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 Nov;18(11):2001-2012.
doi: 10.1007/s11548-023-02965-4. Epub 2023 May 29.

Automated screening of computed tomography using weakly supervised anomaly detection

Atsuhiro Hibi  1   2 , Michael D Cusimano  3 , Alexander Bilbily  2   4 , Rahul G Krishnan  5   6 , Pascal N Tyrrell  7   8   9
Affiliations

Automated screening of computed tomography using weakly supervised anomaly detection

Atsuhiro Hibi et al. Int J Comput Assist Radiol Surg. 2023 Nov.

Abstract

Background: Current artificial intelligence studies for supporting CT screening tasks depend on either supervised learning or detecting anomalies. However, the former involves a heavy annotation workload owing to requiring many slice-wise annotations (ground truth labels); the latter is promising, but while it reduces the annotation workload, it often suffers from lower performance. This study presents a novel weakly supervised anomaly detection (WSAD) algorithm trained based on scan-wise normal and anomalous annotations to provide better performance than conventional methods while reducing annotation workload.

Methods: Based on surveillance video anomaly detection methodology, feature vectors representing each CT slice were trained on an AR-Net-based convolutional network using a dynamic multiple-instance learning loss and a center loss function. The following two publicly available CT datasets were retrospectively analyzed: the RSNA brain hemorrhage dataset (normal scans: 12,862; scans with intracranial hematoma: 8882) and COVID-CT set (normal scans: 282; scans with COVID-19: 95).

Results: Anomaly scores of each slice were successfully predicted despite inaccessibility to any slice-wise annotations. Slice-level area under the curve (AUC), sensitivity, specificity, and accuracy from the brain CT dataset were 0.89, 0.85, 0.78, and 0.79, respectively. The proposed method reduced the number of annotations in the brain dataset by 97.1% compared to an ordinary slice-level supervised learning method.

Conclusion: This study demonstrated a significant annotation reduction in identifying anomalous CT slices compared to a supervised learning approach. The effectiveness of the proposed WSAD algorithm was verified through higher AUC than existing anomaly detection techniques.

Keywords: Anomaly detection; Artificial intelligence; COVID-19; Computed tomography; Machine learning; Traumatic brain injury.

PubMed Disclaimer

Conflict of interest statement

AH is supported by a funding from Nippon Steel Corporation. AB is an officer and shareholder of 16 Bit Inc., and a consultant for Roche. RGK is on the Scientific Advisory Board of Iterative Scopes. PNT is an investigator and consultant of Novo Nordisk, an officer, director and shareholder of SofTx Innovations Inc. and MDC has nothing to disclose.

Figures

Fig. 1
Fig. 1
Pre-processing of CT dataset
Fig. 2
Fig. 2
Outline of the proposed algorithm
Fig. 3
Fig. 3
Anomaly scores and corresponding CT slices in brain hemorrhage CT dataset. Blue and red ranges indicate slices with normal and anomalous annotations, respectively. GT indicates slice-wise ground truth annotated by medical experts
Fig. 4
Fig. 4
Anomaly scores and corresponding CT slices in the COVID-19 lung CT dataset. Blue and red ranges indicate slices with normal and anomalous annotations, respectively. GT indicates slice-wise ground truth annotated by medical experts
Fig. 5
Fig. 5
Anomaly scores of CT scans with ICH estimated by various methods. Blue and red ranges indicate slices with normal and anomalous annotations, respectively
Fig. 6
Fig. 6
Slice-level prediction performance as a function of number of annotations evaluated on the testing dataset
See this image and copyright information in PMC

References

    1. Jonas DE, Reuland DS, Reddy SM, Nagle M, Clark SD, Weber RP, Enyioha C, Malo TL, Brenner AT, Armstrong C, Coker-Schwimmer M, Middleton JC, Voisin C, Harris RP. Screening for lung cancer with low-dose computed tomography: updated evidence report and systematic review for the US Preventive Services Task Force. J Am Med Assoc: JAMA. 2021;325:971–987. doi: 10.1001/jama.2021.0377. - DOI - PubMed
    1. Dai WC, Zhang HW, Yu J, Xu HJ, Chen H, Luo SP, Zhang H, Liang LH, Wu XL, Lei Y, Lin F. CT imaging and differential diagnosis of COVID-19. Can Assoc Radiol J. 2020;71:195–200. doi: 10.1177/0846537120913033. - DOI - PMC - PubMed
    1. Granacher RP. Traumatic brain injury: methods for clinical and forensic neuropsychiatric assessment. Boca Raton: CRC Press; 2015.
    1. Bruls RJM, Kwee RM. Workload for radiologists during on-call hours: dramatic increase in the past 15 years. Insights Imaging. 2020;11:121. doi: 10.1186/s13244-020-00925-z. - DOI - PMC - PubMed
    1. Mushtaq MF, Shahroz M, Aseere AM, Shah H, Majeed R, Shehzad D, Samad A. BHCNet: neural network-based brain hemorrhage classification using head CT scan. IEEE Access. 2021;9:113901–113916. doi: 10.1109/ACCESS.2021.3102740. - DOI