VerSe: A Vertebrae labelling and segmentation benchmark for multi-detector CT images.

Vertebral labelling and segmentation are two fundamental tasks in an automated spine processing pipeline. Reliable and accurate processing of spine images is expected to benefit clinical decision support systems for diagnosis, surgery planning, and population-based analysis of spine and bone health. However, designing automated algorithms for spine processing is challenging predominantly due to considerable variations in anatomy and acquisition protocols and due to a severe shortage of publicly available data. Addressing these limitations, the Large Scale Vertebrae Segmentation Challenge (VerSe) was organised in conjunction with the International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI) in 2019 and 2020, with a call for algorithms tackling the labelling and segmentation of vertebrae. Two datasets containing a total of 374 multi-detector CT scans from 355 patients were prepared and 4505 vertebrae have individually been annotated at voxel level by a human-machine hybrid algorithm (https://osf.io/nqjyw/, https://osf.io/t98fz/). A total of 25 algorithms were benchmarked on these datasets. In this work, we present the results of this evaluation and further investigate the performance variation at the vertebra level, scan level, and different fields of view. We also evaluate the generalisability of the approaches to an implicit domain shift in data by evaluating the top-performing algorithms of one challenge iteration on data from the other iteration. The principal takeaway from VerSe: the performance of an algorithm in labelling and segmenting a spine scan hinges on its ability to correctly identify vertebrae in cases of rare anatomical variations. The VerSe content and code can be accessed at: https://github.com/anjany/verse.

Differentiation Between Benign and Nonbenign Meningiomas by Using Texture Analysis From Multiparametric MRI.

BACKGROUND: It is difficult to prospectively differentiate between benign (World Health Organization [WHO] I) and nonbenign (WHO II and III) meningiomas. PURPOSE: To evaluate the feasibility of preoperative differentiation between benign and nonbenign meningiomas by using texture analysis from multiparametric MR data. STUDY TYPE: Retrospective. SUBJECTS: In all, 184 patients with meningioma (139 benign and 45 nonbenign) were included as the training cohort and 79 patients with meningioma (60 benign and 19 nonbenign) were included as the external validation cohort. FIELD STRENGTH/SEQUENCE: T1 -weighted, T2 -weighted, and contrast-enhanced T1 -weighted imaging were performed on 1.5 or 3.0T MR systems from two centers. ASSESSMENT: Tumor segmentation and radiological characteristic (RC) evaluation were performed by experienced radiologists. The texture features were extracted from preprocessed images and combined with RCs, and then the combined features were reduced by using a two-step feature selection. Three single-sequence models and a multiparametric MRI (the combination of single sequences) model were constructed and then evaluated with the external validation cohort. STATISTICAL TESTS: Area under receiver operating characteristic curve (AUC), accuracy (Acc), f1-score (F1), sensitivity (Sen), and specificity (Spec), were calculated to quantify the performance of the models. RESULTS: Among the four texture models, the multiparametric MRI model demonstrated the best performance for differentiating between benign and nonbenign meningiomas in both the training and external validation cohorts (AUC 0.91, Acc 89%, F1 0.88, Sen 0.93, and Spec 0.87 in the training cohort; AUC 0.83, Acc 80%, F1 0.77, Sen 0.84, and Spec 0.78 in the validation cohort). DATA CONCLUSION: Nonbenign meningiomas might be preoperatively differentiated from benign meningiomas by using texture analysis from multiparametric MR data. LEVEL OF EVIDENCE: 3 Technical Efficacy Stage: 2 J. Magn. Reson. Imaging 2020;51:1810-1820.

Radiomics analysis of multiparametric MRI for the preoperative evaluation of pathological grade in bladder cancer tumors.

