ABSTRACT
Objective:To evaluate the differential diagnostic efficacy of a predictive model of breast imaging reporting and data system (BI-RADS) classification combined with mammography radiomics classifier for various X-ray phenotype of breast lesions.Methods:A retrospective analysis was performed on 2 055 female patients who underwent mammography examination and were confirmed by pathology from May 2013 to August 2020 in Zhongda Hospital, Southeast University. Breast lesion was classified into mass or non-mass according to the fifth edition of BI-RADS. The mass was further divided into small mass (maximum diameter ≤ 2 cm) and large mass (maximum diameter>2 cm), the non-mass was further divided into asymmetric, calcification and structural distortions. By manually segmenting the region of interest of the lesion, the radiomics features were extracted and the model was constructed. Receiver operating characteristic curve and area under the curve (AUC) were used to assess the diagnostic efficacy of the BI-RADS classification, the radiomics model and the combined model for various phenotypes of breast lesions. Differences among the AUC were analyzed by the DeLong test.Results:The AUCs based on the BI-RADS classification, the radiomics model and the combined model were 0.924±0.006, 0.827±0.009 and 0.947±0.005 respectively. Compared with BI-RADS classification and the radiomics model, AUC of the combined model was the highest, and the differences were statistically significant ( Z=9.29, 14.94, P<0.001). For large mass, small mass and non-mass, combined model (AUC=0.958±0.007, 0.933±0.013, 0.939±0.008) showed the best performance when compared to the BI-RADS classification (AUC=0.937±0.010, 0.896±0.020, 0.916±0.011; Z=5.32, 3.90, 5.08, P<0.001) or the radiomics model (AUC=0.872±0.012, 0.851±0.021, 0.758±0.016; Z=7.86, 4.53, 12.13, P<0.001). The AUC of the combined model for benign and malignant asymmetric breast lesions (0.897±0.017) was higher than that of the BI-RADS classification (AUC=0.866±0.020, Z=4.27, P<0.001) and the radiomics model (AUC=0.633±0.029, Z=7.44, P<0.001); however, the AUC of the combined model for benign and malignant calcification and structural distortion of breast lesions (0.971±0.010, 0.811±0.057, respectively) was only higher than that of the radiomics model (AUC=0.827±0.021, 0.586±0.075, Z=7.40, 3.15, P<0.001), and there was no significant difference with the BI-RADS classification (AUC=0.959±0.012, 0.800±0.061, Z=1.87, 0.39, P>0.05). Conclusion:The combined model shows better differential diagnostic performance, which is valued in the clinical application.
ABSTRACT
BACKGROUND@#The detection of pulmonary nodules is a key step to achieving the early diagnosis and therapy of lung cancer. Deep learning based Artificial intelligence (AI) presents as the state of the art in the area of nodule detection, however, a validation with clinical data is necessary for further application. Therefore, the aim of this study is to evaluate the performance of AI in the detection of malignant and non-calcified nodules in chest CT.@*METHODS@#Two hundred chest computed tomography (CT) data were randomly selected from a self-built nodule database from Tianjin Medical University General Hospital. Both the pathology confirmed lung cancers and the nodules in the process of follow-up were included. All CTs were processed by AI and the results were compared with that of radiologists retrieved from the original medical reports. The ground truths were further determined by two experienced radiologists. The size and characteristics of the nodules were evaluated as well. The sensitivity and false positive rate were used to evaluate the effectiveness of AI and radiologists in detecting nodules. The McNemar test was used to determine whether there was a significant difference.@*RESULTS@#A total of 889 non-calcified nodules were determined by experts on chest CT, including 133 lung cancers. Of them, 442 nodules were less than 5 mm. The cancer detection rates of AI and radiologists are 100%. The sensitivity of AI on nodule detection was significantly higher than that of radiologists (99.1% vs 43%, P<0.001). The false-positive rate of AI was 4.9 per CT and decreased to 1.5 when nodules less than 5 mm were excluded.@*CONCLUSIONS@#AI achieves the detection of all malignancies and improve the sensitivity of pulmonary nodules detection beyond radiologists, with a low false positive rate after excluding small nodules.