Independent Test Cohort Research Articles

A new radiomics approach combining the tumor and peri-tumor regions to predict lymph node metastasis and prognosis in gastric cancer.

The development of non-invasive methods for evaluating lymph node metastasis (LNM) preoperatively in gastric cancer (GC) is necessary. In this study, we developed a new radiomics model combining features from the tumor and peri-tumor regions for predicting LNM and prognoses. This was a retrospective observational study. In this study, two cohorts of patients with GC treated in Zhongshan Hospital Fudan University (Shanghai, China) were included. In total, 193 patients were assigned to the internal training/validation cohort; another 98 patients were assigned to the independent testing cohort. The radiomics features were extracted from venous phase computerized tomography (CT) images. The radiomics model was constructed and the output was defined as the radiomics score (RS). The performance of the RS and CT-defined N status (ctN) for predicting LNM was compared using the area under the curve (AUC). The 5-year overall survival and progression-free survival were compared between different subgroups using Kaplan-Meier curves. In both cohorts, the RS was significantly higher in the LNM-positive group than that in the LNM-negative group (all P < 0.001). The radiomics model combining features from the tumor and peri-tumor regions achieved the highest AUC in predicting LNM (AUC, 0.779 and 0.724, respectively), which performed better than the radiomics model based only on the tumor region and ctN (AUC, 0.717, 0.622 and 0.710, 0.603, respectively). The differences in 5-year overall survival and progression-free survival between high-risk and low-risk groups were significant (both P < 0.001). The radiomics model combining features from the tumor and peri-tumor regions could effectively predict the LNM in GC. Risk stratification based on the RS was capable of distinguishing patients with poor prognoses.

Development and validation of a combined model based on dual-sequence MRI radiomics for predicting the efficacy of high-intensity focused ultrasound ablation for hysteromyoma

Objectives To determine the value of dual-sequence magnetic resonance imaging (MRI)-based radiomics in predicting the efficacy of high-intensity focused ultrasound (HIFU) ablation for hysteromyoma. Methods A total of 142 patients with 172 hysteromyomas (95 hysteromyomas from the sufficient ablation group, and 77 hysteromyomas from the insufficient ablation group) were enrolled in the study. The clinical–radiological model was constructed with independent clinical–radiological risk factors, the radiomics model was constructed based on the optimal radiomics features of hysteromyoma from dual sequences, and the two groups of features were incorporated to construct the combined model. A fivefold cross validation procedure was adopted to validate these models. A nomogram was constructed, applying the combined model in the training cohort. The models were assessed with receiver operating characteristic (ROC) curves and integrated discrimination improvement (IDI). An independent test cohort comprising 40 patients was used to evaluate the performance of the optimal model. Results Among the three models, the average areas under the ROC curves (AUC) of the radiomics model and combined model were 0.803 (95% confidence interval (CI): 0.726–0.881) and 0.841 (95% CI: 0.772–0.909), which were better than the clinical–radiological model in the training cohort. The IDI showed that the combined model had the best prediction accuracy. The combined model also showed good discrimination in both the validation cohort (AUC = 0.834) and the independent test cohort (AUC = 0.801). Conclusion The combined model based on the dual-sequence MRI radiomics is the most promising tool from our study to assist clinicians in predicting HIFU ablation efficacy.

Risk of primary osteoporosis score (RPOPs): an algorithm model for primary osteoporosis risk assessment in grass-roots hospital

BackgroundThis study aimed to develop and validate a lasso regression algorithm model which was established by correlation factors of bone mineral density (BMD) and could be accurately predicted a high-risk population of primary osteoporosis (POP). It provides a rapid, economical and acceptable early screening method for osteoporosis in grass-roots hospitals.MethodsWe collected 120 subjects from primary osteoporosis screening population in Zhejiang Quhua Hospital between May 2021 and November 2021 who were divided into three groups (normal, osteopenia and osteoporosis) according to the BMD T-score. The levels of three micro-RNAs in the plasma of these people were detected and assessed by qRT-PCR. At the same time, the levels of β-CTX and t-P1NP in serum of the three groups were determined. Based on the cluster random sampling method, 84 subjects (84/120, 70%) were selected as the training set and the rest were the test set. Lasso regression was used to screen characteristic variables and establish an algorithm model to evaluate the population at high risk of POP which was evaluated and tested in an independent test cohort. The feature variable screening process was used 10-fold cross validation to find the optimal lambda.ResultsThe osteoporosis risk score was established in the training set: Risk of primary osteoporosis score (RPOPs) = -0.1497785 + 2.52Age − 0.19miR21 + 0.35miR182 + 0.17β-CTx. The sensitivity, precision and accuracy of RPOPs in an independent test cohort were 79.17%, 82.61% and 75%, respectively. The AUC in the test set was 0.80. Some risk factors have a significant impact on the abnormal bone mass of the subjects. These risk factors were female (p = 0.00013), older than 55 (p < 2.2e-16) and BMI < 24 (p = 0.0091) who should pay more attention to their bone health.ConclusionIn this study, we successfully constructed and validated an early screening model of osteoporosis that is able to recognize people at high risk for developing osteoporosis and remind them to take preventive measures. But it is necessary to conduct further external and prospective validation research in large sample size for RPOPs prediction models.

