Radiomics to Predict Invasiveness of Part-Solid Adenocarcinoma of the Lung

Toward Generalizability in the Deployment of Artificial Intelligence in Radiology: Role of Computation Stress Testing to Overcome Underspecification

The clinical deployment of artificial intelligence (AI) applications in medical imaging is perhaps the greatest challenge facing radiology in the next decade. One of the main obstacles to the incorporation of automated AI-based decision-making tools in medicine is the failure of models to generalize when deployed across institutions with heterogeneous populations and imaging protocols. The most well-understood pitfall in developing these AI models is overfitting, which has, in part, been overcome by optimizing training protocols. However, overfitting is not the only obstacle to the success and generalizability of AI. Underspecification is also a serious impediment that requires conceptual understanding and correction. It is well known that a single AI pipeline, with prescribed training and testing sets, can produce several models with various levels of generalizability. Underspecification defines the inability of the pipeline to identify whether these models have embedded the structure of the underlying system by using a test set independent of, but distributed identically, to the training set. An underspecified pipeline is unable to assess the degree to which the models will be generalizable. Stress testing is a known tool in AI that can limit underspecification and, importantly, assure broad generalizability of AI models. However, the application of stress tests is new in radiologic applications. This report describes the concept of underspecification from a radiologist perspective, discusses stress testing as a specific strategy to overcome underspecification, and explains how stress tests could be designed in radiology-by modifying medical images or stratifying testing datasets. In the upcoming years, stress tests should become in radiology the standard that crash tests have become in the automotive industry. Keywords: Computer Applications-General, Informatics, Computer-aided Diagnosis © RSNA, 2021.

Value of radiomics model based on enhanced computed tomography in risk grade prediction of gastrointestinal stromal tumors

To explore the application of computed tomography (CT)-enhanced radiomics for the risk-grade prediction of gastrointestinal stromal tumors (GIST). GIST patients (n = 292) confirmed by surgery or endoscopic pathology during June 2013-2019 were reviewed and categorized into low-grade (very low to low risk) and high-grade (medium to high risk) groups. The tumor region of interest (ROI) was depicted layer by layer on each patient's enhanced CT venous phase images using the ITK-SNAP. The texture features were extracted using the Analysis Kit (AK) and then randomly divided into the training (n = 205) and test (n = 87) groups in a ratio of 7:3. After dimension reduction by the least absolute shrinkage and the selection operator algorithm (LASSO), a prediction model was constructed using the logistic regression method. The clinical data of the two groups were statistically analyzed, and the multivariate regression prediction model was constructed by using statistically significant features. The ROC curve was applied to evaluate the prediction performance of the proposed model. A radiomics-prediction model was constructed based on 10 characteristic parameters selected from 396 quantitative feature parameters extracted from the CT images. The proposed radiomics model exhibited effective risk-grade prediction of GIST. For the training group, the area under curve (AUC), sensitivity, specificity, and accuracy rate were 0.793 (95%CI: 0.733-0.854), 83.3%, 64.3%, and 72.7%, respectively; the corresponding values for the test group were 0.791 (95%CI: 0.696-0.886), 84.2%, 69.3%, and 75.9%, respectively. There were significant differences in age (t value: - 3.133, P = 0.008), maximum tumor diameter (Z value: - 12.163, P = 0.000) and tumor morphology (χ2 value:10.409, P = 0.001) between the two groups, which were used to establish a clinical prediction model. The area under the receiver operating characteristic curve of the clinical model was 0.718 (95%CI: 0.659-0.776). The proposed CT-enhanced radiomics model exhibited better accuracy and effective performance than the clinical model, which can be used for the assessment of risk grades of GIST.