Abstract Details

Title Development of a workflow efficient PACS based automated brain tumor segmentation and radiomic feature extraction for clinical implementation

Topic Neuro-oncology

Presentation(s) N2 - Neuroscience in the Clinic: Applications of Artificial Intelligence and Machine Learning Tools in Neurology (4:40 PM-4:50 PM)

Poster/Presentation
Number 003

Background

Deep learning artificial intelligence is a powerful tool for automated segmentation although integrating this tool into the clinical workflow has been challenging. In this work, we implemented a deep learning-based algorithm for automated brain tumor segmentation and embedded it into PACS to accelerate a supervised, end-to-end workflow for radiomic feature extraction.

Objective

To design a clinically applicable workflow for application of AI tools in Neuro-oncology.

Design/Methods

An algorithm was trained to segment primary brain tumors on MRI FLAIR images from the multi-institutional BRATS 2021 dataset which includes segmentations of glioblastoma and lower grade glioma. Afterwards the algorithm was validated using an internal dataset from Yale New Haven Health (YHHH). A UNETR deep-learning network architecture was employed and embedded into Visage 7 (Visage Imaging, Inc., San Diego, CA) diagnostic workstation. The automatically segmented brain tumor was pliable for manual modification. PyRadiomics (Harvard Medical School, Boston, MA) was natively embedded into Visage 7 for feature extraction from segmentations in form of JSON files. Titan Xp, NVIDIA , Santa Clara, CA was used for tumor segmentation.

Results On our system, the AI brain tumor segmentation took on average 4 seconds and the median dice similarity coefficient was 86%. Finally, extraction of radiomic features took on average 5.8±0.01 seconds. The extracted radiomic features did not vary over time of extraction or whether they were extracted within PACS or outside of PACS. The ability to perform segmentation and feature extraction before radiologist opens the study was made available in the workflow.

Conclusions

Integration of advanced image-based algorithms and extraction of imaging biomarkers into PACS systems can accelerate translation of research into development of personalized medicine applications in the clinic. The ability to use familiar clinical tools to revise the AI segmentations on the diagnostic workstation reduce the time needed to generate ground-truth data.

Authors/Disclosures
Mariam Aboian, MD, PhD (Yale University) PRESENTER	Dr. Aboian has a non-compensated relationship as a Principal Investigator with Visage Imaging that is relevant to AAN interests or activities.
	No disclosure on file
	No disclosure on file
	No disclosure on file
	No disclosure on file
Sara Merkaj (Yale School of Medicine)	Mrs. Merkaj has nothing to disclose.
Gabriel I. Cassinelli Petersen (Yale School of Medicine)	Mr. Cassinelli Petersen has nothing to disclose.
Ryan C. Bahar (Yale School of Medicine)	Mr. Bahar has nothing to disclose.
	No disclosure on file
	No disclosure on file
	No disclosure on file
Amit Mahajan (Yale University)	No disclosure on file
Ajay Malhotra	No disclosure on file
Sam Payabvash	No disclosure on file
	No disclosure on file
	No disclosure on file
	No disclosure on file