A radiomics nomogram for the ultrasound-based evaluation of central cervical lymph node metastasis in papillary thyroid carcinoma.

To develop and validate a radiomics nomogram based on ultrasound (US) to predict central cervical lymph node (LN) metastasis in patients with papillary thyroid carcinoma (PTC). PTC patients with pathologically confirmed presence or absence of central cervical LN metastasis in our hospital between March 2021 and November 2021 were enrolled as the training cohort. Radiomics features were extracted from the preoperative US images, and a radiomics signature was constructed. Univariate and multivariate logistic regression analyses were used to screen out the independent risk factors, and a radiomics nomogram was established. The performance of the model was verified in the independent test cohort of PTC patients who underwent thyroidectomy and cervical LN dissection in our hospital from December 2021 to March 2022. In the independent test cohort, the radiomics model based on long-axis cross-section and short-axis cross-section images outperformed the radiomics models based on either one of these sections (the area under the curve (AUC), 0.69 vs. 0.62 and 0.66). The radiomics signature consisted of 4 selected features. The US radiomics nomogram included the radiomics signature, age, gender, BRAF V600E mutation status, and extrathyroidal extension (ETE) status. In the independent test cohort, the AUC of the receiver operating curve(ROC) of this nomogram was 0.76, outperformingthe clinical model and the radiomics model (0.63 and 0.69, respectively), and also much better than preoperative US examination (AUC, 0.60). Decision curve analysis indicated that the radiomics nomogram was clinically useful. This study presents an efficient and useful US radiomics nomogram that can provide comprehensive information to assist clinicians in the individualized preoperative prediction of central cervical LN metastasis in PTC patients.

Automatic detection and classification of treatment deviations in proton therapy from realistically simulated prompt gamma imaging data.

A clinical study regarding the potential of range verification in proton therapy (PT) by prompt gamma imaging (PGI) is carried out at our institution. Manual interpretation of the detected spot-wise range shift information is time-consuming, highly complex, and therefore not feasible in a broad routine application. Here, we present an approach to automatically detect and classify treatment deviations in realistically simulated PGI data for head-and-neck cancer (HNC) treatments using convolutional neural networks (CNNs) and conventional machine learning (ML) approaches. For 12 HNC patients and 1 anthropomorphic head phantom (n=13), pencil beam scanning (PBS) treatment plans were generated, and 1 field per plan was assumed to be monitored with a PGI slit camera system. In total, 386 scenarios resembling different relevant or non-relevant treatment deviations were simulated on planning and control CTs and manually classified into 7 classes: non-relevant changes (NR) and relevant changes (RE) triggering treatment intervention due to range prediction errors (±RP), setup errors in beam direction (±SE), anatomical changes (AC), or a combination of such errors (CB). PBS spots with reliable PGI information were considered with their nominal Bragg peak position for the generation of two 3D spatial maps of 16× 16×16 voxels containing PGI-determined range shift and proton number information. Three complexity levels of simulated PGI data were investigated: (I) optimal PGI data, (II) realistic PGI data with simulated Poisson noise based on the locally delivered proton number, and (III) realistic PGI data with an additional positioning uncertainty of the slit camera following an experimentally determined distribution. For each complexity level, 3D-CNNs were trained on a data subset (n=9) using patient-wise leave-one-out cross-validation and tested on an independent test cohort (n=4). Both the binary task of detecting RE and the multi-class task of classifying the underlying error source were investigated. Similarly, four different conventional ML classifiers (logistic regression, multilayer perceptron, random forest, and support vector machine) were trained using five previously established handcrafted features extracted from the PGI data and used for performance comparison. On the test data, the CNN ensemble achieved a binary accuracy of 0.95, 0.96, and 0.93 and a multi-class accuracy of 0.83, 0.81, and 0.76 for the complexity levels (I), (II), and (III), respectively. In the case of binary classification, the CNN ensemble detected treatment deviations in the most realistic scenario with a sensitivity of 0.95 and a specificity of 0.88. The best performing ML classifiers showed a similar test performance. This study demonstrates that CNNs can reliably detect relevant changes in realistically simulated PGI data and classify most of the underlying sources of treatment deviations. The CNNs extracted meaningful features from the PGI data with a performance comparable to ML classifiers trained on previously established handcrafted features. These results highlight the potential of a reliable, automatic interpretation of PGI data for treatment verification, which is highly desired for a broad clinical application and a prerequisite for the inclusion of PGI in an automated feedback loop for online adaptive PT.

