Segmentation and Measurement of Ventricular and Cranial Vault Volumes in 15 223 Subjects Using Artificial Intelligence. (1st September 2019)

• Record Type: Journal Article
• Authors: Maragkos, Georgios; Chilamkurthy, Sasank; Gomez-Paz, Santiago; Salem, Mohamed M; Tanamala, Swetha; Rao, Pooja; Hackney, David; Thomas, Ajith J
• Is Part Of: Neurosurgery. Volume 66(2010)Supplement 1
• Publication Date: 2019-09-01
• Subjects: Nervous system -- Surgery -- Periodicals; 617.48005
• DOI: 10.1093/neuros/nyz310_415
• Languages: English
• ISSNs: 0148-396X

Abstract

Abstract: INTRODUCTION: True volumetric measurement of the cerebral ventricles could potentially improve the diagnosis and follow-up of hydrocephalus. However, manual image segmentation is an impractically laborious process to be applied routinely. Here, we utilize deep learning algorithms for precise ventricular segmentation and volume quantification in a large dataset. METHODS: A large head computed tomography (CT) dataset was utilized to train and validate convoluted neural network models (U-nets). Lateral ventricles were manually annotated in 103 scans and randomly split 4:1 for training-validation. One U-net model was trained to segment the lateral ventricles in each CT slice and output their volumes, and another for the cranial vaults. Each model was validated against the manually annotated images. Both networks were then used to segment and quantify volumes of a large random subset from our database. RESULTS: Both U-net models showed high fidelity to the manual annotation (Dice Indices 0.909 and 0.983). They were then deployed on 15 223 head CT scans from our database. Among these scans, 1999 (13.1%) had a radiological report of cerebral atrophy and 1404 (9.2%) of hydrocephalus, ascertained through natural language processing (NLP) of the report of the human interpretation. In 13, 388 patients without hydrocephalus, ventricular volume increased with age indefinitely, presumably due to cerebral atrophy. Cranial vault volume increased until the age group of 10 to 20, after which a plateau was observed, and was consistently about 13% larger in males than females. Conversely, ventricle-to-vault ratios showed no difference between the 2 sexes in younger ages, but men had a steeper increase with age than women. CONCLUSION: This is the first study to measure ventricular volumes in such a large dataset, made possible using artificial intelligence. We provide a robust method to establish normal values for ventricular volumes and a tool to routinely report these volumes on CT scans, when evaluating for hydrocephalus.