Fully Automated Segmentation of Head CT Neuroanatomy Using Deep Learning

Abstract

To develop a deep learning model that segments intracranial structures on head CT scans. Materials and Methods:In this retrospective study, a primary dataset containing 62 normal noncontrast head CT scans from 62 patients (mean age, 73 years; age range, 27-95 years) acquired between August and December 2018 was used for model development. Eleven intracranial structures were manually annotated on the axial oblique series. The dataset was split into 40 scans for training, 10 for validation, and 12 for testing. After initial training, eight model configurations were evaluated on the validation dataset and the highest performing model was evaluated on the test dataset. Interobserver variability was reported using multirater consensus labels obtained from the test dataset. To ensure that the model learned generalizable features, it was further evaluated on two secondary datasets containing 12 volumes with idiopathic normal pressure hydrocephalus (iNPH) and 30 normal volumes from a publicly available source. Statistical significance was determined using categorical linear regression with P , .05.Results: Overall Dice coefficient on the primary test dataset was 0.84 6 0.05 (standard deviation). Performance ranged from 0.96 6 0.01 (brainstem and cerebrum) to 0.74 6 0.06 (internal capsule). Dice coefficients were comparable to expert annotations and exceeded those of existing segmentation methods. The model remained robust on external CT scans and scans demonstrating ventricular enlargement. The use of within-network normalization and class weighting facilitated learning of underrepresented classes. Conclusion:Automated segmentation of CT neuroanatomy is feasible with a high degree of accuracy. The model generalized to external CT scans as well as scans demonstrating iNPH.

Automated multiclass tissue segmentation of clinical brain MRIs with lesions

“…While other deep learning methods require less preprocessing ( Bontempi et al, 2020 , Dolz et al, 2018 ), or produce tissue segmentations in less time than ours ( Guha Roy et al, 2019 , Bontempi et al, 2020 ), these methods ( Henschel et al, 2020 , Zhang et al, 2021 , Sendra-Balcells et al, 2020 , Tushar et al, 2019 ) were developed for and tested in healthy subjects or patients with diseases that do not present with focal brain lesions that alter signal and distort anatomy. More recently methods have been developed and tested in patients with a single pathology ( Mendrik et al, 2015 , Moeskops et al, 2016 , Luna and Park, 2018 , Chen et al, 2018 , de Boer et al, 2009 , Cai et al, 2020 , Valverde et al, 2017 , Liu et al, 2021 ). In contrast, our method was developed using a wide variety of gray and white matter pathologies.…”

“…Recently, U-Nets have won multiple image segmentation challenges including the ischemic stroke lesions segmentation (ISLES) ( Song, 2018 ) and multimodal brain tumor segmentation (BraTS) ( Myronenko, 2018 , Bakas et al, 2019 ) challenges. Others have developed U-Nets for tissue segmentation in nonlesional brain MRIs ( Guha Roy et al, 2019 , Bontempi et al, 2020 , Dolz et al, 2018 , Henschel et al, 2020 , Zhang et al, 2021 ), head CT scans with and without hydrocephalus ( Cai et al, 2020 ), and brain MRIs in the context of a variety of specific pathologies including white matter hyperintensities ( Mendrik et al, 2015 , Moeskops et al, 2016 , Luna and Park, 2018 , Chen et al, 2018 , de Boer et al, 2009 ), multiple sclerosis ( Valverde et al, 2017 ), and gliomas ( Liu et al, 2021 ). However, none have been developed that are capable of segmenting brain tissue across MRIs from patients with a wide variety of underlying gray matter and white matter pathologies.…”

Multi-Modal Segmentation of 3D Brain Scans Using Neural Networks

“…Alternatively, the annotations could also be directly added to the DWI, FLAIR, and CT volumes. This was recently done for CT scans (33) and would alleviate the systematic errors introduced by non-perfect coregistration.…”

Head CT: Toward Making Full Use of the Information the X-Rays Give

“…Although still in its early stages, machine learning may prove useful in this regard. 52 If gray and white matter have discrete radiodensities, the radiodensity of a given lobe or anatomic segment of brain would be expected to largely reflect the percentages of gray and white matter in a particular substructure, though this idea has yet to be verified empirically. Dual Source CT. As discussed above, the beam-hardening artifact from the skull has long posed a problem for head CT imaging.…”