Development and external validation of an MRI-based neural network for brain metastasis segmentation in the AURORA multicenter study

BackgroundStereotactic radiotherapy is a standard treatment option for patients with brain metastases. The planning target volume is based on gross tumor volume (GTV) segmentation. The aim of this work is to develop and validate a neural network for automatic GTV segmentation to accelerate clinical daily routine practice and minimize interobserver variability. MethodsWe analyzed MRIs (T1-weighted sequence ± contrast-enhancement, T2-weighted sequence, and FLAIR sequence) from 348 patients with at least one brain metastasis from different cancer primaries treated in six centers. To generate reference segmentations, all GTVs and the FLAIR hyperintense edematous regions were segmented manually. A 3D-U-Net was trained on a cohort of 260 patients from two centers to segment the GTV and the surrounding FLAIR hyperintense region. During training varying degrees of data augmentation were applied. Model validation was performed using an independent international multicenter test cohort (n = 88) including four centers. ResultsOur proposed U-Net reached a mean overall Dice similarity coefficient (DSC) of 0.92 ± 0.08 and a mean individual metastasis-wise DSC of 0.89 ± 0.11 in the external test cohort for GTV segmentation. Data augmentation improved the segmentation performance significantly. Detection of brain metastases was effective with a mean F1-Score of 0.93 ± 0.16. The model performance was stable independent of the center (p = 0.3). There was no correlation between metastasis volume and DSC (Pearson correlation coefficient 0.07). ConclusionReliable automated segmentation of brain metastases with neural networks is possible and may support radiotherapy planning by providing more objective GTV definitions.

Automatic diagnosis and grading of Prostate Cancer with weakly supervised learning on whole slide images

Background:The workflow of prostate cancer diagnosis and grading is cumbersome and the results suffer from substantial inter-observer variability. Recent trials have shown potential in using machine learning to develop automated systems to address this challenge. Most automated deep learning systems for prostate cancer Gleason grading focused on supervised learning requiring demanding fine-grained pixel-level annotations. Methods:A weakly-supervised deep learning model with slide-level labels is presented in this study for the diagnosis and grading of prostate cancer with whole slide image (WSI). WSIs are first cropped into small patches and then processed with a deep learning model to extract patch-level features. A graph convolution network (GCN) is used to aggregate the features for classifications. Throughout the training process, the noisy labels are progressively filtered out to reduce inter-observer variations in clinical reports. Finally, multi-center independent test cohorts with 6,174 slides are collected to evaluate the prostate cancer diagnosis and grading performance of our model. Results:The cancer diagnosis (2-level classification) results on two external test sets (n= 4,675, n= 844) show an area under the receiver operating characteristic curve (AUC) of 0.985 and 0.986. The Gleason grading (6-level classification) results reach 0.931 quadratic weighted kappa on the internal test set (n= 531). It generalizes well on the external test dataset (n= 844) with 0.801 quadratic weighted kappa with the reference standard set independently. The model enables pathological meaningful interpretability by visualizing the most attended lesions which are highly consistent with expert annotations. Conclusion:The proposed model incorporates a graph network in weakly supervised learning with only slide-level reports. A robust learning strategy is also employed to correct the label noise. It is highly accurate (>0.985 AUC for diagnosis) and also interpretable with intuitive heatmap visualization. It can be unified with a digital pathology pipeline to deliver prostate cancer metrics for a pathology report.

Effect of learning parameters on the performance of the U-Net architecture for cell nuclei segmentation from microscopic cell images.

