Radiomics and supervised machine learning in the diagnosis of parkinsonism with FDG PET: promises and challenges

Subregion-specific 18F-FDG PET-CT radiomics for the pre-treatment prediction of EGFR mutation status in solid lung adenocarcinoma

This study aimed to assess the efficacy of fluor-18 fluorodeoxyglucose (18F-FDG) PET/CT using sub-regional-based radiomics in predicting epidermal growth factor receptor (EGFR) mutation status in pretreatment patients with solid lung adenocarcinoma. A retrospective analysis included 269 patients (134 EGFR+ and 135 EGFR-) who underwent pretreatment 18F-FDG PET/CT scans and EGFR mutation testing. The most metabolically active intratumoral sub-region was identified, and radiomics features from whole tumors or sub-regional regions were used to build classification models. The dataset was split into a 7:3 ratio for training and independent testing. Feature subsets were determined by Pearson correlation and the Kruskal Wallis test and radiomics classifiers were built with support vector machines or logistic regressions. Evaluation metrics, including accuracy, area under the curve (AUC), sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were employed for different classifiers. Results indicated that the sub-region-based classifier outperformed the whole-tumor classifier in terms of accuracy (73.8% vs. 66.2%), AUC (0.768 vs. 0.632), specificity (65.0% vs. 50.0%), PPV (70.2% vs. 62.2%), and NPV (78.8% vs. 74.0%). The clinical classifier exhibited an accuracy of 75.0%, AUC of 0.768, sensitivity of 72.5%, specificity of 77.5%, PPV of 76.3%, and NPV of 73.8%. The combined classifier, incorporating sub-region analysis and clinical parameters, demonstrated further improvement with an accuracy of 77.5%, AUC of 0.807, sensitivity of 77.5%, specificity of 77.5%, and NPV of 77.5%. The study suggests that sub-region-based 18F-FDG PET/CT radiomics enhances EGFR mutation prediction in solid lung adenocarcinoma, providing a practical and cost-efficient alternative to invasive EGFR testing.

Random forest analysis of midbrain hypometabolism using [18F]-FDG PET identifies Parkinson's disease at the subject-level

Parkinson's disease (PD) is currently diagnosed largely on the basis of expert judgement with neuroimaging serving only as a supportive tool. In a recent study, we identified a hypometabolic midbrain cluster, which includes parts of the substantia nigra, as the best differentiating metabolic feature for PD-patients based on group comparison of [18F]-fluorodeoxyglucose ([18F]-FDG) PET scans. Longitudinal analyses confirmed progressive metabolic changes in this region and, an independent study showed great potential of nigral metabolism for diagnostic workup of parkinsonian syndromes. In this study, we applied a machine learning approach to evaluate midbrain metabolism measured by [18F]-FDG PET as a diagnostic marker for PD. In total, 51 mid-stage PD-patients and 16 healthy control subjects underwent high-resolution [18F]-FDG PET. Normalized tracer update values of the midbrain cluster identified by between-group comparison were extracted voxel-wise from individuals' scans. Extracted uptake values were subjected to a random forest feature classification algorithm. An adapted leave-one-out cross validation approach was applied for testing robustness of the model for differentiating between patients and controls. Performance of the model across all runs was evaluated by calculating sensitivity, specificity and model accuracy for the validation data set and the percentage of correctly categorized subjects for test data sets. The random forest feature classification of voxel-based uptake values from the midbrain cluster identified patients in the validation data set with an average sensitivity of 0.91 (Min: 0.82, Max: 0.94). For all 67 runs, in which each of the individuals was treated once as test data set, the test data set was correctly categorized by our model. The applied feature importance extraction consistently identified a subset of voxels within the midbrain cluster with highest importance across all runs which spatially converged with the left substantia nigra. Our data suggest midbrain metabolism measured by [18F]-FDG PET as a promising diagnostic imaging tool for PD. Given its close relationship to PD pathophysiology and very high discriminatory accuracy, this approach could help to objectify PD diagnosis and enable more accurate classification in relation to clinical trials, which could also be applicable to patients with prodromal disease.

FDG-PET combined with learning vector quantization allows classification of neurodegenerative diseases and reveals the trajectory of idiopathic REM sleep behavior disorder

BACKGROUND AND OBJECTIVES

¹⁸F-fluorodeoxyglucose (FDG) positron emission tomography (PET) combined with principal component analysis (PCA) has been applied to identify disease-related brain patterns in neurodegenerative disorders such as Parkinson's disease (PD), Dementia with Lewy Bodies (DLB) and Alzheimer's disease (AD). These patterns are used to quantify functional brain changes at the single subject level. This is especially relevant in determining disease progression in idiopathic REM sleep behavior disorder (iRBD), a prodromal stage of PD and DLB. However, the PCA method is limited in discriminating between neurodegenerative conditions. More advanced machine learning algorithms may provide a solution. In this study, we apply Generalized Matrix Learning Vector Quantization (GMLVQ) to FDG-PET scans of healthy controls, and patients with AD, PD and DLB. Scans of iRBD patients, scanned twice with an approximate 4 year interval, were projected into GMLVQ space to visualize their trajectory.

METHODS

We applied a combination of SSM/PCA and GMLVQ as a classifier on FDG-PET data of healthy controls, AD, DLB, and PD patients. We determined the diagnostic performance by performing a ten times repeated ten fold cross validation. We analyzed the validity of the classification system by inspecting the GMLVQ space. First by the projection of the patients into this space. Second by representing the axis, that span this decision space, into a voxel map. Furthermore, we projected a cohort of RBD patients, whom have been scanned twice (approximately 4 years apart), into the same decision space and visualized their trajectories.

RESULTS

The GMLVQ prototypes, relevance diagonal, and decision space voxel maps showed metabolic patterns that agree with previously identified disease-related brain patterns. The GMLVQ decision space showed a plausible quantification of FDG-PET data. Distance traveled by iRBD subjects through GMLVQ space per year (i.e. velocity) was correlated with the change in motor symptoms per year (Spearman's rho =0.62, P=0.004).

CONCLUSION

In this proof-of-concept study, we show that GMLVQ provides a classification of patients with neurodegenerative disorders, and may be useful in future studies investigating speed of progression in prodromal disease stages.