Intensity standardization of MRI prior to radiomic feature extraction for artificial intelligence research in glioma-a systematic review

Quality assessment of critical and non-critical domains of systematic reviews on artificial intelligence in gliomas using AMSTAR II: A systematic review

INTRODUCTION

Gliomas are the most common primary malignant intraparenchymal brain tumors with a dismal prognosis. With growing advances in artificial intelligence, machine learning and deep learning models are being utilized for preoperative, intraoperative and postoperative neurological decision-making. We aimed to compile published literature in one format and evaluate the quality of level 1a evidence currently available.

METHODOLOGY

Using PRISMA guidelines, a comprehensive literature search was conducted within databases including Medline, Scopus, and Cochrane Library, and records with the application of artificial intelligence in glioma management were included. The AMSTAR 2 tool was used to assess the quality of systematic reviews and meta-analyses by two independent researchers.

RESULTS

From 812 studies, 23 studies were included. AMSTAR II appraised most reviews as either low or critically low in quality. Most reviews failed to deliver in critical domains related to the exclusion of studies, appropriateness of meta-analytical methods, and assessment of publication bias. Similarly, compliance was lowest in non-critical areas related to study design selection and the disclosure of funding sources in individual records. Evidence is moderate to low in quality in reviews on multiple neuro-oncological applications, low quality in glioma diagnosis and individual molecular markers like MGMT promoter methylation status, IDH, and 1p19q identification, and critically low in tumor segmentation, glioma grading, and multiple molecular markers identification.

CONCLUSION

AMSTAR 2 is a robust tool to identify high-quality systematic reviews. There is a paucity of high-quality systematic reviews on the utility of artificial intelligence in glioma management, with some demonstrating critically low quality. Therefore, caution must be exercised when drawing inferences from these results.

Artificial intelligence and advanced MRI techniques: A comprehensive analysis of diffuse gliomas

INTRODUCTION

The complexity of diffuse gliomas relies on advanced imaging techniques like MRI to understand their heterogeneity. Utilizing the UCSF-PDGM dataset, this study harnesses MRI techniques, radiomics, and AI to analyze diffuse gliomas for optimizing patient outcomes.

METHODS

The research utilized the dataset of 501 subjects with diffuse gliomas through a comprehensive MRI protocol. After performing intricate tumor segmentation, 82.800 radiomic features were extracted for each patient from nine segmentations across eight MRI sequences. These features informed neural network and XGBoost model training to predict patient outcomes and tumor grades, supplemented by SHAP analysis to pinpoint influential radiomic features.

RESULTS

In our analysis of the UCSF-PDGM dataset, we observed a diverse range of WHO tumor grades and patient outcomes, discarding one corrupt MRI scan. Our segmentation method showed high accuracy when comparing automated and manual techniques. The neural network excelled in prediction of WHO tumor grades with an accuracy of 0.9500 for the necrotic tumor label. The SHAP-analysis highlighted the 3D First Order mean as one of the most influential radiomic features, with features like Original Shape Sphericity and Original Shape Elongation were notably prominent.

CONCLUSION

A study using the UCSF-PDGM dataset highlighted AI and radiomics' profound impact on neuroradiology by demonstrating reliable tumor segmentation and identifying key radiomic features, despite challenges in predicting patient survival. The research emphasizes both the potential of AI in this field and the need for broader datasets of diverse MRI sequences to enhance patient outcomes.

IMPLICATION FOR PRACTICE

The study underline the significant role of radiomics in improving the accuracy of tumor identification through radiomic features.

Impact of intensity standardisation and ComBat batch size on clinical-radiomic prognostic models performance in a multi-centre study of patients with glioblastoma

PURPOSE

To assess the effect of different intensity standardisation techniques (ISTs) and ComBat batch sizes on radiomics survival model performance and stability in a heterogenous, multi-centre cohort of patients with glioblastoma (GBM).

