Machine Assist for Pediatric Posterior Fossa Tumor Diagnosis: A Multinational Study

Abstract

BACKGROUND Clinicians and machine classifiers reliably diagnose pilocytic astrocytoma (PA) on magnetic resonance imaging (MRI) but less accurately distinguish medulloblastoma (MB) from ependymoma (EP). One strategy is to first rule out the most identifiable diagnosis. OBJECTIVE To hypothesize a sequential machine-learning classifier could improve diagnostic performance by mimicking a clinician's strategy of excluding PA before distinguishing MB from EP. METHODS We extracted 1800 total Image Biomarker Standardization Initiative (IBSI)-based features from T2- and gadolinium-enhanced T1-weighted images in a multinational cohort of 274 MB, 156 PA, and 97 EP. We designed a 2-step sequential classifier – first ruling out PA, and next distinguishing MB from EP. For each step, we selected the best performing model from 6-candidate classifier using a reduced feature set, and measured performance on a holdout test set with the microaveraged F1 score. RESULTS Optimal diagnostic performance was achieved using 2 decision steps, each with its own distinct imaging features and classifier method. A 3-way logistic regression classifier first distinguished PA from non-PA, with T2 uniformity and T1 contrast as the most relevant IBSI features (F1 score 0.8809). A 2-way neural net classifier next distinguished MB from EP, with T2 sphericity and T1 flatness as most relevant (F1 score 0.9189). The combined, sequential classifier was with F1 score 0.9179. CONCLUSION An MRI-based sequential machine-learning classifiers offer high-performance prediction of pediatric posterior fossa tumors across a large, multinational cohort. Optimization of this model with demographic, clinical, imaging, and molecular predictors could provide significant advantages for family counseling and surgical planning.

Machine Learning in the Classification of Pediatric Posterior Fossa Tumors: A Systematic Review

“…Table 1 features study method data from all 25 studies of the MLAs applied to the classification of pediatric PFTs. Twenty-two papers used imaging data to classify PFTs, including both non-contrast and contrast-enhanced T1/T2-weighted MRI, DWI, and MR-spectroscopy [ 16 , 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 , 28 , 29 , 30 , 31 , 32 , 33 , 34 , 35 , 36 , 37 ]. Three papers used molecular methods to classify these tumors based on microscopy slides or methylation array data [ 38 , 39 , 40 ].…”

“…Study populations varied significantly, ranging from cohorts of 23 patients to 617 patients [ 31 , 40 ]. Pilocytic astrocytoma and medulloblastoma were the most well-represented PFTs across all reports with inclusion in 19 and 22 studies, respectively [ 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 , 28 , 29 , 30 , 31 , 32 , 34 , 35 , 36 , 37 , 38 , 39 , 40 ]. Ependymomas, while included in most studies, had a small individual sample size per study, with many analyses including fewer than 20 ependymoma patients in training or validation datasets [ 16 , 17 , 18 , 21 , 22 , 23 , 27 , 29 , 30 , 32 , 36 ].…”

“…Features were extracted from the relevant imaging modalities with some papers yielding as few as 13 features while others generated over 11,000 [ 20 , 39 ]. Nineteen studies employed various methods of feature selection to decrease feature dimensionality for the final analysis [ 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 , 29 , 30 , 32 , 33 , 34 , 35 , 36 , 37 , 38 , 40 ]. A machine learning classifier was then applied to discriminate between tumor types based on extracted features.…”

“…Twenty-three studies reported general algorithm performance metrics citing mainly AUCs, accuracies, sensitivities, and specificities ( Table 3 and Table 4 , Supplementary Table S1 ) [ 16 , 17 , 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 , 29 , 30 , 31 , 32 , 34 , 35 , 36 , 37 , 38 , 39 , 40 ]. Algorithms performed well only when differentiating between two tumor types.…”

Dynamic susceptibility contrast and diffusion‐weighted MRI in posterior fossa pilocytic astrocytoma and medulloblastoma

“…Significant efforts have been made to distinguish between these two tumors. Diffusion MRI techniques, including apparent diffusion coefficient (ADC) values/histograms [5][6][7][8][9] and diffusion tensor imaging, [10][11][12] MR spectroscopy, [13][14][15] and machine/deep learning, [16][17][18] have been the main foci of previous studies. However, reports on the diagnostic performance of perfusion MRI are scarce, with heterogeneous results.…”