Nuclei segmentation of cells is the preliminary and essential step of pathological image analysis. However, robust and accurate cell nuclei segmentation is challenging due to the enormous variability of staining, cell sizes, morphologies, cell adhesion or overlapping of the nucleus. The automation process to find the cell's nuclei is a giant leap in this direction and has an important step toward bioimage analysis using software tools. This article extensively analyzes deep U-Net architecture and has been applied to the Data Science Bowl dataset to segment the cell nuclei. The dataset undergoes various preprocessing tasks such as resizing, intensity normalization and data augmentation prior to segmentation. The complete dataset then undergoes the rigorous training and validation process to find the optimized hyperparameters and then the optimized model selection. The mean (m) ± standard deviation (SD) of Intersection over Union (IoU) and F1-score (Dice score) have been calculated along with accuracy during the training and validation process, respectively. The optimized U-Net model results in a training IoU of 0.94 ± 0.16 (m ± SD), an F1-score of 0.94 ± 0.17 (m ± SD), a training accuracy of 95.54 and validation accuracy of 95.45. With this model, we applied a completely independent test cohort of the dataset and obtained the mean IOU of 0.93, F1-score of 0.9311, and mean accuracy of 94.12, respectively to measure the segmentation performance.

Abstract 15312: Automatic Left Atrial Volume and Sphericity Index Calculation From Cardiac CT Using Computer Graphics Imaging and Deep Learning

Introduction: Left atrial (LA) volume and sphericity index are important for atrial fibrillation (AF) recurrence prediction after ablation and AF-associated atrial remodeling analysis. However, calculating these indexes from computed tomography (CT) requires manual outlining, which is subjective and difficult to scale. Hypothesis: We hypothesized LA body can be "virtually" dissected from electroanatomic shells using computer graphics imaging (CGI). We further hypothesized that the results of virtual dissection can be used to train a deep learning (DL) model for automatic LA volume and sphericity index calculation. Methods: First, by encoding the anatomic knowledge with geometric modeling, we developed a mathematical method (termed "virtual dissection") based on CGI techniques to isolate the LA body from electroanatomic shells (Fig A). Second, we used virtual dissection to train a DL model to segment CTs in N=20 patient atria (Fig B-C). Finally, we used this pipeline to segment raw CTs in a test cohort of N=100 patients (30.0% women, 66.0±10.3Y; Fig D). Results: In the independent test cohort (N=100), the LA volume and sphericity index were automatically calculated from DL-based segmentation, with the agreement within 4.04 ml (95% CI: -11.48 to 19.55; r=0.98; p &lt; 0.0001; Fig E-F) and 2.75% (95% CI: -5.71 to 11.22; r=0.88; p &lt; 0.0001; Fig G-H), compared to expert annotation. Conclusions: Encoded anatomical knowledge combined with deep learning in small datasets can accurately extract 3D cardiac models from raw CT scans in large populations, yielding automatic calculation of indexes such as volume and sphericity. This computational pipeline may simplify the outcome analysis and prediction using geometric indexes as predictors.

Abstract 11648: ECG-Derived Features Enable Prediction of Sudden Death in Ischemic Cardiomyopathy

Introduction: Low left ventricular ejection fraction (LVEF) is an imperfect predictor of sudden cardiac death (SCD) in patients with ischemic cardiomyopathy. Novel features identified via automated ECG analysis might improve the prediction of sudden cardiac death risk. Hypothesis: We hypothesized that machine learning of the ECG can be used to predict SCD. Methods: We studied 5603 ECG Lead V1 beats in 41 patients (64±10 Y) with coronary disease and LVEF≤40% in steady-state pacing. Patients were randomly allocated to independent training and test cohorts in a 70:30 ratio, repeated K=10-fold. Support vector machines were trained to predict mortality at 3Y from the top 20 features derived from these beats (Fig C). Patient-level predictions were made by computing an ECG score that indicates the proportion of test set beats in that patient computed by the beat-level model to predict SCD. Results: Fig A shows the data flow in the study. Beat-level predictions in the validation (n=1678 Lead I beats) cohorts yielded c-statistics of 0.78 for SCD (95% CI, 0.62-0.91). In multivariable models, c-statistic was 0.87 for SCD (95% CI, 0.76-0.98) (Fig B). Conclusions: Machine learning of the ECG reveals novel predictors of SCD risk in patients with ischemic cardiomyopathy. This tool may improve risk stratification and allocation for ICD therapy beyond LVEF alone.

