Automatic segmentation of cerebral infarcts in follow-up computed tomography images with convolutional neural networks

Abstract

Background and purposeInfarct volume is a valuable outcome measure in treatment trials of acute ischemic stroke and is strongly associated with functional outcome. Its manual volumetric assessment is, however, too demanding to be implemented in clinical practice.ObjectiveTo assess the value of convolutional neural networks (CNNs) in the automatic segmentation of infarct volume in follow-up CT images in a large population of patients with acute ischemic stroke.Materials and methodsWe included CT images of 1026 patients from a large pooling of patients with acute ischemic stroke. A reference standard for the infarct segmentation was generated by manual delineation. We introduce three CNN models for the segmentation of subtle, intermediate, and severe hypodense lesions. The fully automated infarct segmentation was defined as the combination of the results of these three CNNs. The results of the three-CNNs approach were compared with the results from a single CNN approach and with the reference standard segmentations.ResultsThe median infarct volume was 48 mL (IQR 15–125 mL). Comparison between the volumes of the three-CNNs approach and manually delineated infarct volumes showed excellent agreement, with an intraclass correlation coefficient (ICC) of 0.88. Even better agreement was found for severe and intermediate hypodense infarcts, with ICCs of 0.98 and 0.93, respectively. Although the number of patients used for training in the single CNN approach was much larger, the accuracy of the three-CNNs approach strongly outperformed the single CNN approach, which had an ICC of 0.34.ConclusionConvolutional neural networks are valuable and accurate in the quantitative assessment of infarct volumes, for both subtle and severe hypodense infarcts in follow-up CT images. Our proposed three-CNNs approach strongly outperforms a more straightforward single CNN approach.

Automatic Segmentation of Stroke Lesions in Non-Contrast Computed Tomography Datasets With Convolutional Neural Networks

“…This was also reflected in the fact that manual segmentations also showed less agreement for smaller lesions. A similar result was seen for CNN segmentations of NCCT images by Barros et al, where the DSC scores were lower for subtle injuries with smaller stroke lesions (0.37) [5].…”

“…Other recent studies on automatic NCCT segmentation of stroke lesions have used single-scale CNN models [5], [6], reporting mean DSC scores (0.54-0.57) and volumetric ICCs (0.88) comparable to ours (DSC = 0.45, ICC = 0.88). However, these studies only evaluated their models on test data that belonged to the same distribution as their training set and were only compared against single reference segmentations [5], [6]. How these methods generalize to datasets outside their training distribution and against multiple expert observers, important for evaluating their broader applicability, was not evaluated in detail in the original studies.…”

“…Third, the training of deep learning models is stochastic in nature. Future work may investigate ensembling multiple models trained on subsets of the training data, as this may allow averaging out deficiencies in single models [5], [10], [24]. Addressing these limitations may further improve CNN-based lesion segmentations in NCCT datasets.…”

“…The associate editor coordinating the review of this manuscript and approving it for publication was Wei Wei . require human interaction, which could introduce a bias, while previous work on automatic NCCT lesion segmentation is limited [5], [6].…”

Ischemic infarct detection, localization, and segmentation in noncontrast CT human brain scans: review of automated methods

“… Sales Barros et al (2019) proposed an infarct segmentation method utilizing CNN deep learning. The goal was to segment an infarct to calculate its volume in follow-up NCCT scans acquired between 12 h and 2 weeks after stroke onset.…”