MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel
  1. Home
  2. Meeting Program
  3. All Sessions
  4. All Imaging General ePoster Discussion
  5. Differentiation of Brain Tumors by Characterization of Non-Contrast-Enhanced MR Images Using Topological Data Analysis

Differentiation of Brain Tumors by Characterization of Non-Contrast-Enhanced MR Images Using Topological Data Analysis

J Kotoku1*, A Oyama1 , Y Hiraoka2 , I Obayashi3 , Y Saikawa1 , K Shiraishi4 , S Kumagai5 , A Haga6 ,S Furui4 , H Oba4 , (1) Graduate School of Medical Care and Technology, Teikyo University, Tokyo, , (2) Kyoto University Institute for Advanced Study (KUIAS), Kyoto, ,(3) World Premier International Research Center Initiative-Advanced Institut, Sendai, ,(4) Department of Radiology, Teikyo University School of Medicine, Tokyo, ,(5) Central Radiology Division, Teikyo University Hospital, Tokyo, ,(6) Tokushima University

Presentations

(Wednesday, 7/17/2019) 10:30 AM - 11:00 AM

Room: Exhibit Hall | Forum 8

Purpose: This study differentiated high-grade glioma (HGG) and brain metastases (BM) by characterization of non-contrast-enhanced T1-/T2-weighted magnetic resonance (MR) images using persistent homology and analysis of the data using machine learning models.

Methods: This retrospective study was approved by the ethics committee of our university hospital. Non-contrast-enhanced T1-/T2-weighted MR images of 44 brain tumors, obtained from May 2007 to September 2017, were assessed. The lesions included 22 HGGs and 22 BMs found in 21 and 16 patients, respectively. For classification, persistence diagrams for the tumors were generated from the MR images of each lesion. Then, input vectors were obtained from vectorized persistence diagrams to apply machine learning models. We used two models (i.e., logistic classifier model with an elastic net penalty and extreme gradient boosting [XGBoost]) to classify HGG and BM and evaluate the classification abilities.

Results: In classification of T1-weighted (resp. T2-weighted) MR images of HGG and BM, we obtained the highest classification accuracy (±SD) of 83.0±0.0% (resp. 93.2±0.0%) when the degree 2 (degree 1) vectors were used, with the sensitivity, specificity, and area under the receiver operating characteristic curve of 88.6±6.8% (resp. 93.2±2.3%), 77.3±13.6% (resp. 93.2±2.3%), and 0.860±0.037 (0.868±0.000), respectively. Our method using persistent homology allows for classification of HGG and BM only from non-contrast-enhanced MR images of the brain with considerable accuracy.

Conclusion: This method may increase usefulness when utilized in the computer-aided diagnosis of brain tumors with MR imaging.

Funding Support, Disclosures, and Conflict of Interest: This work was supported in part by Japan Society for the Promotion of Science (JSPS) KAKENHI Grant No. 18K07646.

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email

Additional Links