Classification of Prostate Transitional Zone Cancer and Hyperplasia Using Deep Transfer Learning From Disease-Related Images.

Purpose The diagnosis of prostate transition zone cancer (PTZC) remains a clinical challenge due to their similarity to benign prostatic hyperplasia (BPH) on MRI. The Deep Convolutional Neural Networks (DCNNs) showed high efficacy in diagnosing PTZC on medical imaging but was limited by the small data size. A transfer learning (TL) method was combined with deep learning to overcome this challenge. Materials and methods A retrospective investigation was conducted on 217 patients enrolled from our hospital database (208 patients) and The Cancer Imaging Archive (nine patients). Using T2-weighted images (T2WIs) and apparent diffusion coefficient (ADC) maps, DCNN models were trained and compared between different TL databases (ImageNet vs. disease-related images) and protocols (from scratch, fine-tuning, or transductive transferring). Results PTZC and BPH can be classified through traditional DCNN. The efficacy of TL from natural images was limited but improved by transferring knowledge from the disease-related images. Furthermore, transductive TL from disease-related images had comparable efficacy to the fine-tuning method. Limitations include retrospective design and a relatively small sample size. Conclusion Deep TL from disease-related images is a powerful tool for an automated PTZC diagnostic system. In developing regions where only conventional MR scans are available, the accurate diagnosis of PTZC can be achieved via transductive deep TL from disease-related images.

Differentiation of Pseudoprogression from True Progressionin Glioblastoma Patients after Standard Treatment: A Machine Learning Strategy Combinedwith Radiomics Features from T1-weighted Contrast-enhanced Imaging.

BACKGROUND: Based on conventional MRI images, it is difficult to differentiatepseudoprogression from true progressionin GBM patients after standard treatment, which isa critical issue associated with survival. The aim of this study was to evaluate the diagnostic performance of machine learning using radiomics modelfrom T1-weighted contrast enhanced imaging(T1CE) in differentiating pseudoprogression from true progression after standard treatment for GBM. METHODS: Seventy-sevenGBM patients, including 51 with true progression and 26 with pseudoprogression,who underwent standard treatment and T1CE, were retrospectively enrolled.Clinical information, including sex, age, KPS score, resection extent, neurological deficit and mean radiation dose, were also recorded collected for each patient. The whole tumor enhancementwas manually drawn on the T1CE image, and a total of texture 9675 features were extracted and fed to a two-step feature selection scheme. A random forest (RF) classifier was trained to separate the patients by their outcomes.The diagnostic efficacies of the radiomics modeland radiologist assessment were further compared by using theaccuracy (ACC), sensitivity and specificity. RESULTS: No clinical features showed statistically significant differences between true progression and pseudoprogression.The radiomic classifier demonstrated ACC, sensitivity, and specificity of 72.78%(95% confidence interval [CI]: 0.45,0.91), 78.36%(95%CI: 0.56,1.00) and 61.33%(95%CI: 0.20,0.82).The accuracy, sensitivity and specificity of three radiologists' assessment were66.23%(95% CI: 0.55,0.76), 61.50%(95% CI: 0.43,0.78) and 68.62%(95% CI: 0.55,0.80); 55.84%(95% CI: 0.45,0.66),69.25%(95% CI: 0.50,0.84) and 49.13%(95% CI: 0.36,0.62); 55.84%(95% CI: 0.45,0.66), 69.23%(95% CI: 0.50,0.84) and 47.06%(95% CI: 0.34,0.61), respectively. CONCLUSION: T1CE-based radiomics showed better classification performance compared with radiologists' assessment.The radiomics modelwas promising in differentiating pseudoprogression from true progression.

Amide Proton Transfer Imaging in Predicting Isocitrate Dehydrogenase 1 Mutation Status of Grade II/III Gliomas Based on Support Vector Machine.

