Semi-supervised Long-tail Endoscopic Image Classification.

Objective To explore the semi-supervised learning (SSL) algorithm for long-tail endoscopic image classification with limited annotations. Method We explored semi-supervised long-tail endoscopic image classification in HyperKvasir, the largest gastrointestinal public dataset with 23 diverse classes. Semi-supervised learning algorithm FixMatch was applied based on consistency regularization and pseudo-labeling. After splitting the training dataset and the test dataset at a ratio of 4:1, we sampled 20%, 50%, and 100% labeled training data to test the classification with limited annotations. Results The classification performance was evaluated by micro-average and macro-average evaluation metrics, with the Mathews correlation coefficient (MCC) as the overall evaluation. SSL algorithm improved the classification performance, with MCC increasing from 0.8761 to 0.8850, from 0.8983 to 0.8994, and from 0.9075 to 0.9095 with 20%, 50%, and 100% ratio of labeled training data, respectively. With a 20% ratio of labeled training data, SSL improved both the micro-average and macro-average classification performance; while for the ratio of 50% and 100%, SSL improved the micro-average performance but hurt macro-average performance. Through analyzing the confusion matrix and labeling bias in each class, we found that the pseudo-based SSL algorithm exacerbated the classifier's preference for the head class, resulting in improved performance in the head class and degenerated performance in the tail class. Conclusion SSL can improve the classification performance for semi-supervised long-tail endoscopic image classification, especially when the labeled data is extremely limited, which may benefit the building of assisted diagnosis systems for low-volume hospitals. However, the pseudo-labeling strategy may amplify the effect of class imbalance, which hurts the classification performance for the tail class.

Deep learning for predicting immunotherapeutic efficacy in advanced non-small cell lung cancer patients: a retrospective study combining progression-free survival risk and overall survival risk.

Background: Radiomics based on computed tomography (CT) images is potential in promoting individualized treatment of non-small cell lung cancer (NSCLC), however, its role in immunotherapy needs further exploration. The aim of this study was to develop a CT-based radiomics score to predict the efficacy of immune checkpoint inhibitor (ICI) monotherapy in patients with advanced NSCLC. Methods: Two hundred and thirty-six ICI-treated patients were retrospectively included and divided into a training cohort (n=188) and testing cohort (n=48) at a ratio of 8 to 2. The efficacy outcomes of ICI were evaluated based on overall survival (OS) and progression-free survival (PFS). We designed a survival network and combined it with a Cox regression model to obtain patients' OS risk score (OSRS) and PFS risk score (PFSRS). Results: Based on OSRS and PFSRS, patients were divided into high- and low-risk groups in the training cohort and the test cohort with distinctly different [training cohort, log-rank P<0.001, hazard ratio (HR): 4.14; test cohort, log-rank P=0.014, HR: 4.54] and PFS (training cohort, log-rank P<0.001, HR: 4.52; test cohort, log-rank P<0.001, HR: 6.64). Further joint evaluation of OSRS and PFSRS showed that both were significant in the Cox regression model (P<0.001), and multi-overall survival risk score (MOSRS) displayed more outstanding stratification capabilities than OSRS in both the training (P<0.001) and test cohorts (P=0.002). None of the clinical characteristics were significant in the Cox regression model, and the score that predicted the best immune response was not as good as the risk score from follow-up information in the performance of prognostic stratification. Conclusions: We developed a CT imaging-based score with the potential to become an independent prognostic factor to screen patients who would benefit from ICI treatment, which suggested that CT radiomics could be applied for individualized immunotherapy of NSCLC. Our findings should be further validated by future larger multicenter study.

CT-based deep learning radiomics analysis for evaluation of serosa invasion in advanced gastric cancer.

PURPOSE: This work aimed to develop and validate a deep learning radiomics model for evaluating serosa invasion in gastric cancer. MATERIALS AND METHODS: A total of 572 gastric cancer patients were included in this study. Firstly, we retrospectively enrolled 428 consecutive patients (252 in the training set and 176 in the test set I) with pathological confirmed T3 or T4a. Subsequently, 144 patients who were clinically diagnosed cT3 or cT4a were prospectively allocated to the test set II. Histological verification was based on the surgical specimens. CT findings were determined by a panel of three radiologists. Conventional hand-crafted features and deep learning features were extracted from three phases CT images and were utilized to build radiomics signatures via machine learning methods. Incorporating the radiomics signatures and CT findings, a radiomics nomogram was developed via multivariable logistic regression. Its diagnostic ability was measured using receiver operating characteristiccurve analysis. RESULTS: The radiomics signatures, built with support vector machine or artificial neural network, showed good performance for discriminating T4a in the test I and II sets with area under curves (AUCs) of 0.76-0.78 and 0.79-0.84. The nomogram had powerful diagnostic ability in all training, test I and II sets with AUCs of 0.90 (95 % CI, 0.86-0.94), 0.87 (95 % CI, 0.82-0.92) and 0.90 (95 % CI, 0.85-0.96) respectively. The net reclassification index revealed that the radiomics nomogram had significantly better performance than the clinical model (p-values < 0.05). CONCLUSIONS: The deep learning radiomics model based on CT images is effective at discriminating serosa invasion in gastric cancer.

