Better efficacy in differentiating WHO grade II from III oligodendrogliomas with machine-learning than radiologist’s reading from conventional T1 contrast-enhanced and fluid attenuated inversion recovery images

Abstract

Background: The medical imaging to differentiate World Health Organization (WHO) grade II (ODG2) from III (ODG3) oligodendrogliomas still remains a challenge. We investigated whether combination of machine leaning with radiomics from conventional T1 contrast-enhanced (T1 CE) and fluid attenuated inversion recovery (FLAIR) magnetic resonance imaging (MRI) offered superior efficacy.Methods: Thirty-six patients with histologically confirmed ODGs underwent T1 CE and 33 of them underwent FLAIR MR examination before any intervention from January 2015 to July 2017 were retrospectively recruited in the current study. The volume of interest (VOI) covering the whole tumor enhancement were manually drawn on the T1 CE and FLAIR slice by slice using ITK-SNAP and a total of 1072 features were extracted from the VOI using 3-D slicer software. Random forest (RF) algorithm was applied to differentiate ODG2 from ODG3 and the efficacy was tested with 5-fold cross validation. The diagnostic efficacy of radiomics-based machine learning and radiologist's assessment were also compared. Results: Nineteen ODG2 and 17 ODG3 were included in this study and ODG3 tended to present with prominent necrosis and nodular/ring-like enhancement (P < 0.05). The AUC, ACC, sensitivity, and specificity of radiomics were 0.798, 0.735, 0.672, 0.789 for T1 CE, 0.774, 0.689, 0.700, 0.683 for FLAIR, as well as 0.861, 0.781, 0.778, 0.783 for the combination, respectively. The AUCs of radiologists 1, 2 and 3 were 0.700, 0.687, and 0.714, respectively. The efficacy of machine learning based on radiomics was superior to the radiologists' assessment. Conclusions: Machine-learning based on radiomics of T1 CE and FLAIR offered superior efficacy to that of radiologists in differentiating ODG2 from ODG3.

Diagnostic Accuracy of Machine Learning-Based Radiomics in Grading Gliomas: Systematic Review and Meta-Analysis

“…A summary of the results is presented in Table 3 , while the method-related information is summarized in Table 4 . Three of five studies utilized a single MRI sequence acquired by either conventional or advanced imaging [ 14 , 19 , 20 ], while the remaining studies implemented both conventional and advanced ASL and DWI sequences in [ 21 , 22 ]. An imbalanced ratio was observed between the HGG and LGG datasets in the studies that used a large number of samples [ 19 , 21 ], while the remaining three studies had a ratio equal to the sample [ 14 , 20 , 22 ].…”

“…Three of five studies utilized a single MRI sequence acquired by either conventional or advanced imaging [ 14 , 19 , 20 ], while the remaining studies implemented both conventional and advanced ASL and DWI sequences in [ 21 , 22 ]. An imbalanced ratio was observed between the HGG and LGG datasets in the studies that used a large number of samples [ 19 , 21 ], while the remaining three studies had a ratio equal to the sample [ 14 , 20 , 22 ]. Two of the five studies selected feature numbers equal to or greater than the total sample size [ 14 , 20 ], while the remaining three studies selected fewer than the total sample size [ 19 , 21 , 22 ].…”

“…An imbalanced ratio was observed between the HGG and LGG datasets in the studies that used a large number of samples [ 19 , 21 ], while the remaining three studies had a ratio equal to the sample [ 14 , 20 , 22 ]. Two of the five studies selected feature numbers equal to or greater than the total sample size [ 14 , 20 ], while the remaining three studies selected fewer than the total sample size [ 19 , 21 , 22 ]. Only two studies included a testing set [ 20 ] without reporting the sample size of the testing set [ 19 ].…”

“…Therefore, the imbalanced ratio between LGG and HGG (total sample 240 vs. 389) may increase the sensitivity due to the large sample size of HGG without considering the distribution of data ratio between LGG and HGG, while reducing the specificity performance [ 19 , 21 ]. In contrast, the overall equal sample ratio was found in the group with a sample size of less than 100, as summarized in Table 3 [ 14 , 20 , 22 ]. As a result, the specificity may increase in the small sample size group that did not consider the LGG sample as a minor group during the classifier training.…”

“…The assessment of both mutation status and gene expression, such as O6-methylguanine-DNA-methyltransferase (MGMT) gene expression or isocitrate dehydrogenase (IDH) mutation, is essential for predicting therapeutic responses when treating gliomas. Radiomics has proven the potential for the genotype classification of prognostic factors to predict IDH status, 1/19q codeletion status, or MGMT methylation status [ 12 – 14 ] in glioma-related studies. However, the treatment of gliomas cannot be processed by genotype alone since IDH mutation or 1/19q codeletion status tends to be used for classifying grades II and III and MGMT promoter methylation for grade IV.…”