METHODS

Multi-centre pre-operative MRI acquired between 2014 and 2020 in patients with IDH-wildtype unifocal WHO grade 4 GBM were retrospectively evaluated. WhiteStripe (WS), Nyul histogram matching (HM), and Z-score (ZS) ISTs were applied before radiomic feature (RF) extraction. RFs were realigned using ComBat and minimum batch size (MBS) of 5, 10, or 15 patients. Cox proportional hazards models for overall survival (OS) prediction were produced using five different selection strategies and the impact of IST and MBS was evaluated using bootstrapping. Calibration, discrimination, relative explained variation, and model fit were assessed. Instability was evaluated using 95% confidence intervals (95% CIs), feature selection frequency and calibration curves across the bootstrap resamples.

RESULTS

One hundred ninety-five patients were included. Median OS = 13 (95% CI: 12-14) months. Twelve to fourteen unique MRI protocols were used per MRI sequence. HM and WS produced the highest relative increase in model discrimination, explained variation and model fit but IST choice did not greatly impact on stability, nor calibration. Larger ComBat batches improved discrimination, model fit, and explained variation but higher MBS (reduced sample size) reduced stability (across all performance metrics) and reduced calibration accuracy.

CONCLUSION

Heterogenous, real-world GBM data poses a challenge to the reproducibility of radiomics. ComBat generally improved model performance as MBS increased but reduced stability and calibration. HM and WS tended to improve model performance.

KEY POINTS

Question ComBat harmonisation of RFs and intensity standardisation of MRI have not been thoroughly evaluated in multicentre, heterogeneous GBM data. Findings The addition of ComBat and ISTs can improve discrimination, relative model fit, and explained variance but degrades the calibration and stability of survival models. Clinical relevance Radiomics risk prediction models in real-world, multicentre contexts could be improved by ComBat and ISTs, however, this degrades calibration and prediction stability and this must be thoroughly investigated before patients can be accurately separated into different risk groups.

Deep Learning for MRI Segmentation and Molecular Subtyping in Glioblastoma: Critical Aspects from an Emerging Field

Deep learning (DL) has been applied to glioblastoma (GBM) magnetic resonance imaging (MRI) assessment for tumor segmentation and inference of molecular, diagnostic, and prognostic information. We comprehensively overviewed the currently available DL applications, critically examining the limitations that hinder their broader adoption in clinical practice and molecular research. Technical limitations to the routine application of DL include the qualitative heterogeneity of MRI, related to different machinery and protocols, and the absence of informative sequences, possibly compensated by artificial image synthesis. Moreover, taking advantage from the available benchmarks of MRI, algorithms should be trained on large amounts of data. Additionally, the segmentation of postoperative imaging should be further addressed to limit the inaccuracies previously observed for this task. Indeed, molecular information has been promisingly integrated in the most recent DL tools, providing useful prognostic and therapeutic information. Finally, ethical concerns should be carefully addressed and standardized to allow for data protection. DL has provided reliable results for GBM assessment concerning MRI analysis and segmentation, but the routine clinical application is still limited. The current limitations could be prospectively addressed, giving particular attention to data collection, introducing new technical advancements, and carefully regulating ethical issues.

Whole-lesion assessment of volume and signal changes after sclerotherapy of extremity venous malformations

PURPOSE

To investigate quantitative changes in MRI signal intensity (SI) and lesion volume that indicate treatment response and correlate these changes with clinical outcomes after percutaneous sclerotherapy (PS) of extremity venous malformations (VMs).

METHODS

VMs were segmented manually on pre- and post-treatment T2-weighted MRI using 3D Slicer to assess changes in lesion volume and SI. Clinical outcomes were scored on a 7-point Likert scale according to patient perception of symptom improvement; treatment response (success or failure) was determined accordingly.

RESULTS

Eighty-one patients with VMs underwent 125 PS sessions. Treatment success occurred in 77 patients (95 %). Mean (±SD) changes were -7.9 ± 24 cm3 in lesion volume and -123 ± 162 in SI (both, P <.001). Mean reduction in lesion volume was greater in the success group (-9.4 ± 24 cm3) than in the failure group (21 ± 20 cm3) (P =.006). Overall, lesion volume correlated with treatment response (ρ = -0.3, P =.004). On subgroup analysis, volume change correlated with clinical outcomes in children (ρ = -0.3, P =.03), in sodium tetradecyl sulfate-treated lesions (ρ = -0.5, P =.02), and in foot lesions (ρ = -0.6, P =.04). SI change correlated with clinical outcomes in VMs treated in 1 PS session (ρ = -0.3, P =.01) and in bleomycin-treated lesions (ρ = -0.4, P =.04).