A Subregion-Based Survival Prediction Based on Multi-Parametric MRIs for Glioblastoma

<h3>Purpose/Objective(s)</h3> To predict the survival of GBM patients by using multi-parametric MR images, we developed a novel framework of subregion-based radiomic analysis. This study aims to search for the best combinations among tumor subregions and various MRI sequences for GBM survival prediction. <h3>Materials/Methods</h3> 119 GBM patients provided by BraTS 2020 were used in the study. 100 cases were used for training and evaluating under 5-fold cross-validation and the external 19 cases were used for independent testing. For each tumor subregion, we used open-source software to extract 680 subregional radiomic features from one MRI sequence for subsequent analysis. We proposed a novel algorithm to uncover an optimal MRIs combination from multiple MRI sequences. By a one-to-one matching relation between 4 sequences MRIs (T1/T1ce/T2/FLAIR) and 3 tumor subregions (gadolinium-enhancing tumor (ET), peritumoral edema (ED), and non-enhancing tumor core (NET)), we evaluated the 64 (4³) kinds of subregional radiomic features combinations (680 × 3) based on the evaluation of maximizing the performance of survival stage classification between long-stage (>365 days) and short-stage (<365 days) groups using the SVC (Support Vector Classification) classifier. The best combination with a quarter size compared to the initial imaging data (680 × 12) can be determined and achieved precisely prediction. <h3>Results</h3> As shown in table, the specific matching relation of NET-FLAIR, ED-T1ce, ET-T2 which used 3-sequence resulted in the highest AUCs of 0.998 and 0.983 on the training and independent test cohort respectively. The model generated from optimized 3 MR sequences of features was superior to those models using one single-sequence, 2-sequence, other combinations of 3-sequence and 4-sequence MRIs. <h3>Conclusion</h3> We demonstrated the efficacy of a novel subregion-based framework for survival prediction of GBM by using multi-parametric MRIs. The proposed approach effectively combined multiple MR images and achieved survival prediction precisely.

Head and neck cancer predictive risk estimator to determine control and therapeutic outcomes of radiotherapy (HNC-PREDICTOR): development, international multi-institutional validation, and web implementation of clinic-ready model-based risk stratification for head and neck cancer

BackgroundPersonalised radiotherapy can improve treatment outcomes of patients with head and neck cancer (HNC), where currently a ‘one-dose-fits-all’ approach is the standard. The aim was to establish individualised outcome prediction based on multi-institutional international ‘big-data’ to facilitate risk-based stratification of patients with HNC. MethodsThe data of 4611 HNC radiotherapy patients from three academic cancer centres were split into four cohorts: a training (n = 2241), independent test (n = 786), and external validation cohorts 1 (n = 1087) and 2 (n = 497). Tumour- and patient-related clinical variables were considered in a machine learning pipeline to predict overall survival (primary end-point) and local and regional tumour control (secondary end-points); serially, imaging features were considered for optional model improvement. Finally, patients were stratified into high-, intermediate-, and low-risk groups. ResultsPerformance score, AJCC8thstage, pack-years, and Age were identified as predictors for overall survival, demonstrating good performance in both the training cohort (c-index = 0.72 [95% CI, 0.66–0.77]) and in all three validation cohorts (c-indices: 0.76 [0.69–0.83], 0.73 [0.68–0.77], and 0.75 [0.68–0.80]). Excellent stratification of patients with HNC into high, intermediate, and low mortality risk was achieved; with 5-year overall survival rates of 17–46% for the high-risk group compared to 92–98% for the low-risk group. The addition of morphological image feature further improved the performance (c-index = 0.73 [0.64–0.81]). These models are integrated in a clinic-ready interactive web interface: https://uic-evl.github.io/hnc-predictor/ ConclusionsRobust model-based prediction was able to stratify patients with HNC in distinct high, intermediate, and low mortality risk groups. This can effectively be capitalised for personalised radiotherapy, e.g., for tumour radiation dose escalation/de-escalation.

Feasibility of Using High-Resolution Computed Tomography Features for Invasiveness Differentiation of Malignant Nodules Manifesting as Ground-Glass Nodules.