OBJECTIVES: To develop and validate an MRI-based radiomics strategy for the preoperative estimation of pathological grade in bladder cancer (BCa) tumors. METHODS: A primary cohort of 70 patients (31 high-grade BCa and 39 low-grade BCa) with BCa were retrospectively enrolled. Three sets of radiomics features were separately extracted from tumor volumes on T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), and apparent diffusion coefficient (ADC) maps. Two sets of multimodal features were separately generated by the maxout and concatenation of the above mentioned single-modality features. Each feature set was subjected to a two-sample t test and the least absolute shrinkage and selection operator (LASSO) algorithm for feature selection. Multivariable logistic regression (LR) analysis was used to obtain five corresponding radiomics models. The diagnostic abilities of the radiomics models were evaluated using receiver operating characteristic (ROC) curve analysis and compared using the DeLong test. Validation was performed on a time-independent cohort containing 30 consecutive patients. RESULTS: The areas under the ROC curves (AUCs) of single-modality T2WI, DWI, and ADC models in the training cohort were 0.7933 (95% confidence interval [CI] 0.7471-0.8396), 0.8083 (95% CI 0.7565-0.8601), and 0.8350 (95% CI 0.7924-0.8776), respectively. Both multimodality models achieved higher AUCs (maxout 0.9233, 95% CI 0.9001-0.9466; concatenation 0.9233, 95% CI 0.9001-0.9466) than single-modality models. The AUCs of the maxout and concatenation models in the validation cohort were 0.9186 and 0.9276, respectively. CONCLUSIONS: The MRI-based multiparametric radiomics approach has the potential to be used as a noninvasive imaging tool for preoperative grading of BCa tumors. Multicenter validation is needed to acquire high-level evidence for its clinical application. KEY POINTS: • Multiparametric MRI may help in the preoperative grading of BCa tumors. • The Joint_Model established from T2WI, DWI, and ADC feature subsets demonstrated a high diagnostic accuracy for preoperative prediction of pathological grade in BCa tumors. • The radiomics approach has the potential to preoperatively assess tumor grades in BCa and avoid subjectivity.

Multiple mathematical models of diffusion-weighted magnetic resonance imaging combined with prognostic factors for assessing the response to neoadjuvant chemotherapy and radiation therapy in locally advanced rectal cancer.

PURPOSE: To investigate whether the apparent diffusion coefficient (ADC), intravoxel incoherent motion (IVIM), and stretched exponential model (SEM) based on histogram analyses derived from the whole-tumor volume combined with prognostic factors can be used to assess the response to chemotherapy and radiation therapy (CRT) in locally advanced rectal cancer (LARC). MATERIALS AND METHODS: This study included 60 patients with LARC who underwent diffusion-weighted imaging with 9b values (0-1000s/mm2) before CRT. Histograms derived from the whole-tumor volume were used to obtain the ADC, IVIM (Dslow, Dfast, and f), and SEM parameters (distributed diffusion coefficient (DDC) and α). The histogram metrics and prognostic factors before CRT were compared between pathological complete response (pCR) and non-pCR patients. The receiver operating characteristic (ROC) and the area under the ROC curve (AUC) were generated to analyze the histogram metrics and prognostic factors. RESULTS: A significant difference was only found in the tumor volume between the pCR and non-pCR groups (p = 0.033, AUC = 0.740). The ADC mean, DDC median, and most of the histogram metrics were significantly lower in the pCR group than the non-pCR group (p = 0.000-0.025), and AUC was highest for the ADC mean (0.890). Only the Dslow median differed significantly between the two groups (p = 0.023, AUC = 0.721). However, the Dfast, f, and α histogram metrics did not differ significantly between the pCR and non-pCR groups. The AUC for the ADC mean combined with the tumor volume was 0.908, with a sensitivity of 100% and specificity of 81%. The inter-observer agreements were good or excellent for the ADC and SEM histogram parameters but generally fair for IVIM. CONCLUSION: The whole-tumor ADC mean combined with the tumor volume was highly accurate for predicting pCR. The IVIM models were inferior to ADC and SEM at predicting pCR.