A deep learning MR-based radiomic nomogram may predict survival for nasopharyngeal carcinoma patients with stage T3N1M0.

PURPOSE: To estimate the prognostic value of deep learning (DL) magnetic resonance (MR)-based radiomics for stage T3N1M0 nasopharyngeal carcinoma (NPC) patients receiving induction chemotherapy (ICT) prior to concurrent chemoradiotherapy (CCRT). METHODS: A total of 638 stage T3N1M0 NPC patients (training cohort: n = 447; test cohort: n = 191) were enrolled and underwent MRI scans before receiving ICT + CCRT. From the pretreatment MR images, DL-based radiomic signatures were developed to predict disease-free survival (DFS) in an end-to-end way. Incorporating independent clinical prognostic parameters and radiomic signatures, a radiomic nomogram was built through multivariable Cox proportional hazards method. The discriminative performance of the radiomic nomogram was assessed using the concordance index (C-index) and the Kaplan-Meier estimator. RESULTS: Three DL-based radiomic signatures were significantly correlated with DFS in the training (C-index: 0.695-0.731, all p < 0.001) and test (C-index: 0.706-0.755, all p < 0.001) cohorts. Integrating radiomic signatures with clinical factors significantly improved the predictive value compared to the clinical model in the training (C-index: 0.771 vs. 0.640, p < 0.001) and test (C-index: 0.788 vs. 0.625, p = 0.001) cohorts. Furthermore, risk stratification using the radiomic nomogram demonstrated that the high-risk group exhibited short-lived DFS compared to the low-risk group in the training cohort (hazard ratio [HR]: 6.12, p < 0.001), which was validated in the test cohort (HR: 6.90, p < 0.001). CONCLUSIONS: Our DL-based radiomic nomogram may serve as a noninvasive and useful tool for pretreatment prognostic prediction and risk stratification in stage T3N1M0 NPC.

Heterogeneity of metastatic gastrointestinal stromal tumor on texture analysis: DWI texture as potential biomarker of overall survival.

PURPOSE: To determine if texture features of diffusion weighted imaging (DWI) on MRI of metastatic gastrointestinal stromal tumor (mGIST) have correlation with overall survival (OS). METHOD: Fifty-one GIST patients with metastatic lesions who received imatinib targeted therapy were included. Texture features of the largest metastatic lesion were analyzed using inhouse software. Three types of texture features were assessed: fractal features, gray-level co-occurrence matrix (GLCM) features, and gray-level run-length matrix (GLRLM) features. The features were extracted from the regions of interest (ROIs) on T2-weighted imaging (T2WI), DWI and apparent diffusion coefficient (ADC) maps. Histogram analysis was performed on ADC maps. Patients were followed up until death. Kaplan-Meier analysis was performed to determine the correlation of texture features with OS. The curves of the high- and low-risk groups were compared using log-rank test. The prognostic efficacy of the predictors was assessed by calculating the concordance probability. RESULTS: The median survival time was 43.5 months (range, 3.97-120.90 m). Four DWI and three ADC texture features showed significant correlation with OS on univariate analysis (p < 0.05). DWI_L_GLCM_maximum_probability [hazard ratio (HR): 2.062 (1.357-3.131)], ADC_H_GLRLM_mean [HR: 2.174 (1.457-3.244)], and ADC_O_GLCM_cluster_shade [HR: 1.882 (1.324-2.674)] were identified as representative prognostic indicators. The optimum threshold levels for these three features were 1.19×100, 1.71×10 and 2.19×0.1, respectively. Neither histogram analysis values nor fractal features revealed significant correlation with survival status (p > 0.05). CONCLUSIONS: Texture features of the mGIST on DWI exhibited correlation with overall survival. High-grade heterogeneity was associated with poor prognosis.

Development and validation of a novel MR imaging predictor of response to induction chemotherapy in locoregionally advanced nasopharyngeal cancer: a randomized controlled trial substudy (NCT01245959).

