Assessing the utility of low resolution brain imaging: treatment of infant hydrocephalus

Joshua R Harper, Venkateswararao Cherukuri, Tom O'Reilly, Mingzhao Yu, Edith Mbabazi-Kabachelor, Ronald Mulando, Kevin N Sheth, Andrew G Webb, Benjamin C Warf, Abhaya V Kulkarni, Vishal Monga, Steven J Schiff, Joshua R Harper, Venkateswararao Cherukuri, Tom O'Reilly, Mingzhao Yu, Edith Mbabazi-Kabachelor, Ronald Mulando, Kevin N Sheth, Andrew G Webb, Benjamin C Warf, Abhaya V Kulkarni, Vishal Monga, Steven J Schiff

Abstract

As low-field MRI technology is being disseminated into clinical settings around the world, it is important to assess the image quality required to properly diagnose and treat a given disease and evaluate the role of machine learning algorithms, such as deep learning, in the enhancement of lower quality images. In this post hoc analysis of an ongoing randomized clinical trial, we assessed the diagnostic utility of reduced-quality and deep learning enhanced images for hydrocephalus treatment planning. CT images of post-infectious infant hydrocephalus were degraded in terms of spatial resolution, noise, and contrast between brain and CSF and enhanced using deep learning algorithms. Both degraded and enhanced images were presented to three experienced pediatric neurosurgeons accustomed to working in low- to middle-income countries (LMIC) for assessment of clinical utility in treatment planning for hydrocephalus. In addition, enhanced images were presented alongside their ground-truth CT counterparts in order to assess whether reconstruction errors caused by the deep learning enhancement routine were acceptable to the evaluators. Results indicate that image resolution and contrast-to-noise ratio between brain and CSF predict the likelihood of an image being characterized as useful for hydrocephalus treatment planning. Deep learning enhancement substantially increases contrast-to-noise ratio improving the apparent likelihood of the image being useful; however, deep learning enhancement introduces structural errors which create a substantial risk of misleading clinical interpretation. We find that images with lower quality than is customarily acceptable can be useful for hydrocephalus treatment planning. Moreover, low quality images may be preferable to images enhanced with deep learning, since they do not introduce the risk of misleading information which could misguide treatment decisions. These findings advocate for new standards in assessing acceptable image quality for clinical use.

Trial registration: ClinicalTrials.gov NCT01936272.

Keywords: Deep learning; Hydrocephalus treatment planning; Image quality; Low field MRI; Risk assessment.

Conflict of interest statement

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Copyright © 2021. Published by Elsevier Inc.