BACKGROUND: To compare the efficacies of univariate and radiomics analyses of amide proton transfer weighted (APTW) imaging in predicting isocitrate dehydrogenase 1 (IDH1) mutation of grade II/III gliomas. METHODS: Fifty-nine grade II/III glioma patients with known IDH1 mutation status were prospectively included (IDH1 wild type, 16; IDH1 mutation, 43). A total of 1044 quantitative radiomics features were extracted from APTW images. The efficacies of univariate and radiomics analyses in predicting IDH1 mutation were compared. Feature values were compared between two groups with independent t-test and receiver operating characteristic (ROC) analysis was applied to evaluate the predicting efficacy of each feature. Cases were randomly assigned to either the training (n = 49) or test cohort (n = 10) for the radiomics analysis. Support vector machine with recursive feature elimination (SVM-RFE) was adopted to select the optimal feature subset. The adverse impact of the imbalance dataset in the training cohort was solved by synthetic minority oversampling technique (SMOTE). Subsequently, the performance of SVM model was assessed on both training and test cohort. RESULTS: As for univariate analysis, 18 features were significantly different between IDH1 wild-type and mutant groups (P < 0.05). Among these parameters, High Gray Level Run Emphasis All Direction offset 8 SD achieved the biggest area under the curve (AUC) (0.769) with the accuracy of 0.799. As for radiomics analysis, SVM model was established using 19 features selected with SVM-RFE. The AUC and accuracy for IDH1 mutation on training set were 0.892 and 0.952, while on the testing set were 0.7 and 0.84, respectively. CONCLUSION: Radiomics strategy based on APT image features is potentially useful for preoperative estimating IDH1 mutation status.

Better efficacy in differentiating WHO grade II from III oligodendrogliomas with machine-learning than radiologist's reading from conventional T1 contrast-enhanced and fluid attenuated inversion recovery images.

BACKGROUND: The medical imaging to differentiate World Health Organization (WHO) grade II (ODG2) from III (ODG3) oligodendrogliomas still remains a challenge. We investigated whether combination of machine leaning with radiomics from conventional T1 contrast-enhanced (T1 CE) and fluid attenuated inversion recovery (FLAIR) magnetic resonance imaging (MRI) offered superior efficacy. METHODS: Thirty-six patients with histologically confirmed ODGs underwent T1 CE and 33 of them underwent FLAIR MR examination before any intervention from January 2015 to July 2017 were retrospectively recruited in the current study. The volume of interest (VOI) covering the whole tumor enhancement were manually drawn on the T1 CE and FLAIR slice by slice using ITK-SNAP and a total of 1072 features were extracted from the VOI using 3-D slicer software. Random forest (RF) algorithm was applied to differentiate ODG2 from ODG3 and the efficacy was tested with 5-fold cross validation. The diagnostic efficacy of radiomics-based machine learning and radiologist's assessment were also compared. RESULTS: Nineteen ODG2 and 17 ODG3 were included in this study and ODG3 tended to present with prominent necrosis and nodular/ring-like enhancement (P < 0.05). The AUC, ACC, sensitivity, and specificity of radiomics were 0.798, 0.735, 0.672, 0.789 for T1 CE, 0.774, 0.689, 0.700, 0.683 for FLAIR, as well as 0.861, 0.781, 0.778, 0.783 for the combination, respectively. The AUCs of radiologists 1, 2 and 3 were 0.700, 0.687, and 0.714, respectively. The efficacy of machine learning based on radiomics was superior to the radiologists' assessment. CONCLUSIONS: Machine-learning based on radiomics of T1 CE and FLAIR offered superior efficacy to that of radiologists in differentiating ODG2 from ODG3.

Perfusion, Diffusion, Or Brain Tumor Barrier Integrity: Which Represents The Glioma Features Best?

PURPOSE: This study aims to incorporate informative histogram indicator analyses and advanced multimodal MRI parameters to differentiate low-grade gliomas (LGGs) from high-grade gliomas (HGGs) and to explore the features associated with patients' survival. PATIENTS AND METHODS: A total of 120 patients with pathologically confirmed LGGs or HGGs receiving conventional and advanced MRI such as three-dimensional arterial spin labeling (3D-ASL), intravoxel incoherent motion-diffusion weighted imaging (IVIM-DWI), and dynamic contrast-enhanced MRI (DCE-MRI) were included. The mean and histogram indicators from advanced MRI were calculated from the entire tumor. The efficacies of a single indicator or multiple parameters were tested in distinguishing HGGs from LGGs and predicting patients' survival. Receiver operating characteristic (ROC) curve and multivariable stepwise logistic regression were used to evaluate the diagnostic efficacies. Leave-one-out cross-validation was further used to validate the accuracy of the parameter sets in glioma grading. Log-rank test using the Kaplan-Meier curve was utilized to predict patients' survival. RESULTS: Overall, parameters from DCE-MRI performed better than those from 3D-ASL or IVIM-DWI in both glioma grading and survival prediction. The histogram metrics of Ve were demonstrated to have higher accuracies (the accuracies for Extended Tofts_Ve mean and Extended Tofts_Ve median were 68.33% and 71.67%, respectively, while those for the Incremental_Ve mean and Incremental_Ve 75th were 68.33% and 72.50%, respectively) in grading LGGs from HGGs. The combination of Tofts_Ve histogram metrics was the one with the highest accuracy (81.67%) and area under ROC curve (AUC = 0.840). On the other hand, Patlak_Ktrans 95th (AUC = 0.9265) and Extended Tofts_Ve 95th (AUC = 0.9154) performed better than their corresponding means (Patlak_Ktrans mean: AUC = 0.9118 and Extended Tofts_Ve mean: AUC = 0.9044) in predicting patients' overall survival (OS) at 18-month follow-up. CONCLUSION: DCE-MRI-derived histogram features from the entire tumor were promising metrics for glioma grading and OS prediction. Combining single modal histogram features improved glioma grading. TRIAL REGISTRATION: NCT02622620.

