We sought to evaluate the utility of radiomics for Amide Proton Transfer weighted (APTw) imaging by assessing its value in differentiating brain metastases from high- and low grade glial brain tumors. We retrospectively identified 48 treatment-naïve patients (10 WHO grade 2, 1 WHO grade 3, 10 WHO grade 4 primary glial brain tumors and 27 metastases) with either primary glial brain tumors or metastases who had undergone APTw MR imaging. After image analysis with radiomics feature extraction and post-processing, machine learning algorithms (multilayer perceptron machine learning algorithm; random forest classifier) with stratified tenfold cross validation were trained on features and were used to differentiate the brain neoplasms. The multilayer perceptron achieved an AUC of 0.836 (receiver operating characteristic curve) in differentiating primary glial brain tumors from metastases. The random forest classifier achieved an AUC of 0.868 in differentiating WHO grade 4 from WHO grade 2/3 primary glial brain tumors. For the differentiation of WHO grade 4 tumors from grade 2/3 tumors and metastases an average AUC of 0.797 was achieved. Our results indicate that the use of radiomics for APTw imaging is feasible and the differentiation of primary glial brain tumors from metastases is achievable with a high degree of accuracy.

Treatment planning and prognosis in glioma treatment are based on the classification into low- and high-grade oligodendroglioma or astrocytoma, which is mainly based on molecular characteristics (IDH1/2- and 1p/19q codeletion status). It would be of great value if this classification could be made reliably before surgery, without biopsy. Machine learning algorithms (MLAs) could play a role in achieving this by enabling glioma characterization on magnetic resonance imaging (MRI) data without invasive tissue sampling. The aim of this study is to provide a performance evaluation and meta-analysis of various MLAs for glioma characterization. Systematic literature search and meta-analysis were performed on the aggregated data, after which subgroup analyses for several target conditions were conducted. This study is registered with PROSPERO, CRD42020191033. We identified 724 studies; 60 and 17 studies were eligible to be included in the systematic review and meta-analysis, respectively. Meta-analysis showed excellent accuracy for all subgroups, with the classification of 1p/19q codeletion status scoring significantly poorer than other subgroups (AUC: 0.748, p = 0.132). There was considerable heterogeneity among some of the included studies. Although promising results were found with regard to the ability of MLA-tools to be used for the non-invasive classification of gliomas, large-scale, prospective trials with external validation are warranted in the future.

To investigate the applicability and performance of automated machine learning (AutoML) for potential applications in diagnostic neuroradiology. In the medical sector, there is a rapidly growing demand for machine learning methods, but only a limited number of corresponding experts. The comparatively simple handling of AutoML should enable even non-experts to develop adequate machine learning models with manageable effort. We aim to investigate the feasibility as well as the advantages and disadvantages of developing AutoML models compared to developing conventional machine learning models. We discuss the results in relation to a concrete example of a medical prediction application. In this retrospective IRB-approved study, a cohort of 107 patients who underwent gross total meningioma resection and a second cohort of 31 patients who underwent subtotal resection were included. Image segmentation of the contrast enhancing parts of the tumor was performed semi-automatically using the open-source software platform 3D Slicer. A total of 107 radiomic features were extracted by hand-delineated regions of interest from the pre-treatment MRI images of each patient. Within the AutoML approach, 20 different machine learning algorithms were trained and tested simultaneously. For comparison, a neural network and different conventional machine learning algorithms were trained and tested. With respect to the exemplary medical prediction application used in this study to evaluate the performance of Auto ML, namely the pre-treatment prediction of the achievable resection status of meningioma, AutoML achieved remarkable performance nearly equivalent to that of a feed-forward neural network with a single hidden layer. However, in the clinical case study considered here, logistic regression outperformed the AutoML algorithm. Using independent test data, we observed the following classification results (AutoML/neural network/logistic regression): mean area under the curve = 0.849/0.879/0.900, mean accuracy = 0.821/0.839/0.881, mean kappa = 0.465/0.491/0.644, mean sensitivity = 0.578/0.577/0.692 and mean specificity = 0.891/0.914/0.936. The results obtained with AutoML are therefore very promising. However, the AutoML models in our study did not yet show the corresponding performance of the best models obtained with conventional machine learning methods. While AutoML may facilitate and simplify the task of training and testing machine learning algorithms as applied in the field of neuroradiology and medical imaging, a considerable amount of expert knowledge may still be needed to develop models with the highest possible discriminatory power for diagnostic neuroradiology.