BACKGROUND: In locoregionally advanced nasopharyngeal carcinoma (LANPC) patients, variance of tumor response to induction chemotherapy (ICT) was observed. We developed and validated a novel imaging biomarker to predict which patients will benefit most from additional ICT compared with chemoradiotherapy (CCRT) alone. METHODS: All patients, including retrospective training (n = 254) and prospective randomized controlled validation cohorts (a substudy of NCT01245959, n = 248), received ICT+CCRT or CCRT alone. Primary endpoint was failure-free survival (FFS). From the multi-parameter magnetic resonance images of the primary tumor at baseline, 819 quantitative 2D imaging features were extracted. Selected key features (according to their interaction effect between the two treatments) were combined into an Induction Chemotherapy Outcome Score (ICTOS) with a multivariable Cox proportional hazards model using modified covariate method. Kaplan-Meier curves and significance test for treatment interaction were used to evaluate ICTOS, in both cohorts. RESULTS: Three imaging features were selected and combined into ICTOS to predict treatment outcome for additional ICT. In the matched training cohort, patients with a high ICTOS had higher 3-year and 5-year FFS in ICT+CCRT than CCRT subgroup (69.3% vs. 45.6% for 3-year FFS, and 64.0% vs. 36.5% for 5-year FFS; HR = 0.43, 95% CI = 0.25-0.74, p = 0.002), whereas patients with a low ICTOS had no significant difference in FFS between the subgroups (p = 0.063), with a significant treatment interaction (pinteraction <  0.001). This trend was also found in the validation cohort with high (n = 73, ICT+CCRT 89.7% and 89.7% vs. CCRT 61.8% and 52.8% at 3-year and 5-year; HR = 0.17, 95% CI = 0.06-0.51, p <  0.001) and low ICTOS (n = 175, p = 0.31), with a significant treatment interaction (pinteraction = 0.019). Compared with 12.5% and 16.6% absolute benefit in the validation cohort (3-year FFS from 69.9 to 82.4% and 5-year FFS from 63.4 to 80.0% from additional ICT), high ICTOS group in this cohort had 27.9% and 36.9% absolute benefit. Furthermore, no significant survival improvement was found from additional ICT in both groups after stratifying low ICTOS patients into low-risk and high-risks groups, by clinical risk factors. CONCLUSION: An imaging biomarker, ICTOS, as proposed, identified patients who were more likely to gain additional survival benefit from ICT+CCRT (high ICTOS), which could influence clinical decisions, such as the indication for ICT treatment. TRIAL REGISTRATION: ClinicalTrials.gov , NCT01245959 . Registered 23 November 2010.

Prognostic Value of Deep Learning PET/CT-Based Radiomics: Potential Role for Future Individual Induction Chemotherapy in Advanced Nasopharyngeal Carcinoma.

PURPOSE: We aimed to evaluate the value of deep learning on positron emission tomography with computed tomography (PET/CT)-based radiomics for individual induction chemotherapy (IC) in advanced nasopharyngeal carcinoma (NPC). EXPERIMENTAL DESIGN: We constructed radiomics signatures and nomogram for predicting disease-free survival (DFS) based on the extracted features from PET and CT images in a training set (n = 470), and then validated it on a test set (n = 237). Harrell's concordance indices (C-index) and time-independent receiver operating characteristic (ROC) analysis were applied to evaluate the discriminatory ability of radiomics nomogram, and compare radiomics signatures with plasma Epstein-Barr virus (EBV) DNA. RESULTS: A total of 18 features were selected to construct CT-based and PET-based signatures, which were significantly associated with DFS (P < 0.001). Using these signatures, we proposed a radiomics nomogram with a C-index of 0.754 [95% confidence interval (95% CI), 0.709-0.800] in the training set and 0.722 (95% CI, 0.652-0.792) in the test set. Consequently, 206 (29.1%) patients were stratified as high-risk group and the other 501 (70.9%) as low-risk group by the radiomics nomogram, and the corresponding 5-year DFS rates were 50.1% and 87.6%, respectively (P < 0.0001). High-risk patients could benefit from IC while the low-risk could not. Moreover, radiomics nomogram performed significantly better than the EBV DNA-based model (C-index: 0.754 vs. 0.675 in the training set and 0.722 vs. 0.671 in the test set) in risk stratification and guiding IC. CONCLUSIONS: Deep learning PET/CT-based radiomics could serve as a reliable and powerful tool for prognosis prediction and may act as a potential indicator for individual IC in advanced NPC.