Machine Learning Tools for Image-Based Glioma Grading and the Quality of Their Reporting: Challenges and Opportunities

Machine Learning Algorithms for Neurosurgical Preoperative Planning: A Scoping Review

BACKGROUND AND OBJECTIVE

Preoperative neurosurgical planning is an important step in avoiding surgical complications, reducing morbidity, and improving patient safety. The incursion of machine learning (ML) in this domain has recently gained attention, given the notable advantages in processing large datasets and potentially generating efficient and accurate algorithms in patient care. We explored the evolving applications of ML algorithms in the preoperative planning of brain and spine surgery.

METHODS

In accordance with the Arksey and O'Malley framework, a scoping review was conducted using 3 databases (PubMed, Embase, and Web of Science). Articles that described the use of ML for preoperative planning in brain and spine surgery were included. Relevant data were collected regarding the neurosurgical field of application, patient baseline features, disease description, type of ML technology, study's aim, preoperative ML algorithm description, and advantages and limitations of ML algorithms.

RESULTS

Our search strategy yielded 7407 articles, of which 8 studies (5 retrospective, 2 prospective, and 1 experimental) satisfied the inclusion criteria. Clinical information from 518 patients (62.7% female; mean age: 44.8 years) was used for generating ML algorithms, including convolutional neural networks (14.3%), logistic regression (14.3%), and random forest (14.3%), among others. Neurosurgical fields of applications included functional neurosurgery (37.5%), tumor surgery (37.5%), and spine surgery (25%). The main advantages of ML included automated processing of clinical and imaging information, selection of an individualized patient surgical approach, and data-driven support for treatment decision-making. All studies reported technical limitations, such as long processing time, algorithmic bias, limited generalizability, and the need for database updating and maintenance.

CONCLUSIONS

ML algorithms for preoperative neurosurgical planning are being developed for efficient, automated, and safe treatment decision-making. However, future studies are necessary to validate their objective performance across diverse clinical scenarios. Enhancing the robustness, transparency, and understanding of ML applications will be crucial for their successful integration into neurosurgical practice.

Auto-segmentation of Adult-Type Diffuse Gliomas: Comparison of Transfer Learning-Based Convolutional Neural Network Model vs. Radiologists

Segmentation of glioma is crucial for quantitative brain tumor assessment, to guide therapeutic research and clinical management, but very time-consuming. Fully automated tools for the segmentation of multi-sequence MRI are needed. We developed and pretrained a deep learning (DL) model using publicly available datasets A (n = 210) and B (n = 369) containing FLAIR, T2WI, and contrast-enhanced (CE)-T1WI. This was then fine-tuned with our institutional dataset (n = 197) containing ADC, T2WI, and CE-T1WI, manually annotated by radiologists, and split into training (n = 100) and testing (n = 97) sets. The Dice similarity coefficient (DSC) was used to compare model outputs and manual labels. A third independent radiologist assessed segmentation quality on a semi-quantitative 5-scale score. Differences in DSC between new and recurrent gliomas, and between uni or multifocal gliomas were analyzed using the Mann-Whitney test. Semi-quantitative analyses were compared using the chi-square test. We found that there was good agreement between segmentations from the fine-tuned DL model and ground truth manual segmentations (median DSC: 0.729, std-dev: 0.134). DSC was higher for newly diagnosed (0.807) than recurrent (0.698) (p < 0.001), and higher for unifocal (0.747) than multi-focal (0.613) cases (p = 0.001). Semi-quantitative scores of DL and manual segmentation were not significantly different (mean: 3.567 vs. 3.639; 93.8% vs. 97.9% scoring ≥ 3, p = 0.107). In conclusion, the proposed transfer learning DL performed similarly to human radiologists in glioma segmentation on both structural and ADC sequences. Further improvement in segmenting challenging postoperative and multifocal glioma cases is needed.

