ABSTRACT
Objective: To investigate the prediction value of prognosis of resectable gastric cancer patients based on texture features of preoperative enhanced CT images. Methods:: Data of 197 patients with gastric cancer confirmed by surgical pathology were retrospectively analyzed. The patients were randomly divided into training group (n=147) and validation group (n=50). A total of 90 3-dimensional quantitative features on portal venous phase images of preoperative enhanced CT were extracted of all patients, and intraclass correlation coefficient was used to select better repetitive features. LASSO COX regression analysis was used to reduce dimensionality and screen features related to patients' overall survival (OS). A image tag was built to classify patients in 2 groups. The patients were stratified into high-risk and low-risk groups according to the median of signature score, and the difference of OS was analyzed. A nomogram integrating image tag and pathological features was constructed after analyzing the relationship of clinical, pathological features or image texture labels and prognosis of gastric cancer patients, and the efficacy in predicting prognosis of gastric cancer patients was evaluated. Clinical decision curve was plotted to evaluate relative clinical value. Results: The image tag was established with 2 OS-related CT features. Statistical differences of OS were found between high-risk and low-risk patients in both training group (χ2=9.25) and validation group (χ2=8.49, both P<0.01). The image tag and TNM staging were independent risk factors of gastric cancer. For patients in training group and validation group, AUC of image tag predicting 3-year OS was 0.72 (P=0.02) and 0.67 (P=0.07), of nomogram integrated image tag and TNM staging was 0.78 and 0.81, respectively (both P<0.01). The decision curve analysis showed that the nomogram model had higher net benefit than image tag alone with the threshold probabilities of 0.13-0.59. Conclusion: Image labels based on texture features of enhanced CT image can be used for postoperative risk stratification of gastric cancer patients. Nomogram constructed with image tag combining pathological features can help to predict the prognosis of patient with resectable gastric cancer.
ABSTRACT
To develop and validate a fat-suppressed (T2 weighted-magnetic resonance imaging, T2W-MRI) based radiomics signature to preoperatively evaluate the histologic grade (grade I/II VS. grade III) of invasive breast cancer. Methods: A total of 202 patients with MRI examination and pathologically confirmed invasive breast cancer from June 2011 to February 2017 were retrospectively enrolled. After retrieving fat-suppressed T2W images and tumor segmentation, radiomics features were extracted and valuable features were selected to build a radiomic signature with the least absolute shrinkage and selection operator (LASSO) method. Mann-Whitney U test was used to explore the correlation between radiomics signature and histologic grade. Receiver operating characteristics (ROC) curve was applied to determine the discriminative performance of the radiomics signature [area under curre (AUC), sensitivity, specificity, and accuracy]. An independent validation dataset was used to confirm the discriminatory power of radiomics signature. Results: Eight radiomics features were selected to build a radiomics signature, which showed good performance for preoperatively evaluating histologic grade of invasive breast cancer, with an AUC of 0.802 (95% CI 0.729 to 0.875), sensitivity of 78.7%, specificity of 70.3% and accuracy of 73.7% in training dataset and AUC of 0.812 (95% CI 0.686 to 0.938), sensitivity of 80.0%, specificity of 73.3% and accuracy of 76.0% in the validation dataset. Conclusion: The fat-suppressed T2W-MRI based radiomics signature can be used to preoperatively evaluate the histologic grade of invasive breast cancer, which may assist clinical decision-maker.