Abstract Details

Title Integration of Machine Learning Into Clinical Radiology Practice – Development of a Machine Learning Tool for Preoperative Glioma Grade Prediction

Topic Neuro-oncology

Presentation(s) P14 - Poster Session 14 (11:45 AM-12:45 PM)

Poster/Presentation
Number 4-002

Objective

To develop a non-invasive, clinically integrated, preoperative glioma grade prediction tool, which can assist clinicians in their daily clinical practice.

Background

The most common primary brain malignancy, gliomas, are graded according to the World Health Organization (WHO) criteria into grades 1/2 (low-grade gliomas (LGG)) and grades 3/4 (high-grade gliomas (HGG)). As prognosis and treatment plan vary between HGGs and LGGs, there’s a need for preoperative glioma grading. The current gold standard for diagnosis is histopathology, which is a lengthy, invasive procedure, which carries associated surgical risks. Machine learning (ML) models and radiomics present novel solutions to overcome these current obstacles.

Design/Methods We trained a PACS integrated deep learning algorithm (U-Net), which was pretrained on BraTS and retrained on Yale hospital data, to auto-segment whole, core and necrotic portions of gliomas on Visage 7 (Visage Imaging, Inc., San Diego, CA) platform. Segmentations were made available on FLAIR, T1, T1ce, T2, ADC and SWI MRI sequences, with FLAIR segmentations only being included in the preliminary predictions. We extracted 1605 quantitative features from each segmentation using PyRadiomics. Features with <50% presence across all segmentations and where the Pearson correlation between two features exceeded a threshold of 0.8 were dropped. A ML algorithm, XGBoost, performed HGG vs LGG predictions. Our model was internally validated using the five-fold cross-validation method.

Results PACS based auto-segmentation tool was used to segment whole 3D tumor volumes with dice score of 0.89. 139 final radiomic features were identified from 202 segmented gliomas (111 HGG, 91 LGG) and incorporated into XGBoost algorithm. After training on 80% of the data, XGBoost was able to achieve an AUC of 0.84 ± 0.06 for HGG vs. LGG prediction on the test set.

Conclusions

PACS based auto-segmentation linked to XGBoost classification of HGG vs LGG allows clinical translation of image-based glioma grade prediction into clinical practice.

Authors/Disclosures
Sara Merkaj (Yale School of Medicine) PRESENTER	Mrs. Merkaj has nothing to disclose.
	No disclosure on file
	No disclosure on file
	No disclosure on file
	No disclosure on file
	No disclosure on file
Gabriel I. Cassinelli Petersen (University of Tübingen)	Mr. Cassinelli Petersen has nothing to disclose.
Leon Jekel	Mr. Jekel has nothing to disclose.
Ryan C. Bahar (Yale School of Medicine)	Mr. Bahar has nothing to disclose.
Niklas J. Tillmanns	Mr. Tillmanns has received research support from Heinrich Heine University Duesseldorf.
	No disclosure on file
	No disclosure on file
Richard A. Bronen, MD (Yale Univ Sch of Med)	Dr. Bronen has received personal compensation in the range of $500-$4,999 for serving as an Expert Witness for Blankingship & Keith. Dr. Bronen has received personal compensation in the range of $10,000-$49,999 for serving as an Expert Witness for Hinshaw Culbertson. Dr. Bronen has received personal compensation in the range of $500-$4,999 for serving as an Expert Witness for Triplett. Dr. Bronen has received personal compensation in the range of $500-$4,999 for serving as an Expert Witness for Weinmuller. Dr. Bronen has received personal compensation in the range of $500-$4,999 for serving as an Expert Witness for Farmers Insurance. Dr. Bronen has received personal compensation in the range of $500-$4,999 for serving as an Expert Witness for Aldrich, Hanks & Sheehan. An immediate family member of Dr. Bronen has received stock or an ownership interest from Sed Med.
	No disclosure on file
	No disclosure on file
Antonio M. Omuro, MD, FAAN (Stanford University)	Dr. Omuro has received personal compensation in the range of $500-$4,999 for serving as a Consultant for Ono Therapeutics. Dr. Omuro has received personal compensation in the range of $500-$4,999 for serving as a Consultant for Telix. Dr. Omuro has received personal compensation in the range of $500-$4,999 for serving on a Scientific Advisory or Data Safety Monitoring board for Curevac. The institution of Dr. Omuro has received research support from NIH. The institution of Dr. Omuro has received research support from Arcus Biosciences. The institution of Dr. Omuro has received research support from Denovo Biopharma. The institution of Dr. Omuro has received research support from Ono Pharmaceutical. The institution of Dr. Omuro has received research support from Servier. The institution of Dr. Omuro has received research support from Nanopharmaceuticals. The institution of Dr. Omuro has received research support from Denovo.
Mariam Aboian, MD, PhD (Yale University)	Dr. Aboian has a non-compensated relationship as a Principal Investigator with Visage Imaging that is relevant to AAN interests or activities.

��ɫ��

��ɫ��