Comparative study of radiologists vs machine learning in differentiating biopsy-proven pseudoprogression and true progression in diffuse gliomas

• DOI: 10.1016/j.neuri.2022.100088
• Publication Date: 2022-06-13T16:57:54Z
• Authors: Sevcan Turk; Nicholas C. Wang; Omer Kitis; Shariq Mohammed; Tianwen Ma; Remy Lobo; John Kim; Sandra Camelo-Pir...; Timothy D. Johnson; Michelle M. Kim; Larry Junck; Toshio Moritani; Ashok Srinivasan; Arvind Rao; Jayapalli R. Bapuraj
• Keywords: Fluid-attenuated inversion recovery

MRI features of tumor progression and pseudoprogression may be indistinguishable especially without enhancing portion the diffuse gliomas. Our aim is to discriminate these two conditions using radiomics machine learning algorithm compare them with human observations. Three consecutive studies before a definitive biopsy in 43 glioma patients (7 36 true cases) who underwent treatment were evaluated. Two neuroradiologists reviewed pre- post-contrast T1, T2, FLAIR, ADC, rCBV, rCBF, K2, MTT maps. Patterns enhancement, ADC maps, MTT, K2 values, perilesional FLAIR signal intensity changes recorded. Odds ratios (OR) for each descriptor, raters' success predicting pseudoprogression, inter-observer reliability calculated R statistics software. Unpaired Student's t-test receiver operating characteristic (ROC) analysis applied texture parameters histogram pseudo- groups. All first-order second-order image shape used train test Random Forest classifier (RFC). Observers' RFC compared. Observers could not identify first visit. However, accuracy model was 81%. For second third visits, rater's prediction between 62% 72%. The last visit 75% more successful than observations MRI.