Figures

Fig. 1
Fig. 1
A comparison of the image quality between a high-field (3T) and a low-field (0.05 T) image of the brain of the same volunteer taken at the Leiden University Medical Center. A) A 256 × 256 3D T1 weighted TFE with Field of View: 200 × 175 × 156 mm, Resolution: 1.15 × 1.15 × 1.2 mm, TR/TE/TI  = 9.8 ms/4.6 ms/1050 ms, ETL  = 166, scan duration: 3 min 13 s; B) A 128 × 128 image at 0.05 T with Field of view: 256 × 256 × 200mm, Resolution: 2 × 2 × 4 mm, TR/TE  = 400 ms/15 ms, echo train length  = 6, scan duration: 7 min 7 s.
Fig. 2
Fig. 2
Schematic of study. In A) the image parameter space describing all possible combinations of noise, contrast between brain and CSF, and image resolution are visualized. There is likely to be a region of parameter combinations yielding images which are useful for hydrocephalus treatment planning (green volume), a region of parameter combinations that are not useful (red volume), and a region of uncertainty in between (orange volume). In B) we show a single plane from image parameter space in which all images have 512 × 512 resolution. The lower right corner has maximum contrast between brain and CSF and least noise considered in this study and the upper left corner has the lowest contrast and most noise. In C) the starred image from panel B) is chosen to be enhanced with a single encoder dual decoder (SEDD) architecture following the DenseNet network described in (Guo et al., 2019, Cherukuri et al., 2019). The output of such enhancement is seen in the upper panel of D) with corresponding segmentation in the lower panel of D). The ground truth version of the enhancement and segmentation from the original image without degradation or enhancement is shown in E) and called “ground truth”.
Fig. 3
Fig. 3
The figure shows results from Part 1 of the Assessment. In A) we show an example panel from Part 1 of the assessment. The lower left image is an enhanced image and all other images are degraded. The experts must indicate which (if any) is useful. The left panel of B) shows raw classification data from Part 1 for 64 × 64 images. Solid lines are lines of constant contrast-to-noise ratio (CNR). Dashed lines show lines of constant usefulness likelihood from the multivariate logistic regression. The right panel of B) shows the receiver operating characteristic curves. In C) we show the univariate logistic regression models for each resolution with CNR as the predictor. The diamond and circle datapoints show the calculated CNR values for the low-field and high-field MRI images shown in Fig. 1, respectively. Resolution for these images lie between the 128 × 128 and 512 × 512 curves, which overlap for the CNR values reported. The bottom four panels of C) show the raw classification data for each resolution.
Fig. 4
Fig. 4
The figure shows results from Part 2 of the assessment. A) An example panel from Part 2 of the assessment. The left column of images are ground truth and the right column are the enhanced versions. B) shows the usefulness likelihood curves based on image CNR. The triangles show the average CNR for each network location before enhancement and the circles show the average CNR for each network after enhancement. C) shows the predicted usefulness likelihood of the enhanced images based on CNR after enhancement, the actual Part 1 classification of the enhanced images, and the Part 2 re-classification of the enhanced images after comparison with ground truth. In D) we compare the usefulness likelihood of the degraded images with the risk of a misleading result if the image is enhanced for 128 × 128 images. The left vertical axis shows the usefulness likelihood of the degraded image and the right vertical axis shows the risk of a misleading result if the corresponding degraded image were enhanced. In D) we also show an example degraded image on the left with CNR  = 1, the enhanced version of this image on the right with CNR` = 8 after enhancement and corresponding high likelihood of misleading results after enhancement. Finally, E) shows the ground truth version of the example image in D) for comparison.

