RESUMEN
BACKGROUND: Hemangioblastoma (HB) is a highly vascularized tumor most commonly occurring in the posterior cranial fossa, requiring accurate preoperative diagnosis to avoid accidental intraoperative hemorrhage and even death. PURPOSE: To accurately distinguish HBs from other cerebellar-and-brainstem tumors using a convolutional neural network model based on a contrast-enhanced brain MRI dataset. STUDY TYPE: Retrospective. POPULATION: Four hundred five patients (182 = HBs; 223 = other cerebellar-and brainstem tumors): 305 cases for model training, and 100 for evaluation. FIELD STRENGTH/SEQUENCE: 3 T/contrast-enhanced T1-weighted imaging (T1WI + C). ASSESSMENT: A CNN-based 2D classification network was trained by using sliced data along the z-axis. To improve the performance of the network, we introduced demographic information, various data-augmentation methods and an auxiliary task to segment tumor region. Then, this method was compared with the evaluations performed by experienced and intermediate-level neuroradiologists, and the heatmap of deep feature, which indicates the contribution of each pixel to model prediction, was visualized by Grad-CAM for analyzing the misclassified cases. STATISTICAL TESTS: The Pearson chi-square test and an independent t-test were used to test for distribution difference in age and sex. And the independent t-test was exploited to evaluate the performance between experts and our proposed method. P value <0.05 was considered significant. RESULTS: The trained network showed a higher accuracy for identifying HBs (accuracy = 0.902 ± 0.031, F1 = 0.891 ± 0.035, AUC = 0.926 ± 0.040) than experienced (accuracy = 0.887 ± 0.013, F1 = 0.868 ± 0.011, AUC = 0.881 ± 0.008) and intermediate-level (accuracy = 0.827 ± 0.037, F1 = 0.768 ± 0.068, AUC = 0.810 ± 0.047) neuroradiologists. The recall values were 0.910 ± 0.050, 0.659 ± 0.084, and 0.828 ± 0.019 for the trained network, intermediate and experienced neuroradiologists, respectively. Additional ablation experiments verified the utility of the introduced demographic information, data augmentation, and the auxiliary-segmentation task. DATA CONCLUSION: Our proposed method can successfully distinguish HBs from other cerebellar-and-brainstem tumors and showed diagnostic efficiency comparable to that of experienced neuroradiologists. EVIDENCE LEVEL: 3 TECHNICAL EFFICACY: Stage 2.
RESUMEN
BACKGROUND: Measuring non-parametric intravoxel mean diffusivity distributions (MDDs) using magnetic resonance imaging (MRI) is a sensitive method for detecting intracellular diffusivity changes during physiological alterations. Histological and molecular glioma classifications are essential for prognosis and treatment, with distinct water diffusion dynamics among subtypes. PURPOSE: We developed a data-driven approach using a fully connected network (FCN) to enhance the speed and stability of calculating MDDs across varying SNRs, enable tumor microstructural mapping, and test its reliability in identifying MIB-1 labeling index (LI) levels and molecular status of gliomas. METHODS: An FCN was trained to learn the mapping between the simulated diffusion decay curves and the ground truth MDDs. We performed 5 000 000 simulation curves with various diffusivity components and random SNR ∈ [ 30 , 300 ] $ \in [ {30,\ 300} ]$ . Eighty percent of simulation curves were used for the FCN training, 10% for validation, and the others were external tests for the FCN performance evaluation. In vivo data were collected to evaluate its clinical reliability. One hundred one patients (44 years ± $ \pm $ 14, 67 men) with gliomas and six healthy controls underwent a 3.0 T MRI examination with a spin echo-echo planar imaging (SE-EPI) diffusion-weighted imaging (DWI) sequence. The trained FCN was employed to calculate MDDs of each brain voxel by voxel. We used the Fuzzy C-means algorithm to cluster the MDDs of tumor voxels, facilitating the characterization of distinct glioma tissues. Quantitative assessments were conducted through sectional integrals of the MDDs, demarcated by six bands to derive signal fractions ( f n , n = 1 - 6 ${{f}_n},\ n = 1 -6$ ) and diffusivities of the maximum peaks ( D p e a k ${{D}_{peak}}$ ). Cosine similarity scores (CSS) were used for MDD similarity. ANOVA and Mann-Whitney U test were used for difference analysis. Logistic regression and area under the receiver operator characteristic curve (AUC) were used for classification evaluation. RESULTS: The simulation results showed that the FCN-based MDD approach (FCN-MDD) achieved higher CSS than non-negative least squares-based MDD (NNLS-MDD). For in vivo data, the spectra of ET and NET obtained by FCN-MDD are more distinguishable than NNLS-MDD. Fraction maps delineate the characteristics of different tumor tissues (enhancing and non-enhancing tumor, edema, and necrosis). f 3 , f 4 , D p e a k ${{f}_3},\ {{f}_4},{{D}_{peak}}$ showed a positive and negative correlation with MIB-1 respectively ( r = 0.568 , r = - 0.521 , r = - 0.654 $r = 0.568,\ r = - 0.521,\ r = - 0.654$ , all p < 0.001 $p < 0.001$ ). The AUC of D p e a k ${{D}_{peak}}$ for predicting MIB-1 LI levels was 0.900 (95% CI, 0.826-0.974), versus 0.781 (0.677-0.886) of ADC. The highest AUC of isocitrate dehydrogenase (IDH) mutation status, assessed by a logistic regression model ( f 1 + f 3 ${{f}_1} + {{f}_3}$ ) was 0.873 (95% CI, 0.802-0.944). CONCLUSION: The proposed FCN-MDD method was more robust to variations in SNR and less reliant on empirically set regularization values than the NNLS-MDD method. FCN-MDD also enabled qualitative and quantitative evaluation of the composition of gliomas.