Ground-glass nodule (GGN)-like adenocarcinoma is a special subtype of lung cancer. The invasiveness of the nodule correlates well with the patient's prognosis. This study aimed to establish a radiomic model for invasiveness differentiation of malignant nodules manifesting as ground glass on high-resolution computed tomography (HRCT). Between January 2014 and July 2019, 276 pulmonary nodules manifesting as GGNs on preoperative HRCTs, whose histological results were available, were collected. The nodules were randomly classified into training (n = 221) and independent testing (n = 55) cohorts. Three logistic models using features derived from HRCT were fit in the training cohort and validated in both aforementioned cohorts for invasive adenocarcinoma and preinvasive-minimally invasive adenocarcinoma (MIA) differentiation. The model with the best performance was presented as a nomogram and was validated using a calibration curve before performing a decision curve analysis. The benefit of using the proposed model was also shown by groups of management strategies recommended by The Fleischner Society. The combined model showed the best differentiation performance (area under the curve (AUC), training set = 0.89, and testing set = 0.92). The quantitative texture model showed better performance (AUC, training set = 0.87, and testing set = 0.91) than the semantic model (AUC, training set = 0.83, and testing set = 0.79). Of the 94 type 2 nodules that were IACs, 66 were identified by this model. Models using features derived from imaging are effective for differentiating between preinvasive-MIA and IACs among lung adenocarcinomas appearing as GGNs on CT images.

Can Persistent Homology Features Capture More Intrinsic Information about Tumors from 18F-Fluorodeoxyglucose Positron Emission Tomography/Computed Tomography Images of Head and Neck Cancer Patients?

This study hypothesized that persistent homology (PH) features could capture more intrinsic information about the metabolism and morphology of tumors from 18F-fluorodeoxyglucose positron emission tomography (PET)/computed tomography (CT) images of patients with head and neck (HN) cancer than other conventional features. PET/CT images and clinical variables of 207 patients were selected from the publicly available dataset of the Cancer Imaging Archive. PH images were generated from persistent diagrams obtained from PET/CT images. The PH features were derived from the PH PET/CT images. The signatures were constructed in a training cohort from features from CT, PET, PH-CT, and PH-PET images; clinical variables; and the combination of features and clinical variables. Signatures were evaluated using statistically significant differences (p-value, log-rank test) between survival curves for low- and high-risk groups and the C-index. In an independent test cohort, the signature consisting of PH-PET features and clinical variables exhibited the lowest log-rank p-value of 3.30 × 10−5 and C-index of 0.80, compared with log-rank p-values from 3.52 × 10−2 to 1.15 × 10−4 and C-indices from 0.34 to 0.79 for other signatures. This result suggests that PH features can capture the intrinsic information of tumors and predict prognosis in patients with HN cancer.

Automatic left atrial segmentation from cardiac CT using computer graphics imaging and deep learning

Abstract Introduction Segmenting left atrial (LA) substructures, including the LA body, appendage (LAA), and pulmonary veins (PVs), from computed tomography (CT) is central to electroanatomic mapping for ablation and functional studies in patients with atrial fibrillation (AF). However, this process requires manual outlining which needs special training, is subjective, and is difficult to scale. Computer graphics imaging (CGI) has been applied in media, film, and computer-aided design to reliably segment complex structures using their basic geometric representations. Purpose We hypothesized that LA substructures can be “virtually” dissected using CGI to separate geometric contours of the “convex ellipsoid” LA, “tubular” PVs, and “conical” LAA. We further hypothesized that the results of virtual dissection can be used to train a deep learning (DL) model to segment raw CT scans. Methods First, a mathematical method based on CGI techniques – erosion and dilation – was developed to “virtually dissect” the convex LA body from the original concave shell in publicly available digital atria with diverse simulated morphologies (Fig. 1A). The PVs and LAA were then automatically revealed and labeled by a 3D subtraction approach. Second, we refined precise LA/PV/LAA boundaries by tuning hyper-parameters from N=5 patient shells (Fig. 1B). Third, we used virtual dissection to train a DL model to segment CTs in N=20 patient atria (Fig. 1C). Finally, we applied this pipeline to segment raw CTs in a validation cohort of N=105 patients (23.8% women, 63.8±10.3Y; Fig. 1D). Results Virtual dissection accurately identified LA/PV/LAA boundaries in the training set (Dice coefficients 89–98%). In the independent test cohort (N=105), this automated pipeline accurately segmented raw CTs with Dice 81–95% (Fig. 1D) compared to a panel of experts (p&amp;lt;0.001). Conclusion CGI of basic cardiac geometry combined with deep learning in small datasets can accurately segment raw CT scans in large populations. This computational pipeline may automate and simplify cardiac image processing and ablation procedures, and could be applied to the ventricle or other organ systems for diverse therapeutic strategies or to train machine learning. Funding Acknowledgement Type of funding sources: Public grant(s) – National budget only. Main funding source(s): National Institutes of Health

Wavelet transformation can enhance computed tomography texture features: a multicenter radiomics study for grade assessment of COVID-19 pulmonary lesions.