Our aim is to predict possible gross total and subtotal resections of skull meningiomas from pre-treatment T1 post contrast MR-images using radiomics and machine learning in a representative patient cohort. We analyse the accuracy of our model predictions depending on the tumor location within the skull and the postoperative tumor volume. In this retrospective, IRB-approved study, image segmentation of the contrast enhancing parts of the tumor was semi-automatically performed using the 3D Slicer open-source software platform. Imaging data were split into training data and independent test data at random. We extracted a total of 107 radiomic features by hand-delineated regions of interest on T1 post contrast MR images. Feature preselection and model construction were performed with eight different machine learning algorithms. Each model was estimated 100 times on new training data and then tested on a previously unknown, independent test data set to avoid possible overfitting. Our cohort included 138 patients. A gross total resection of the meningioma was performed in 107 cases and a subtotal resection in the remaining 31 cases. Using the training data, the mean area under the curve (AUC), mean accuracy, mean kappa, mean sensitivity and mean specificity were 0.901, 0.875, 0.629, 0.675 and 0.933 respectively. We obtained very similar results with the independent test data: mean AUC = 0.900, mean accuracy = 0.881, mean kappa = 0.644, mean sensitivity = 0.692 and mean specificity = 0.936. Thus, our model exposes good and stable predictive performance with both training and test data. Our radiomics approach shows that with machine learning algorithms and comparatively few explanatory factors such as the location of the tumor within the skull as well as its shape, it is possible to make accurate predictions about whether a meningioma can be completely resected by surgery. Complete resections and resections with larger postoperative tumor volumes can be predicted with very high accuracy. However, cases with very small postoperative tumor volumes are comparatively difficult to predict correctly.

To compare the diagnostic performance of texture analysis (TA) against visual qualitative assessment in the differentiation of spondyloarthritis (SpA) from degenerative changes in the sacroiliac joints (SIJ). Ninety patients referred for suspected inflammatory lower back pain from the rheumatology department were retrospectively included at our university hospital institution. MRI at 3 T of the lumbar spine and SIJ was performed with oblique coronal T1-weighted (w), fluid-sensitive fat-saturated (fs) TIRM and fsT1w intravenously contrast-enhanced (CE) images. Subjects were divided into three age- and gender-matched groups (30 each) based on definite clinical diagnosis serving as clinical reference standard with either degenerative, inflammatory (SpA) or no changes of the SIJ. SIJ were rated qualitatively by two independent radiologists and quantitatively by region-of-interest-based TA with 304 features subjected to machine learning logistic regression with randomized ten-fold selection of training and validation data. Qualitative and quantitative results were evaluated for diagnostic performance and compared against clinical reference standard. Agreement of radiologist's diagnose with clinical reference was fair for both readers (κ = 0.32 and 0.44). ROC statistics revealed significant outperformance of TA compared to qualitative ratings for differentiation of SpA from remainder (AUC = 0.89 vs. 0.75), SpA from degenerative (AUC = 0.91 vs. 0.67) and TIRM-positive SpA (i.e. with bone marrow edema) from remainder cases (AUC = 0.95 vs. 0.76). T1w-CE images were the most important discriminator for detection of SpA. TA is superior to qualitative assessment for the differentiation of inflammatory from degenerative changes of the SIJ. Intravenous CE-images increase diagnostic yield in quantitative TA.