A multi-institutional pediatric dataset of clinical radiology MRIs by the Children's Brain Tumor Network

Pediatric brain and spinal cancers remain the leading cause of cancer-related death in children. Advancements in clinical decision-support in pediatric neuro-oncology utilizing the wealth of radiology imaging data collected through standard care, however, has significantly lagged other domains. Such data is ripe for use with predictive analytics such as artificial intelligence (AI) methods, which require large datasets. To address this unmet need, we provide a multi-institutional, large-scale pediatric dataset of 23,101 multi-parametric MRI exams acquired through routine care for 1,526 brain tumor patients, as part of the Children's Brain Tumor Network. This includes longitudinal MRIs across various cancer diagnoses, with associated patient-level clinical information, digital pathology slides, as well as tissue genotype and omics data. To facilitate downstream analysis, treatment-naïve images for 370 subjects were processed and released through the NCI Childhood Cancer Data Initiative via the Cancer Data Service. Through ongoing efforts to continuously build these imaging repositories, our aim is to accelerate discovery and translational AI models with real-world data, to ultimately empower precision medicine for children.

CSF-Glioma: A Causal Segmentation Framework for Accurate Grading and Subregion Identification of Gliomas

Deep networks have shown strong performance in glioma grading; however, interpreting their decisions remains challenging due to glioma heterogeneity. To address these challenges, the proposed solution is the Causal Segmentation Framework (CSF). This framework aims to accurately predict high- and low-grade gliomas while simultaneously highlighting key subregions. Our framework utilizes a shrinkage segmentation method to identify subregions containing essential decision information. Moreover, we introduce a glioma grading module that combines deep learning and traditional approaches for precise grading. Our proposed model achieves the best performance among all models, with an AUC of 96.14%, an F1 score of 93.74%, an accuracy of 91.04%, a sensitivity of 91.83%, and a specificity of 88.88%. Additionally, our model exhibits efficient resource utilization, completing predictions within 2.31s and occupying only 0.12 GB of memory during the test phase. Furthermore, our approach provides clear and specific visualizations of key subregions, surpassing other methods in terms of interpretability. In conclusion, the Causal Segmentation Framework (CSF) demonstrates its effectiveness at accurately predicting glioma grades and identifying key subregions. The inclusion of causality in the CSF model enhances the reliability and accuracy of preoperative decision-making for gliomas. The interpretable results provided by the CSF model can assist clinicians in their assessment and treatment planning.

Performance of deep learning algorithms to distinguish high-grade glioma from low-grade glioma: A systematic review and meta-analysis

This study aims to evaluate deep learning (DL) performance in differentiating low- and high-grade glioma. Search online database for studies continuously published from 1st January 2015 until 16th August 2022. The random-effects model was used for synthesis, based on pooled sensitivity (SE), specificity (SP), and area under the curve (AUC). Heterogeneity was estimated using the Higgins inconsistency index (I²). 33 were ultimately included in the meta-analysis. The overall pooled SE and SP were 94% and 93%, with an AUC of 0.98. There was great heterogeneity in this field. Our evidence-based study shows DL achieves high accuracy in glioma grading. Subgroup analysis reveals several limitations in this field: 1) Diagnostic trials require standard method for data merging for AI; 2) small sample size; 3) poor-quality image preprocessing; 4) not standard algorithm development; 5) not standard data report; 6) different definition of HGG and LGG; and 7) poor extrapolation.

Machine-Learning-Based Radiomics for Classifying Glioma Grade from Magnetic Resonance Images of the Brain