Optimizing Texture Retrieving Model for Multimodal MR Image-Based Support Vector Machine for Classifying Glioma.

BACKGROUND: Accurate glioma grading plays an important role in patient treatment. PURPOSE: To investigate the influence of varied texture retrieving models on the efficacy of grading glioma with support vector machine (SVM). STUDY TYPE: Retrospective. POPULATION: In all, 117 glioma patients including 25, 29, and 63 grade II, III, and IV gliomas, respectively, based on WHO 2007. FIELD STRENGTH/SEQUENCE: 3.0T MRI/ T1 WI, T2 fluid-attenuated inversion recovery, contrast enhanced T1 , arterial spinal labeling, diffusion-weighted imaging (0, 30, 50, 100, 200, 300, 500, 800, 1000, 1500, 2000, 3000, and 3500 sec/mm2 ), and dynamic contrast-enhanced. ASSESSMENT: Texture attributes from 30 parametric maps were retrieved using four models, including Global, gray-level co-occurrence matrix (GLCM), gray-level run-length matrix (GLRLM), and gray-level size-zone matrix (GLSZM). Attributes derived from varied models were input into radial basis function SVM (RBF-SVM) combined with attribute selection using SVM-recursive feature elimination (SVM-RFE). The SVM model was trained and established with 80% randomly selected data of each category using 10-fold crossvalidation. The model performance was further tested using the remaining 20% data. STATISTICAL TESTS: Ten-fold crossvalidation was used to validate the model performance. RESULTS: Based on 30 parametric maps, 90, 240, 390, or 390 texture attributes were retrieved using the Global, GLCM, GLRLM, or GLSZM model, respectively. SVM-RFE was able to reduce attribute redundancy as well as improve RBF-SVM performance. Training data were oversampled by applying the Synthetic Minority Oversampling Technique (SMOTE) method to overcome the data imbalance problem; test results were able to further demonstrate the classifying performance of the final models. GLSZM using gray-level 64 was the optimal model to retrieve powerful image texture attributes to produce enough classifying power with an accuracy / area under the curve of 0.760/0.867 for the training and 0.875/0.971 for the independent test. Fifteen attributes were selected with SVM-RFE to provide comparable classifying efficacy. DATA CONCLUSION: When using image textures-based SVM classification of gliomas, the GLSZM model in combination with gray-level 64 and attribute selection may be an optimized solution. LEVEL OF EVIDENCE: 2 Technical Efficacy: Stage 2 J. Magn. Reson. Imaging 2019;49:1263-1274.

Radiomics strategy for glioma grading using texture features from multiparametric MRI.