BackgroundThis study set out to develop a computed tomography (CT)-based wavelet transforming radiomics approach for grading pulmonary lesions caused by COVID-19 and to validate it using real-world data.MethodsThis retrospective study analyzed 111 patients with 187 pulmonary lesions from 16 hospitals; all patients had confirmed COVID-19 and underwent non-contrast chest CT. Data were divided into a training cohort (72 patients with 127 lesions from nine hospitals) and an independent test cohort (39 patients with 60 lesions from seven hospitals) according to the hospital in which the CT was performed. In all, 73 texture features were extracted from manually delineated lesion volumes, and 23 three-dimensional (3D) wavelets with eight decomposition modes were implemented to compare and validate the value of wavelet transformation for grade assessment. Finally, the optimal machine learning pipeline, valuable radiomic features, and final radiomic models were determined. The area under the receiver operating characteristic (ROC) curve (AUC), calibration curve, and decision curve were used to determine the diagnostic performance and clinical utility of the models.ResultsOf the 187 lesions, 108 (57.75%) were diagnosed as mild lesions and 79 (42.25%) as moderate/severe lesions. All selected radiomic features showed significant correlations with the grade of COVID-19 pulmonary lesions (P<0.05). Biorthogonal 1.1 (bior1.1) LLL was determined as the optimal wavelet transform mode. The wavelet transforming radiomic model had an AUC of 0.910 in the test cohort, outperforming the original radiomic model (AUC =0.880; P<0.05). Decision analysis showed the radiomic model could add a net benefit at any given threshold of probability.ConclusionsWavelet transformation can enhance CT texture features. Wavelet transforming radiomics based on CT images can be used to effectively assess the grade of pulmonary lesions caused by COVID-19, which may facilitate individualized management of patients with this disease.

Development and validation of a clinical-radiomics nomogram for predicting a poor outcome and 30-day mortality after a spontaneous intracerebral hemorrhage

BackgroundNoncontrast computed tomography (NCCT) is often performed for patients with a suspected spontaneous intracerebral hemorrhage (ICH) at the time of admission. Both clinical and radiomic features on the initial NCCT can predict the outcomes of those with ICH, but satisfactory model performance remains challenging.MethodsA total of 258 acute ICH patients from the Central Hospital of Wuhan (CHW) between January 2018 and December 2020 were retrospectively assigned to training and internal validation cohorts at a ratio of 7:3. An independent external testing cohort of 87 patients from January 2021 to July 2021 from the Fifth Affiliated Hospital of Nanchang University (FAHNU) was also used. Based on the least absolute shrinkage and selection operator (LASSO) algorithm, radiomics (rad)-scores were generated from 9 quantitative features on the initial NCCT images. Three models (radiomics, clinical, and hybrid) were established using stepwise logistic regression analysis. The Akaike information criterion and the likelihood ratio test were used to compare the goodness of fit of the three models. Receiver operating characteristic (ROC) curve analysis was performed and bar charts were constructed to evaluate the discrimination of constructed model for predicting a poor outcome following ICH.ResultsThe three cohorts had similar baseline clinical characteristics, including demographic features and outcomes. In the clinical model, hematoma expansion [2.457 (0.297, 2.633); P=0.014], intracerebral ventricular hemorrhage [2.374 (0.180, 1.882); P=0.018], and location [−2.268 (−2.578, −0.188); P=0.023] were independently associated with a poor clinical outcome. In the hybrid model, location [−2.291 (−2.925, −0.228); P=0.022], and rad-score [5.255 (0.680, 11.460); P<0.001] were independently associated with a poor outcome. The hybrid model achieved satisfactory discriminability, with areas under curve (AUCs) of 0.892 [95% confidence interval (CI): 0.847 to 0.937], 0.893 (95% CI: 0.820 to 0.966), and 0.838 (95% CI: 0.755 to 0.920) in the training, internal validation, and external testing cohorts, respectively. The hybrid model also achieved good discriminability in the prediction of 30-day mortality, with AUCs of 0.840, 0.823, and 0.883 in the training, internal validation, and external testing cohorts, respectively. The rad-score [2.861 (1.940, 4.220); P<0.001] was the predominant risk factor associated with 30-day mortality.ConclusionsRadiomic analysis based on initial NCCT scans showed added value in predicting a poor outcome after ICH. A clinical-radiomics model yielded improved accuracy in predicting a poor outcome and 30-day death following ICH compared with radiomics alone.

