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
. 2023 Jul 10;3(1):95.
doi: 10.1038/s43856-023-00327-4.

Automatic comprehensive radiological reports for clinical acute stroke MRIs

Chin-Fu Liu  1   2 , Yi Zhao  3 , Vivek Yedavalli  4 , Richard Leigh  5 , Vitor Falcao  6 ; STIR and VISTA Imaging investigators; Michael I Miller  1   2   7 , Argye E Hillis  5   8 , Andreia V Faria  9
Collaborators, Affiliations

Automatic comprehensive radiological reports for clinical acute stroke MRIs

Chin-Fu Liu et al. Commun Med (Lond). .

Abstract

Background: Although artificial intelligence systems that diagnosis among different conditions from medical images are long term aims, specific goals for automation of human-labor, time-consuming tasks are not only feasible but equally important. Acute conditions that require quantitative metrics, such as acute ischemic strokes, can greatly benefit by the consistency, objectiveness, and accessibility of automated radiological reports.

Methods: We used 1,878 annotated brain MRIs to generate a fully automated system that outputs radiological reports in addition to the infarct volume, 3D digital infarct mask, and the feature vector of anatomical regions affected by the acute infarct. This system is associated to a deep-learning algorithm for segmentation of the ischemic core and to parcellation schemes defining arterial territories and classically-identified anatomical brain structures.

Results: Here we show that the performance of our system to generate radiological reports was comparable to that of an expert evaluator. The weight of the components of the feature vectors that supported the prediction of the reports, as well as the prediction probabilities are outputted, making the pre-trained models behind our system interpretable. The system is publicly available, runs in real time, in local computers, with minimal computational requirements, and it is readily useful for non-expert users. It supports large-scale processing of new and legacy data, enabling clinical and translational research.

Conclusion: The generation of reports indicates that our fully automated system is able to extract quantitative, objective, structured, and personalized information from stroke MRIs.

Plain language summary

Artificial intelligence (AI) uses computer software to solve problems that normally require human input. It is likely that AI will take over, or help with, certain tasks in medical imaging, particularly where these tasks are time-consuming and laborious for clinicians. Here, we demonstrate the possibility of using AI to generate radiological reports for brain scans from patients who have had a stroke. These reports provide a summary of what is shown in the scans, and are normally written by clinicians. Our system performs similarly to human experts, is fast, publicly available, and runs on normal computers with minimal computational requirements, meaning that it might be a useful tool for researchers and clinicians to use when assessing and treating patients with stroke.

PubMed Disclaimer

Conflict of interest statement

The authors declare the following competing interests: Michael I. Miller owns “AnatomyWorks”. This arrangement is managed by Johns Hopkins University in accordance with its conflict-of-interest policies. The remaining authors declare no competing interests.

Figures

Fig. 1
Fig. 1. Data description and study design.
The flowchart describes data inclusion and exclusion, and of the design used for developing and testing of machine learning models.
Fig. 2
Fig. 2. Atlases defining the arterial territories (A) and brain structures (B) used in this study.
The regions of interest (ROIs) are overlaid in the template T1-WI.
Fig. 3
Fig. 3. Illustrative example of the infarct location prediction to generate automated radiological report.
The figures show a large acute ischemic infarct in DWI (a). The infarct core, automatically segmented, is overlaid in ADC (b). Brain atlases representing classical anatomical structures (c) and arterial territories (d) allow to quantify the injury in diverse regions of interest (ROIs). The calculated quantitative feature vectors (QFV) are in Table 2.
Fig. 4
Fig. 4. Feature importance, as revealed by the Mean Decrease in Impurity (MDI) of the Random Forest (RF) models (n = 1414).
The MDI is proportional to the importance of the features (the QFVs and lesion volume, in the x-axis) to predict the injury of the region in question (title of each graph). The QFVs represent the proportion of each ROI affected by the infarct. Note that the dominant QFV component agrees with the prediction of injury in the corresponding region and is followed by the QFV of its spatially neighboring regions.
Fig. 5
Fig. 5. Agreement between human and machine on the definition of infarct location in the testing set (n = 464).
Intraclass correlations (ICCs, y-axis) among human evaluators (E1, E2, E3) and among evaluators and our automated model for infarct location classification (auto). ICC = 1 is perfect agreement.
See this image and copyright information in PMC

References

    1. Hinton, G. On radiology. In Machine Learning and the Market for Intelligence Conference (Synced, Canada, 2016).
    1. Hinton G. Deep learning—a technology with the potential to transform health care. JAMA. 2018;320:1101–1102. doi: 10.1001/jama.2018.11100. - DOI - PubMed
    1. Oren O, Gersh BJ, Bhatt DL. Artificial intelligence in medical imaging: switching from radiographic pathological data to clinically meaningful endpoints. The Lancet Digit. Heal. 2020;2:e486–e488. doi: 10.1016/S2589-7500(20)30160-6. - DOI - PubMed
    1. Esteva A, et al. Dermatologist-level classification of skin cancer with deep neural networks. Nature. 2017;542:115–118. doi: 10.1038/nature21056. - DOI - PMC - PubMed
    1. Poplin R, et al. Prediction of cardiovascular risk factors from retinal fundus photographs via deep learning. Nat. Biomed. Eng. 2018;2:158–164. doi: 10.1038/s41551-018-0195-0. - DOI - PubMed