Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2022 Dec 1;17(12):e0277573.
doi: 10.1371/journal.pone.0277573. eCollection 2022.

Artificial intelligence for localization of the acute ischemic stroke by non-contrast computed tomography

Natsuda Kaothanthong  1 , Kamin Atsavasirilert  1 , Soawapot Sarampakhul  2 , Pantid Chantangphol  1 , Dittapong Songsaeng  2 , Stanislav Makhanov  1
Affiliations

Artificial intelligence for localization of the acute ischemic stroke by non-contrast computed tomography

Natsuda Kaothanthong et al. PLoS One. .

Abstract

A non-contrast cranial computer tomography (ncCT) is often employed for the diagnosis of the early stage of the ischemic stroke. However, the number of false negatives is high. More accurate results are obtained by an MRI. However, the MRI is not available in every hospital. Moreover, even if it is available in the clinic for the routine tests, emergency often does not have it. Therefore, this paper proposes an end-to-end framework for detection and segmentation of the brain infarct on the ncCT. The computer tomography perfusion (CTp) is used as the ground truth. The proposed ensemble model employs three deep convolution neural networks (CNNs) to process three end-to-end feature maps and a hand-craft features characterized by specific contra-lateral features. To improve the accuracy of the detected infarct area, the spatial dependencies between neighboring slices are employed at the postprocessing step. The numerical experiments have been performed on 18 ncCT-CTp paired stroke cases (804 image-pairs). The leave-one-out approach is applied for evaluating the proposed method. The model achieves 91.16% accuracy, 65.15% precision, 77.44% recall, 69.97% F1 score, and 0.4536 IoU.

PubMed Disclaimer

Conflict of interest statement

The authors have declared that no competing interests exist.

Figures

Fig 1
Fig 1. Examples of nCCT images, (a)-(d), and the corresponding CTp, (e)-(h).
The color shows the cerebral blood volume (CBV), cerebral blood flow (CBF), the mean transit time (MTT) generated by CTp, the ischemic area- blue, the dead tissue-red.
Fig 2
Fig 2. Three basic steps of the proposed model 1) label assignment, 2) feature map extraction, 3) infarct localization.
Fig 3
Fig 3. CTp with the ground truth, ncCT aligned with the CTp.
Fig 4
Fig 4. The architecture of the input for the proposed DNN.
Fig 5
Fig 5. Input for feature extraction (a) brain region on an ncCT slice (b) corresponding CTp, (c) the mask of the brain region.
(a) ncCT Slice. (b) Label. (c) Mask.
Fig 6
Fig 6. CTp and ncCT slices, and the result of aligning CTp with ncCT.
Fig 7
Fig 7. The feature maps.
(a) DeepLabv3+ with. (b) DeepLabv3+ with MobileNet ResNet50. (c) DeepLabv3+ with. (d) Pixel-wised Feature ResNet101 and Neural Network.
Fig 8
Fig 8. Localizing the infarct region using the extracted feature maps.
Fig 9
Fig 9. Precision and recall values of Ensemble3 (left) and Ensemble5 (right) with different classification threshold.
Fig 10
Fig 10. Accuracy (a), precision (b), recall (c), and F1 score (d) of Ensemble model with 3- and 5- blocks by MobileNet, ResNet101, and Resnet50.
(a) Accuracy of all model with (left) and without (right) post processing. (b) Precision of all model with (left) and without (right) post processing. (c) Recall of all model with (left) and without (right) post processing. (d) F1 score of all model with (left) and without (right) post processing.
Fig 11
Fig 11. Average IoU of regions obtained from each model using different classification threshold values with postproceessing (left) and without it (right).
Fig 12
Fig 12. Box plots of IoU for MobileNet, hand-crafted feature map, ResNet101 and Resnet50.
Ensemble model with 1, 3 and 5 extended output blocks. Each plot shows the median values. IoU of all model with (left) and without (right) post processing.
Fig 13
Fig 13
Lesion detection of the original ncCT (a) label on CTp (b) of the case that achieved the best segmentation result. (c) Ensemble3 (e) Ensemble5 without postproceessing (d) Ensemble3 (f) Ensemble5 with postproceessing.
Fig 14
Fig 14. Lesion detection (a) CTp, (b), (d) Ensemble3 and Ensemble5 without postproceessing, (c), (e) Ensemble3 and Ensemble5 with postprocessing.
Fig 15
Fig 15. Lesion detection (a) ncCT and its corresponding CTp (b) of the case that achieved the lowest result due to the failed postprocessing, (b), (d) Ensemble3 and Ensemble5 without postprocessing (c), (e)Ensemble3 and Ensemble5 with postproceessing.
Fig 16
Fig 16
Lesion detection of the label on CTp (a) that achieve the low result. The result by Ensemble3 and 5 without postproceessing is depicted in (b) and (d), respectively. The result with postproceessing is in (c) and (e) respectively.
See this image and copyright information in PMC

Similar articles

See all similar articles

Cited by

References

    1. Zhao H, Shi J, Qi X, Wang X, Jia J. Pyramid Scene Parsing Network. In: 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR); 2017. p. 6230–6239.
    1. Çiçek Ö, Abdulkadir A, Lienkamp SS, Brox T, Ronneberger O. 3D U-net: Learning dense volumetric segmentation from sparse annotation. In: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). vol. 9901 LNCS. Springer; Verlag; 2016. p. 424–432.
    1. Clèrigues A, Valverde S, Bernal J, Freixenet J, Oliver A, Lladó X. Acute ischemic stroke lesion core segmentation in CT perfusion images using fully convolutional neural networks. Computers in Biology and Medicine. 2019;115:103487. doi: 10.1016/j.compbiomed.2019.103487 - DOI - PubMed
    1. Subbanna NK, Rajashekar D, Cheng B, Thomalla G, Fiehler J, Arbel T, et al.. Stroke lesion segmentation in FLAIR MRI datasets using customized Markov random fields. Frontiers in Neurology. 2019;10(MAY). doi: 10.3389/fneur.2019.00541 - DOI - PMC - PubMed
    1. Milletari F, Navab N, Ahmadi SA. V-Net: Fully convolutional neural networks for volumetric medical image segmentation. In: Proceedings—2016 4th International Conference on 3D Vision, 3DV 2016. Institute of Electrical and Electronics Engineers Inc.; 2016. p. 565–571.

Publication types