CONCLUSIONS

Change in lesion volume is a reliable indicator of treatment response. Lesion volume and SI correlate with clinical outcomes in specific subgroups.

Advancements in Image-Based Models for High-Grade Gliomas Might Be Accelerated

The first half of 2022 saw the publication of several major research advances in image-based models and artificial intelligence applications to optimize treatment strategies for high-grade gliomas, the deadliest brain tumors. We review them and discuss the barriers that delay their entry into clinical practice; particularly, the small sample size and the heterogeneity of the study designs and methodologies used. We will also write about the poor and late palliation that patients suffering from high-grade glioma can count on at the end of life, as well as the current legislative instruments, with particular reference to Italy. We suggest measures to accelerate the gradual progress in image-based models and end of life care for patients with high-grade glioma.

Impact of signal intensity normalization of MRI on the generalizability of radiomic-based prediction of molecular glioma subtypes

OBJECTIVES

Radiomic features have demonstrated encouraging results for non-invasive detection of molecular biomarkers, but the lack of guidelines for pre-processing MRI-data has led to poor generalizability. Here, we assessed the influence of different MRI-intensity normalization techniques on the performance of radiomics-based models for predicting molecular glioma subtypes.

METHODS

Preoperative MRI-data from n = 615 patients with newly diagnosed glioma and known isocitrate dehydrogenase (IDH) and 1p/19q status were pre-processed using four different methods: no normalization (naive), N4 bias field correction (N4), N4 followed by either WhiteStripe (N4/WS), or z-score normalization (N4/z-score). A total of 377 Image-Biomarker-Standardisation-Initiative-compliant radiomic features were extracted from each normalized data, and 9 different machine-learning algorithms were trained for multiclass prediction of molecular glioma subtypes (IDH-mutant 1p/19q codeleted vs. IDH-mutant 1p/19q non-codeleted vs. IDH wild type). External testing was performed in public glioma datasets from UCSF (n = 410) and TCGA (n = 160).

RESULTS

Support vector machine yielded the best performance with macro-average AUCs of 0.84 (naive), 0.84 (N4), 0.87 (N4/WS), and 0.87 (N4/z-score) in the internal test set. Both N4/WS and z-score outperformed the other approaches in the external UCSF and TCGA test sets with macro-average AUCs ranging from 0.85 to 0.87, replicating the performance of the internal test set, in contrast to macro-average AUCs ranging from 0.19 to 0.45 for naive and 0.26 to 0.52 for N4 alone.

CONCLUSION

Intensity normalization of MRI data is essential for the generalizability of radiomic-based machine-learning models. Specifically, both N4/WS and N4/z-score approaches allow to preserve the high model performance, yielding generalizable performance when applying the developed radiomic-based machine-learning model in an external heterogeneous, multi-institutional setting.

CLINICAL RELEVANCE STATEMENT

Intensity normalization such as N4/WS or N4/z-score can be used to develop reliable radiomics-based machine learning models from heterogeneous multicentre MRI datasets and provide non-invasive prediction of glioma subtypes.

KEY POINTS

• MRI-intensity normalization increases the stability of radiomics-based models and leads to better generalizability. • Intensity normalization did not appear relevant when the developed model was applied to homogeneous data from the same institution. • Radiomic-based machine learning algorithms are a promising approach for simultaneous classification of IDH and 1p/19q status of glioma.

Artificial intelligence research in radiation oncology: a practical guide for the clinician on concepts and methods

The use of artificial intelligence (AI) holds great promise for radiation oncology, with many applications being reported in the literature, including some of which are already in clinical use. These are mainly in areas where AI provides benefits in efficiency (such as automatic segmentation and treatment planning). Prediction models that directly impact patient decision-making are far less mature in terms of their application in clinical practice. Part of the limited clinical uptake of these models may be explained by the need for broader knowledge, among practising clinicians within the medical community, about the processes of AI development. This lack of understanding could lead to low commitment to AI research, widespread scepticism, and low levels of trust. This attitude towards AI may be further negatively impacted by the perception that deep learning is a "black box" with inherently low transparency. Thus, there is an unmet need to train current and future clinicians in the development and application of AI in medicine. Improving clinicians' AI-related knowledge and skills is necessary to enhance multidisciplinary collaboration between data scientists and physicians, that is, involving a clinician in the loop during AI development. Increased knowledge may also positively affect the acceptance and trust of AI. This paper describes the necessary steps involved in AI research and development, and thus identifies the possibilities, limitations, challenges, and opportunities, as seen from the perspective of a practising radiation oncologist. It offers the clinician with limited knowledge and experience in AI valuable tools to evaluate research papers related to an AI model application.