References

    1. Alexander D.C., Zikic D., Zhang J., Zhang H., Criminisi A. International Conference on Medical Image Computing and Computer-Assisted Intervention. Springer; 2014. Image quality transfer via random forest regression: applications in diffusion MRI; pp. 225–232.
    1. Brenner D.J., Eric J. Hall. Computed tomography–an increasing source of radiation exposure. N. Engl. J. Med. 2007;357(22):2277–2284.
    1. Byrt T., Bishop J., Carlin J.B. Bias, prevalence and kappa. J. Clinical Epidemiol. 1993;46(5):423–429.
    1. Chen, Y., Y. Xie, Z. Zhou, F. Shi, A.G. Christodoulou, and D. Li (2018) ”Brain MRI super resolution using 3D deep densely connected neural networks,” in 2018 IEEE 15th International Symposium on Biomedical Imaging (ISBI 2018), IEEE, pp. 739–742.
    1. Cherukuri V., Ssenyonga P., Warf B.C., Kulkarni A.V., Monga V., Schiff S.J. Learning based segmentation of CT brain images: application to postoperative hydrocephalic scans. IEEE Trans. Biomed. Eng. 2017;65(8):1871–1884.
    1. Cherukuri V., Guo T., Schiff S.J., Monga V. Deep MR Brain Image Super-Resolution Using Spatio-Structural Priors. IEEE Trans. Image Process. 2019;29:1368–1383.
    1. Cooley C.Z., McDaniel P.C., Stockmann J.P., Srinivas S.A., Cauley S.F., Śliwiak M., Sappo C.R., Vaughn C.F., Guerin B., Rosen M.S., et al. A portable scanner for magnetic resonance imaging of the brain. Nature Biomed. Eng. 2021;5(3):229–239.
    1. Dewan M.C., Rattani A., Mekary R., Glancz L.J., Yunusa I., Baticulon R.E., Fieggen G., Wellons J.C., Park K.B., Warf B.C. Global hydrocephalus epidemiology and incidence: systematic review and meta-analysis. J. Neurosurg. 2018;130(4):1065–1079.
    1. Diwakar M., Kumar M. A review on CT image noise and its denoising. Biomed. Signal Process. Control. 2018;42:73–88.
    1. Durand D.J., Carrino J.A., Fayad L.M., Huisman T.A., El-Sharkawy A.-M.M., Edelstein W.A. MRI pyschophysics: An experimental framework relating image quality to diagnostic performance metrics. J. Magn. Reson. Imaging. 2013;37(6):1402–1408.
    1. Frush D.P., Donnelly L.F., Rosen N.S. Computed tomography and radiation risks: what pediatric health care providers should know. Pediatrics. 2003;112(4):951–957.
    1. Gatrad A., Gatrad S., Gatrad A. Equipment donation to developing countries. Anaesthesia. 2007;62:90–95.
    1. Guo T., Cherukuri V., Monga V. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. 2019. Dense123’color enhancement dehazing network.
    1. Guo T., Li X., Cherukuri V., Monga V. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops. 2019. Dense scene information estimation network for dehazing.
    1. Hallgren K.A. Computing inter-rater reliability for observational data: an overview and tutorial. Tutorials Quantitative Methods Psychol. 2012;8(1):23.
    1. Jack J.C., Berquist T.H., Miller G.M., Forbes G.S., Gray J.E., Morin R.L., Ilstrup D.M. Field strength in neuro-MR imaging: a comparison of 0.5 T and 1.5 T. J. Computer Assisted Tomography. 1990;14(4):505–513.
    1. Jhaveri K. Image quality versus outcomes. J. Magn. Reson. Imaging. 2015;41(4):866–869.
    1. Jia Y., Gholipour A., He Z., Warfield S.K. A new sparse representation framework for reconstruction of an isotropic high spatial resolution MR volume from orthogonal anisotropic resolution scans. IEEE Trans. Med. Imaging. 2017;36(5):1182–1193.
    1. Klein H.-M. Springer; 2015. clinical low field strength magnetic resonance imaging: a practical guide to accessible MRI.
    1. Kulkarni A.V., Schiff S.J., Mbabazi-Kabachelor E., Mugamba J., Ssenyonga P., Donnelly R., Levenbach J., Monga V., Peterson M., MacDonald M., et al. Endoscopic treatment versus shunting for infant hydrocephalus in Uganda. N. Engl. J. Med. 2017;377(25):2456–2464.
    1. Lee D.H., Vellet A.D., Eliasziw M., Vidito L., Ebers G.C., Rice G.P., Hewett L., Dunlavy S. MR imaging field strength: prospective evaluation of the diagnostic accuracy of MR for diagnosis of multiple sclerosis at 0.5 and 1.5 T. Radiology. 1995;194(1):257–262.
    1. Lehmann T.M., Gonner C., Spitzer K. Survey: Interpolation methods in medical image processing. IEEE Trans. Med. Imaging. 1999;18(11):1049–1075.