BACKGROUND: Accurate glioma grading plays an important role in the clinical management of patients and is also the basis of molecular stratification nowadays. PURPOSE/HYPOTHESIS: To verify the superiority of radiomics features extracted from multiparametric MRI to glioma grading and evaluate the grading potential of different MRI sequences or parametric maps. STUDY TYPE: Retrospective; radiomics. POPULATION: A total of 153 patients including 42, 33, and 78 patients with Grades II, III, and IV gliomas, respectively. FIELD STRENGTH/SEQUENCE: 3.0T MRI/T1 -weighted images before and after contrast-enhanced, T2 -weighted, multi-b-value diffusion-weighted and 3D arterial spin labeling images. ASSESSMENT: After multiparametric MRI preprocessing, high-throughput features were derived from patients' volumes of interests (VOIs). The support vector machine-based recursive feature elimination was adopted to find the optimal features for low-grade glioma (LGG) vs. high-grade glioma (HGG), and Grade III vs. IV glioma classification tasks. Then support vector machine (SVM) classifiers were established using the optimal features. The accuracy and area under the curve (AUC) was used to assess the grading efficiency. STATISTICAL TESTS: Student's t-test or a chi-square test were applied on different clinical characteristics to confirm whether intergroup significant differences exist. RESULTS: Patients' ages between LGG and HGG groups were significantly different (P < 0.01). For each patient, 420 texture and 90 histogram parameters were derived from 10 VOIs of multiparametric MRI. SVM models were established using 30 and 28 optimal features for classifying LGGs from HGGs and grades III from IV, respectively. The accuracies/AUCs were 96.8%/0.987 for classifying LGGs from HGGs, and 98.1%/0.992 for classifying grades III from IV, which were more promising than using histogram parameters or using the single sequence MRI. DATA CONCLUSION: Texture features were more effective for noninvasively grading gliomas than histogram parameters. The combined application of multiparametric MRI provided a higher grading efficiency. The proposed radiomic strategy could facilitate clinical decision-making for patients with varied glioma grades. LEVEL OF EVIDENCE: 3 Technical Efficacy: Stage 2 J. Magn. Reson. Imaging 2018;48:1518-1528.

Structural and advanced imaging in predicting MGMT promoter methylation of primary glioblastoma: a region of interest based analysis.

BACKGROUND: The methylation status of oxygen 6-methylguanine-DNA methyltransferase (MGMT) promoter has been associated with treatment response in glioblastoma(GBM). Using pre-operative MRI techniques to predict MGMT promoter methylation status remains inconclusive. In this study, we investigated the value of features from structural and advanced imagings in predicting the methylation of MGMT promoter in primary glioblastoma patients. METHODS: Ninety-two pathologically confirmed primary glioblastoma patients underwent preoperative structural MR imagings and the efficacy of structural image features were qualitatively analyzed using Fisher's exact test. In addition, 77 of the 92 patients underwent additional advanced MRI scans including diffusion-weighted (DWI) and 3-diminsional pseudo-continuous arterial spin labeling (3D pCASL) imaging. Apparent diffusion coefficient (ADC) and relative cerebral blood flow (rCBF) values within the manually drawn region-of-interest (ROI) were calculated and compared using independent sample t test for their efficacies in predicting MGMT promoter methylation. Receiver operating characteristic curve (ROC) analysis was used to investigate the predicting efficacy with the area under the curve (AUC) and cross validations. Multiple-variable logistic regression model was employed to evaluate the predicting performance of multiple variables. RESULTS: MGMT promoter methylation was associated with tumor location and necrosis (P <  0.05). Significantly increased ADC value (P <  0.001) and decreased rCBF (P <  0.001) were associated with MGMT promoter methylation in primary glioblastoma. The ADC achieved the better predicting efficacy than rCBF (ADC: AUC, 0.860; sensitivity, 81.1%; specificity, 82.5%; vs rCBF: AUC, 0.835; sensitivity, 75.0%; specificity, 78.4%; P = 0.032). The combination of tumor location, necrosis, ADC and rCBF resulted in the highest AUC of 0.914. CONCLUSION: ADC and rCBF are promising imaging biomarkers in clinical routine to predict the MGMT promoter methylation in primary glioblastoma patients.

Optimizing a machine learning based glioma grading system using multi-parametric MRI histogram and texture features.

Current machine learning techniques provide the opportunity to develop noninvasive and automated glioma grading tools, by utilizing quantitative parameters derived from multi-modal magnetic resonance imaging (MRI) data. However, the efficacies of different machine learning methods in glioma grading have not been investigated.A comprehensive comparison of varied machine learning methods in differentiating low-grade gliomas (LGGs) and high-grade gliomas (HGGs) as well as WHO grade II, III and IV gliomas based on multi-parametric MRI images was proposed in the current study. The parametric histogram and image texture attributes of 120 glioma patients were extracted from the perfusion, diffusion and permeability parametric maps of preoperative MRI. Then, 25 commonly used machine learning classifiers combined with 8 independent attribute selection methods were applied and evaluated using leave-one-out cross validation (LOOCV) strategy. Besides, the influences of parameter selection on the classifying performances were investigated. We found that support vector machine (SVM) exhibited superior performance to other classifiers. By combining all tumor attributes with synthetic minority over-sampling technique (SMOTE), the highest classifying accuracy of 0.945 or 0.961 for LGG and HGG or grade II, III and IV gliomas was achieved. Application of Recursive Feature Elimination (RFE) attribute selection strategy further improved the classifying accuracies. Besides, the performances of LibSVM, SMO, IBk classifiers were influenced by some key parameters such as kernel type, c, gama, K, etc. SVM is a promising tool in developing automated preoperative glioma grading system, especially when being combined with RFE strategy. Model parameters should be considered in glioma grading model optimization.