Preoperative histogram analysis of intravoxel incoherent motion (IVIM) for predicting microvascular invasion in patients with single hepatocellular carcinoma.

PURPOSE: To evaluate the value of intravoxel incoherent motion (IVIM) histogram analysis based on whole tumor volume in predicting microvascular invasion (MVI) of single hepatocellular carcinoma (HCC). MATERIALS AND METHODS: The study enrolled 41 patients with pathologically proven HCCs who underwent IVIM diffusion-weighted imaging with nine b values and contrast-enhanced magnetic resonance imaging (MRI). Histogram parameters including mean; skewness; kurtosis; and percentiles (5th, 10th, 25th, 50th, 75th, 90th, 95th) were derived from apparent diffusion coefficient (ADC), perfusion fraction (f), true diffusion coefficient (D), and pseudo diffusion coefficient (D*). Quantitative histogram parameters and clinical data were compared between HCCs with and without MVI. For significant parameters, receiver operating characteristic (ROC) curves were further plotted to compare the diagnosis performance for identifying MVI. RESULTS: The mean, 5th, 10th, 25th, 50th, and 75th percentiles of D, and the 5th, 10th, and 25th percentiles of ADC between HCCs with and without MVI were statistically significant (all P＜0.05). The histogram parameters of D* and f showed no statistically significant differences between HCCs with and without MVI (all P＞0.05). The areas under the ROC curves (AUCs) were 0.707-0.874 for D and 0.668-0.720 for ADC. The largest AUC of D (5th percentile) showed significantly higher accuracy than that of ADC or tumor size (P = 0.009-0.046). With a cut-off of 0.403 × 10-3 mm²/s, the 5th percentile of D value provided a sensitivity of 81% and a specificity of 85% in the prediction of MVI. CONCLUSIONS: Histogram analysis of IVIM based on whole tumor volume can be useful for predicting MVI. The 5th percentile of D was most useful value to predict MVI of HCC.

[Differential diagnosis of hepatocellular carcinoma and hepatic hemangiomas based on radiomic features of gadoxetate disodium-enhanced magnetic resonance imaging].

OBJECTIVE: To evaluate the feasibility of using radiomic features for differential diagnosis of hepatocellular carcinoma (HCC) and hepatic cavernous hemangioma (HHE). METHODS: Gadoxetate disodium-enhanced magnetic resonance imaging data were collected from a total of 135 HCC and HHE lesions. The radiomic texture features of each lesion were extracted on the hepatobiliary phase images, and the performance of each feature was assessed in differentiation and classification of HCC and HHE. In multivariate analysis, the performance of 3 feature selection algorithms (namely minimum redundancy-maximum relevance, mRmR; neighborhood component analysis, NCA; and sequence forward selection, SFS) was compared. The optimal feature subset was determined according to the optimal feature selection algorithm and used for testing the 3 classifier algorithms (namely the support vector machine, RBF-SVM; linear discriminant analysis, LDA; and logistic regression). All the tests were repeated 5 times with 10-fold cross validation experiments. RESULTS: More than 50% of the radiomic features exhibited strong distinguishing ability, among which gray level co-occurrence matrix feature S (3, -3) SumEntrp showed a good classification performance with an AUC of 0.72 (P<0.01), a sensitivity of 0.83 and a specificity of 0.57. For the multivariate analysis, 15 features were selected based on the SFS algorithm, which produced better results than the other two algorithms. Testing of these 15 selected features for their average cross-validation performance with RBF-SVM classifier yielded a test accuracy of 0.82∓0.09, an AUC of 0.86∓0.12, a sensitivity of 0.88∓0.11, and a specificity of 0.76∓0.18. CONCLUSION: The radiomic features based on gadoxetate disodium-enhanced magnetic resonance images allow efficient differential diagnosis of HCC and HHE, and can potentially provide important assistance in clinical diagnosis of the two diseases.