Machine learning predicts histologic type and grade of canine gliomas based on MRI texture analysis

Abstract

Conventional MRI features of canine gliomas subtypes and grades significantly overlap. Texture analysis (TA) quantifies image texture based on spatial arrangement of pixel intensities. Machine learning (ML) models based on MRI-TA demonstrate high accuracy in predicting brain tumor types and grades in human medicine. The aim of this retrospective, diagnostic accuracy study was to investigate the accuracy of ML-based MRI-TA in predicting canine gliomas histologic types and grades. Dogs with histopathological diagnosis of intracranial glioma and available brain MRI were included. Tumors were manually segmented across their entire volume in enhancing part, non-enhancing part, and peri-tumoral vasogenic edema in T2-weighted (T2w), T1-weighted (T1w), FLAIR, and T1w postcontrast sequences. Texture features were extracted and fed into three ML classifiers. Classifiers' performance was assessed using a leave-one-out cross-validation approach. Multiclass and binary models were built to predict histologic types (oligodendroglioma vs. astrocytoma vs. oligoastrocytoma) and grades (high vs. low), respectively. Thirty-eight dogs with a total of 40 masses were included.Machine learning classifiers had an average accuracy of 77% for discriminating tumor types and of 75.6% for predicting high-grade gliomas. The support vector machine classifier had an accuracy of up to 94% for predicting tumor types and up to 87% for predicting high-grade gliomas. The most discriminative texture features of tumor types and grades appeared related to the peri-tumoral edema in T1w images and to the non-enhancing part of the tumor in T2w images, respectively. In conclusion, ML-based MRI-TA has the potential to discriminate intracranial canine gliomas types and grades.

An Illustrated Scoping Review of the Magnetic Resonance Imaging Characteristics of Canine and Feline Brain Tumors

“…After the exclusion of studies describing experimentally induced brain tum other irrelevant records, de-duplication of records appearing repeatedly in our searches using the different MeSH search phrases, and removal of studies that c < 3 cases, unconfirmed diagnoses, or limited MRI data, we identified 67 articles (F that fulfilled the inclusion criteria, with these studies reporting a total of 1630 cani 59 studies) and 125 feline cases (from 9 studies) with tumors [2,[6][7][8][9][10][11][12][13]25,26, of 62/67 studies were classified using Oxford hierarchal evidentiary criteria [2 (73%) were level 2b retrospective cohort studies [2,6-11,25,26,29,31-33,35-37,40-4 59,63-67,69-72,74,77,78,81-83], 14/62 (23%) were level 4 case series each describ animals [30,38,43,47,49,50,58,61,62,73,75,76,79,80], and 3/62 (4%) were level 3a 'm views containing source data quality heterogeneity [12][13][14]. Five radiomic stud identified [34,[44][45][46]68], with all radiomic studies having QUADAS-2 scores ≤ 6, were considered to provide low-level evidence [28]. The radiomic studies includ high risk of bias within the patient selection domain, manifesting as small sam given the number of variables tested or imbalances among analytical subgroups For the generation of a prioritized list of differential diagnoses based on characteristics of brain lesions, a recurrent theme that emerged from the literatur identify the number of lesions (solitary; multifocal/diffuse) present and then cla neuroanatomic locati...…”

“…The infarct is heterogeneously iso-to hypointense on T2*GRE, is rin hancing, associated with significant mass effect, and demonstrates T2W blackout (lesion hypoi sity on DWI and ADC) due to susceptibility effects of hemorrhage. On DWI and ADC image peripheral hyperintensity surrounding the hypointense lesion core represents perilesional ede The primary neoplastic differential diagnostic considerations for solitary intramasses are neuroepithelial tumors, among which oligodendrogliomas and astrocyto (i.e., gliomas) predominate in the dog (Figure 5, Cases 16-20), with other uncommo rare possible differentials including undefined glioma (oligoastrocytoma), brain met sis, ependymoma, lymphoma, HS, and embryonal tumors [2,9,10,31,32,55,[66][67][68][69][70][71][72][73][74][75][76][77][78][79][80]. Glio can have wide-ranging MRI appearances, resulting in imaging features that may ove substantially with brain abscesses, ischemic and hemorrhagic brain infarctions, fu granulomas, immune-mediated encephalitides, leukoencephalopathies, and meningi [2,[6][7][8]10,[12][13][14]67,77,78].…”

“…The primary neoplastic differential diagnostic considerations for solitary intra-axial masses are neuroepithelial tumors, among which oligodendrogliomas and astrocytomas (i.e., gliomas) predominate in the dog (Figure 5, Cases 16-20), with other uncommon to rare possible differentials including undefined glioma (oligoastrocytoma), brain metastasis, ependymoma, lymphoma, HS, and embryonal tumors [2,9,10,31,32,55,[66][67][68][69][70][71][72][73][74][75][76][77][78][79][80]. Gliomas can have wide-ranging MRI appearances, resulting in imaging features that may overlap substantially with brain abscesses, ischemic and hemorrhagic brain infarctions, fungal granulomas, immune-mediated encephalitides, leukoencephalopathies, and meningioma [2,[6][7][8]10,[12][13][14]67,77,78].…”

“…However, the classification accuracies of texture-based features to identify the type and grade of gliomas and the specific subtypes of autoimmune meningoencephalitides were all too low to be clinically useful [34]. Barge et al extracted texture-based features from 40 canine gliomas, analyzed those features using three machine learning classifiers, and built models to predict tumor type (astrocytoma vs. oligodendroglioma) and grade (high vs. low) [68]. The average reported accuracies for the machine learning classifiers were 77% for discriminating tumor types and 76% for predicting high-grade gliomas.…”

“…The average reported accuracies for the machine learning classifiers were 77% for discriminating tumor types and 76% for predicting high-grade gliomas. Across the machine learning methods, the support vector machine classifier demonstrated superior performance, with accuracies of 94% for predicting tumor types and 87% for tumor grade [68]. Texture features extracted from peritumoral edema in T1W images and from non-enhancing tumors in T2W were demonstrated to be more discriminative for the tumor type and grade, respectively [68].…”