Combination of IVIM-DWI and 3D-ASL for differentiating true progression from pseudoprogression of Glioblastoma multiforme after concurrent chemoradiotherapy: study protocol of a prospective diagnostic trial.

BACKGROUND: Standard therapy for Glioblastoma multiforme (GBM) involves maximal safe tumor resection followed with radiotherapy and concurrent adjuvant temozolomide. About 20 to 30% patients undergoing their first post-radiation MRI show increased contrast enhancement which eventually recovers without any new treatment. This phenomenon is referred to as pseudoprogression. Differentiating tumor progression from pseudoprogression is critical for determining tumor treatment, yet this capacity remains a challenge for conventional magnetic resonance imaging (MRI). Thus, a prospective diagnostic trial has been established that utilizes multimodal MRI techniques to detect tumor progression at its early stage. The purpose of this trial is to explore the potential role of intravoxel incoherent motion diffusion-weighted imaging (IVIM-DWI) and three-dimensional arterial spin labeling imaging (3D-ASL) in differentiating true progression from pseudoprogression of GBM. In addition, the diagnostic performance of quantitative parameters obtained from IVIM-DWI and 3D-ASL, including apparent diffusion coefficient (ADC), slow diffusion coefficient (D), fast diffusion coefficient (D*), perfusion fraction (f), and cerebral blood flow (CBF), will be evaluated. METHODS: Patients that recently received a histopathological diagnosis of GBM at our hospital are eligible for enrollment. The patients selected will receive standard concurrent chemoradiotherapy and adjuvant temozolomide after surgery, and then will undergo conventional MRI, IVIM-DWI, 3D-ASL, and contrast-enhanced MRI. The quantitative parameters, ADC, D, D*, f, and CBF, will be estimated for newly developed enhanced lesions. Further comparisons will be made with unpaired t-tests to evaluate parameter performance in differentiating true progression from pseudoprogression, while receiver-operating characteristic (ROC) analyses will determine the optimal thresholds, as well as sensitivity and specificity. Finally, relationships between these parameters will be assessed with Pearson's correlation and partial correlation analyses. DISCUSSION: The results of this study may demonstrate the potential value of using multimodal MRI techniques to differentiate true progression from pseudoprogression in its early stages to help decision making in early intervention and improve the prognosis of GBM. TRIAL REGISTRATION: This study has been registered at ClinicalTrials.gov ( NCT02622620 ) on November 18, 2015 and published on March 28, 2016.

Comparison between ultra-high and conventional mono b-value DWI for preoperative glioma grading.

To compare the efficacy of ultra-high and conventional mono-b-value DWI for glioma grading, in 109 pathologically confirmed glioma patients, ultra-high apparent diffusion coefficient (ADCuh)was calculated using a tri-exponential mode, distributed diffusion coefficients (DDCs) and α values were calculated using a stretched-exponential model, and conventional ADC values were calculated using a mono-exponential model. The efficacy and reliability of parameters for grading gliomas were investigated using receiver operating characteristic (ROC) curve and intra-class correlation (ICC) analyses, respectively. The ADCuh values differed (P < 0.001) between low-grade gliomas (LGGs; 0.436 ×10-3 mm2/sec) and high-grade gliomas (HGGs; 0.285 × 10-3 mm2/sec). DDC, a and various conventional ADC values were smaller in HGGs (all P ≤ 0.001, vs. LGGs). The ADCuh parameter achieved the highest diagnostic efficacy with an area under curve (AUC) of 0.993, 92.9% sensitivity and 98.8% specificity for glioma grading at a cutoff value of 0.362×10-3 mm2/sec. ADCuh measurement appears to be an easy-to-perform technique with good reproducibility (ICC = 0.9391, P < 0.001). The ADCuh value based in a tri-exponential model exhibited greater efficacy and reliability than other DWI parameters, making it a promising technique for glioma grading.