years) by postoperative pathology or clinical imaging/follow-up, and 1,988 participants with normal bone radiographs (1,152 males, 836 females; mean age 27.9±12.2 years). The dataset was split into a training set for model development, an internal independent and an external test set for model validation. The trained model located bone tumor lesions and then detected tumor patients. Receiver operating characteristic curves and Cohen's kappa coefficient were used for evaluating detection performance. We compared the model's detection performance with that of two junior radiologists in the internal test set using permutation tests.ResultsThe DL model correctly localized 94.5% and 92.9% bone tumors on radiographs in the internal and external test set, respectively. An accuracy of 0.964/0.920, and an area under the receiver operating characteristic curve (AUC) of 0.981/0.990 in DL detection of bone tumor patients were for the internal and external test set, respectively. Cohen's kappa coefficient of the model in the internal test set was significantly higher than that of the two junior radiologists with 4 and 3 years of experience in musculoskeletal radiology (Model vs. Reader A, 0.927 vs. 0.777, P<0.001; Model vs. Reader B, 0.927 vs. 0.841, P=0.033).ConclusionsThe DL model achieved good performance in detecting primary bone tumors around the knee joint. This model had better performance than those of junior radiologists, indicating the potential for the detection of bone tumors on radiographs.

Project description:BackgroundCoronavirus disease 2019 (COVID-19) is a pandemic disease. Fast and accurate diagnosis of COVID-19 from chest radiography may enable more efficient allocation of scarce medical resources and hence improved patient outcomes. Deep learning classification of chest radiographs may be a plausible step towards this. We hypothesize that bone suppression of chest radiographs may improve the performance of deep learning classification of COVID-19 phenomena in chest radiographs.MethodsTwo bone suppression methods (Gusarev et al. and Rajaraman et al.) were implemented. The Gusarev and Rajaraman methods were trained on 217 pairs of normal and bone-suppressed chest radiographs from the X-ray Bone Shadow Suppression dataset (https://www.kaggle.com/hmchuong/xray-bone-shadow-supression). Two classifier methods with different network architectures were implemented. Binary classifier models were trained on the public RICORD-1c and RSNA Pneumonia Challenge datasets. An external test dataset was created retrospectively from a set of 320 COVID-19 positive patients from Queen Elizabeth Hospital (Hong Kong, China) and a set of 518 non-COVID-19 patients from Pamela Youde Nethersole Eastern Hospital (Hong Kong, China), and used to evaluate the effect of bone suppression on classifier performance. Classification performance, quantified by sensitivity, specificity, negative predictive value (NPV), accuracy and area under the receiver operating curve (AUC), for non-suppressed radiographs was compared to that for bone suppressed radiographs. Some of the pre-trained models used in this study are published at (https://github.com/danielnflam).ResultsBone suppression of external test data was found to significantly (P<0.05) improve AUC for one classifier architecture [from 0.698 (non-suppressed) to 0.732 (Rajaraman-suppressed)]. For the other classifier architecture, suppression did not significantly (P>0.05) improve or worsen classifier performance.ConclusionsRajaraman suppression significantly improved classification performance in one classification architecture, and did not significantly worsen classifier performance in the other classifier architecture. This research could be extended to explore the impact of bone suppression on classification of different lung pathologies, and the effect of other image enhancement techniques on classifier performance.

Project description:BackgroundMorphological parameters of the lumbar spine are valuable in assessing lumbar spine diseases. However, manual measurement of lumbar morphological parameters is time-consuming. Deep learning has automatic quantitative and qualitative analysis capabilities. To develop a deep learning-based model for the automatic quantitative measurement of morphological parameters from anteroposterior digital radiographs of the lumbar spine and to evaluate its performance.MethodsThis study used 1,368 anteroposterior digital radiographs of the lumbar spine to train a deep learning model to measure the quantitative morphological indicators, including L1 to L5 vertebral body height (VBH) and L1-L2 to L4-L5 intervertebral disc height (IDH). The means of the manual measurements by three radiologists were used as the reference standard. The parameters predicted by the model were analyzed against the manual measurements using paired t-tests. Percentage of correct key points (PCK), intra-class correlation coefficient (ICC), Pearson correlation coefficient (r), mean absolute error (MAE), root mean square error (RMSE), and Bland-Altman plots were performed to assess the performance of the model.ResultsWithin the 3-mm distance threshold, the model had a PCK range of 99.77-99.46% for the L1 to L4 vertebrae and 77.37% for the L5 vertebrae. Except for VBH-L5 and IDH_L3-L4, IDH_L4-L5 (P<0.05), the estimated values of the model in the remaining parameters were not statistically significant compared with the reference standard (P>0.05). Except for VBH-L5 and IDH_L4-L5, the model showed good correlation and consistency with the reference standard (ICC =0.84-0.96, r=0.85-0.97, MAE =0.5-0.66, RMSE =0.66-0.95). The model outperformed other models (EfficientDet + Unet, EfficientDet + DarkPose, HRNet, and Unet) in predicting landmarks within a distance threshold of 1.5 to 5 mm.ConclusionsThe model developed in this study can automatically measure the morphological parameters of the L1 to L4 vertebrae from anteroposterior digital radiographs of the lumbar spine. Its performance is close to the level of radiologists.