Differentiation of benign versus malignant indistinguishable vertebral compression fractures by different machine learning with MRI-based radiomic features

OBJECTIVES

To explore an optimal machine learning (ML) model trained on MRI-based radiomic features to differentiate benign from malignant indistinguishable vertebral compression fractures (VCFs).

METHODS

This retrospective study included patients within 6 weeks of back pain (non-traumatic) who underwent MRI and were diagnosed with benign and malignant indistinguishable VCFs. The two cohorts were retrospectively recruited from the Affiliated Hospital of Qingdao University (QUH) and Qinghai Red Cross Hospital (QRCH). Three hundred seventy-six participants from QUH were divided into the training (n = 263) and validation (n = 113) cohort based on the date of MRI examination. One hundred three participants from QRCH were used to evaluate the external generalizability of our prediction models. A total of 1045 radiomic features were extracted from each region of interest (ROI) and used to establish the models. The prediction models were established based on 7 different classifiers.

RESULTS

These models showed favorable efficacy in differentiating benign from malignant indistinguishable VCFs. However, our Gaussian naïve Bayes (GNB) model attained higher AUC and accuracy (0.86, 87.61%) than the other classifiers in validation cohort. It also remains the high accuracy and sensitivity for the external test cohort.

CONCLUSIONS

Our GNB model performed better than the other models in the present study, suggesting that it may be more useful for differentiating indistinguishable benign form malignant VCFs.

KEY POINTS

• The differential diagnosis of benign and malignant indistinguishable VCFs based on MRI is rather difficult for spine surgeons or radiologists. • Our ML models facilitate the differential diagnosis of benign and malignant indistinguishable VCFs with improved diagnostic efficacy. • Our GNB model had the high accuracy and sensitivity for clinical application.

MR Intensity Normalization Methods Impact Sequence Specific Radiomics Prognostic Model Performance in Primary and Recurrent High-Grade Glioma

PURPOSE

This study investigates the impact of different intensity normalization (IN) methods on the overall survival (OS) radiomics models' performance of MR sequences in primary (pHGG) and recurrent high-grade glioma (rHGG).

METHODS

MR scans acquired before radiotherapy were retrieved from two independent cohorts (rHGG C1: 197, pHGG C2: 141) from multiple scanners (15, 14). The sequences are T1 weighted (w), contrast-enhanced T1w (T1wce), T2w, and T2w-FLAIR. Sequence-specific significant features (SF) associated with OS, extracted from the tumour volume, were derived after applying 15 different IN methods. Survival analyses were conducted using Cox proportional hazard (CPH) and Poisson regression (POI) models. A ranking score was assigned based on the 10-fold cross-validated (CV) concordance index (C-I), mean square error (MSE), and the Akaike information criterion (AICs), to evaluate the methods' performance.

RESULTS

Scatter plots of the 10-CV C-I and MSE against the AIC showed an impact on the survival predictions between the IN methods and MR sequences (C1/C2 C-I range: 0.62-0.71/0.61-0.72, MSE range: 0.20-0.42/0.13-0.22). White stripe showed stable results for T1wce (C1/C2 C-I: 0.71/0.65, MSE: 0.21/0.14). Combat (0.68/0.62, 0.22/0.15) and histogram matching (HM, 0.67/0.64, 0.22/0.15) showed consistent prediction results for T2w models. They were also the top-performing methods for T1w in C2 (Combat: 0.67, 0.13; HM: 0.67, 0.13); however, only HM achieved high predictions in C1 (0.66, 0.22). After eliminating IN impacted SF using Spearman's rank-order correlation coefficient, a mean decrease in the C-I and MSE of 0.05 and 0.03 was observed in all four sequences.

CONCLUSION

The IN method impacted the predictive power of survival models; thus, performance is sequence-dependent.