Differentiation of malignant from benign pleural effusions based on artificial intelligence

IntroductionThis study aimed to construct artificial intelligence models based on thoracic CT images to perform segmentation and classification of benign pleural effusion (BPE) and malignant pleural effusion (MPE).MethodsA total of...

Development and validation of a deep learning signature for predicting lymph node metastasis in lung adenocarcinoma: comparison with radiomics signature and clinical-semantic model.

To develop and validate a deep learning (DL) signature for predicting lymph node (LN) metastasis in patients with lung adenocarcinoma. A total of 612 patients with pathologically-confirmed lung adenocarcinoma were retrospectively enrolled and were randomly divided into training cohort (n = 489) and internal validation cohort (n = 123). Besides, 108 patients were enrolled and constituted an independent test cohort (n = 108). Patients' clinical characteristics and CT semantic features were collected. The radiomics features were derived from contrast-enhanced CT images. The clinical-semantic model and radiomics signature were built to predict LN metastasis. Furthermore, Swin Transformer was adopted to develop a DL signature predictive of LN metastasis. Model performance was evaluated by area under the receiver operating characteristic curve (AUC), sensitivity, specificity, calibration curve, and decision curve analysis. The comparisons of AUC were conducted by the DeLong test. The proposed DL signature yielded an AUC of 0.948-0.961 across all three cohorts, significantly superior to both clinical-semantic model and radiomics signature (all p < 0.05). The calibration curves show that DL signature predicted probabilities fit well the actual observed probabilities of LN metastasis. DL signature gained a higher net benefit than both clinical-semantic model and radiomics signature. The incorporation of radiomics signature or clinical-semantic risk predictors failed to reveal an incremental value over the DL signature. The proposed DL signature based on Swin Transformer achieved a promising performance in predicting LN metastasis and could confer important information in noninvasive mediastinal LN staging and individualized therapeutic options. • Accurate prediction for lymph node metastasis is crucial to formulate individualized therapeutic options for patients with lung adenocarcinoma. • The deep learning signature yielded an AUC of 0.948-0.961 across all three cohorts in predicting lymph node metastasis, superior to both radiomics signature and clinical-semantic model. • The incorporation of radiomics signature or clinical-semantic risk predictors into deep learning signature failed to reveal an incremental value over deep learning signature.

Gut metagenome-derived signature predicts hepatic decompensation and mortality in NAFLD-related cirrhosis.

There are limited data on the diagnostic accuracy of gut microbial signatures for predicting hepatic decompensation in patients with cirrhosis. To determine whether a stool metagenome-derived signature accurately detects hepatic decompensation and mortality risk in cirrhosis secondary to non-alcoholic fatty liver disease (NAFLD) METHODS: Shotgun metagenomic sequencing was performed on faecal samples collected at study entry from a prospective cohort of adults with NAFLD-related cirrhosis. A Random Forest machine learning algorithm was utilised to identify a metagenomic signature of decompensated cirrhosis (defined by ascites, hepatic encephalopathy or variceal haemorrhage) and subsequently validated in an external cohort. A Cox proportional hazards regression model was used to examine predictors of all-cause mortality. In all, 25 adults with NAFLD-related cirrhosis (training cohort) were included. Among the 16 participants with decompensated cirrhosis, 33% had ascites, 56% had hepatic encephalopathy and 22% had experienced a variceal haemorrhage (not mutually exclusive). We identified a stool metagenomic signature comprising 13 discriminatory species that reliably distinguished decompensated NAFLD-related cirrhosis (diagnostic accuracy, 0.97, 95% confidence interval [CI] 0.96-0.99). Diagnostic accuracy of the 13-species signature remained high after adjustment for lactulose (area under the curve [AUC] 0.99) and rifaximin use (AUC 0.93). The discriminative ability of 13-species metagenomic signature was robust in an independent test cohort (AUC 0.95, 95% CI 0.81-1.00). The 13-species metagenomic signature (hazard ratio [HR] 1.54, 95% CI 1.10-2.15, p= 0.01) was a stronger predictor of mortality than the Model for End-Stage Liver Disease score (HR 1.25, 95% CI 1.03-1.53, p= 0.03). This study provides evidence for a gut metagenome-derived signature with high diagnostic accuracy for hepatic decompensation that predicts risk of mortality in NAFLD-related cirrhosis.