Grading of gliomas is a piece of critical information related to prognosis and survival. Classifying glioma grade by semantic radiological features is subjective, requires multiple MRI sequences, is quite complex and clinically demanding, and can very often result in erroneous radiological diagnosis. We used a radiomics approach with machine learning classifiers to determine the grade of gliomas. Eighty-three patients with histopathologically proven gliomas underwent MRI of the brain. Whenever available, immunohistochemistry was additionally used to augment the histopathological diagnosis. Segmentation was performed manually on the T2W MR sequence using the TexRad texture analysis softwareTM, Version 3.10. Forty-two radiomics features, which included first-order features and shape features, were derived and compared between high-grade and low-grade gliomas. Features were selected by recursive feature elimination using a random forest algorithm method. The classification performance of the models was measured using accuracy, precision, recall, f1 score, and area under the curve (AUC) of the receiver operating characteristic curve. A 10-fold cross-validation was adopted to separate the training and the test data. The selected features were used to build five classifier models: support vector machine, random forest, gradient boost, naive Bayes, and AdaBoost classifiers. The random forest model performed the best, achieving an AUC of 0.81, an accuracy of 0.83, f1 score of 0.88, a recall of 0.93, and a precision of 0.85 for the test cohort. The results suggest that machine-learning-based radiomics features extracted from multiparametric MRI images can provide a non-invasive method for predicting glioma grades preoperatively. In the present study, we extracted the radiomics features from a single cross-sectional image of the T2W MRI sequence and utilized these features to build a fairly robust model to classify low-grade gliomas from high-grade gliomas (grade 4 gliomas).

Vasari Scoring System in Discerning between Different Degrees of Glioma and IDH Status Prediction: A Possible Machine Learning Application?

(1) The aim of our study is to evaluate the capacity of the Visually AcceSAble Rembrandt Images (VASARI) scoring system in discerning between the different degrees of glioma and Isocitrate Dehydrogenase (IDH) status predictions, with a possible application in machine learning. (2) A retrospective study was conducted on 126 patients with gliomas (M/F = 75/51; mean age: 55.30), from which we obtained their histological grade and molecular status. Each patient was analyzed with all 25 features of VASARI, blinded by two residents and three neuroradiologists. The interobserver agreement was assessed. A statistical analysis was conducted to evaluate the distribution of the observations using a box plot and a bar plot. We then performed univariate and multivariate logistic regressions and a Wald test. We also calculated the odds ratios and confidence intervals for each variable and the evaluation matrices with receiver operating characteristic (ROC) curves in order to identify cut-off values that are predictive of a diagnosis. Finally, we did the Pearson correlation test to see if the variables grade and IDH were correlated. (3) An excellent ICC estimate was obtained. For the grade and IDH status prediction, there were statistically significant results by evaluation of the degree of post-contrast impregnation (F4) and the percentage of impregnated area (F5), not impregnated area (F6), and necrotic (F7) tissue. These models showed good performances according to the area under the curve (AUC) values (>70%). (4) Specific MRI features can be used to predict the grade and IDH status of gliomas, with important prognostic implications. The standardization and improvement of these data (aim: AUC > 80%) can be used for programming machine learning software.

Cathepsin B, D and S as Potential Biomarkers of Brain Glioma Malignancy

Brain gliomas constitute the vast majority of malignant tumors of the nervous system. There is still a lack of fast, reliable and non-invasive methods of diagnostics. Our work focuses on the quantification of cathepsin B, D and S in glioma. The research was conducted with the use of SPRi biosensors sensitive to individual cathepsins. Changes in the quantity of selected cathepsins (cathepsins B, D and S), depending on the advancement of glioma and the presence or absence of important features or comorbidities in the selected patient, were examined. The results were statistically analyzed and interpreted based on the available clinical description. Statistical significance was observed in the difference in the concentration of the studied cathepsins, mainly between the groups Control and G3/G4 and G1/G2 and G3/G4. The strength of the correlation between the concentrations of individual cathepsins and the age of the patient and the size of the tumor, as well as the correlation between individual proteins, was investigated. The influence of IDH 1/2 status on the concentration of determined cathepsins was investigated and ROC analysis was performed. As a result of our research, we have developed a method for the diagnosis of brain glioma that allows us to distinguish grades G1/G2 from G3/G4 and the control group from G3/G4. We found an average positive correlation between the concentrations of the proteins tested and the age of the patient and a high positive correlation between the cathepsins tested. Comparative analysis of the effect of the presence of IDH 1/2 mutations on the number of proteins tested allowed us to demonstrate that the cathepsins assayed can be independent markers.