    1. Malkin R.A. “Design of health care technologies for the developing world, “ Annu. Rev. Biomed. Eng. 2007;9:567–587.
    1. Manjón J.V., Coupé P., Buades A., Fonov V., Collins D.L., Robles M. Non-local MRI upsampling. Med. Image Anal. 2010;14(6):784–792.
    1. Manjón, J.V., P. Coupé, A. Buades, D.L. Collins, and M. Robles (2010) ”MRI superresolution using self-similarity and image priors,” International journal of biomedical imaging, 2010.
    1. Mazurek M., Cahn B., Yuen Mercy, Cahn Bradley, Yuen Matthew, Prabhat Anjali, Chavva Isha, Shah Jill, Crawford Anna, Welch E., Rothburg Jonathan, Sacolick Laura, Poole Michael, Wira Charles, Matouk Charles, Ward Adrienne, Timario Nona, Leasure Audrey, Beekman Rachel, Peng Teng, Witsch Jens, Antonios Joseph, Falcone Guido, Gobeske Kevin, Petersen Nils, Schindler Joseph, Sansing Lauren, Gilmore Emily, Hwang David, Kim Jennifer, Malhotra Ajay, Sze Gordon, Rosen Matthew, Kimberly Taylor, Sheth Kevin. Portable, bedside, low-field magnetic resonance imaging for evaluation of intracerebral hemorrhage. Nat. Commun. 2021 doi: 10.1038/s41467-021-25441-6.
    1. Obungoloch J., Harper J.R., Consevage S., Savukov I.M., Neuberger T., Tadigadapa S., Schiff S.J. Design of a sustainable prepolarizing magnetic resonance imaging system for infant hydrocephalus. Magn. Reson. Mater. Phys., Biol. Med. 2018;31(5):665–676.
    1. O’Reilly T., Teeuwisse W.M., de Gans D., Koolstra K., Webb A.G. In vivo 3D brain and extremity MRI at 50 mT using a permanent magnet Halbach array. Magn. Reson. Med. 2020
    1. Orrison W., Jr, Stimac G., Stevens E., LaMasters D., Espinosa M., Cobb L., Mettler F., Jr Comparison of CT, low-field-strength MR imaging, and high-field-strength MR imaging. Work in progress. Radiology. 1991;181(1):121–127.
    1. Paulson J.N., Williams B.L., Hehnly C., Mishra N., Sinnar S.A., Zhang L., Ssentongo P., Mbabazi-Kabachelor E., Wijetunge D.S., von Bredow B., et al. The Bacterial and Viral Complexity of Postinfectious Hydrocephalus in Uganda. Sci. Transl. Med. 2020;12(563)
    1. Pham, C.-H., A. Ducournau, R. Fablet, and F. Rousseau (2017) ”Brain MRI super-resolution using deep 3D convolutional networks,” in 2017 IEEE 14th International Symposium on Biomedical Imaging (ISBI 2017), IEEE, pp. 197–200.
    1. Rutland J., Delman B., Gill C., Zhu C., Shrivastava R., Balchandani P. Emerging Use of Ultra-High-Field 7T MRI in the Study of Intracranial Vascularity: State of the Field and Future Directions. Am. J. Neuroradiol. 2020;41(1):2–9.
    1. Schiff, S., A. Kulkarni, K.E. Mbabazi, J. Mugamba, P. Ssenyonga, R. Donnelly, J. Levenbach, V. Monga, M. Peterson, V. Cherukuri, and B. Warf (2021) ”Brain growth after surgical treatment of infant post-infectious hydrocephalus in sub-Saharan Africa: two-year results of a randomized trial.” Journal of Neurosurgery Pediatrics.
    1. Sheth, K.N., M.H. Mazurek, M.M. Yuen, B.A. Cahn, J.T. Shah, A. Ward, J.A. Kim, E.J. Gilmore, G.J. Falcone, N. Petersen, et al. (2020) ”Assessment of Brain Injury Using Portable, Low-Field Magnetic Resonance Imaging at the Bedside of Critically Ill Patients,” JAMA neurology.
    1. Shi F., Cheng J., Wang L., Yap P.-T., Shen D. LRTV: MR image super-resolution with low-rank and total variation regularizations. IEEE Trans. Med. Imaging. 2015;34(12):2459–2466.
    1. Steinberg H., Alarcon J., Bernardino M. Focal hepatic lesions: comparative MR imaging at 0.5 and 1.5 T. Radiology. 1990;174(1):153–156.
    1. Wang Y.-H., Qiao J., Li J.-B., Fu P., Chu S.-C., Roddick J.F. Sparse representation-based MRI super-resolution reconstruction. Measurement. 2014;47:946–953.
    1. Warf B.C. Educate one to save a few. Educate a few to save many. World Neurosurg. 2013;79(2):S15–e15.
    1. World Health Organization Baseline country survey on medical devices 2010. Tech. rep., World Health Organization. 2011
    1. Yang J., Wright J., Huang T.S., Ma Y. Image super-resolution via sparse representation. IEEE Trans. Image Process. 2010;19(11):2861–2873.
    1. Zhu B., Liu J.Z., Cauley S.F., Rosen B.R., Rosen M.S. Image reconstruction by domain-transform manifold learning. Nature. 2018;555(7697):487–492.

Source: PubMed

Sponzoři a spolupracovníci

Zdravotní podmínky

Drogové intervence

3
Předplatit