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
. 2024 Nov 21;19(11):e0310305.
doi: 10.1371/journal.pone.0310305. eCollection 2024.

Investigation of the effectiveness of no-reference metric in image evaluation in nuclear medicine

Shigeaki Higashiyama  1 , Yutaka Katayama  2 , Atsushi Yoshida  1 , Nahoko Inoue  3 , Takashi Yamanaga  2 , Takao Ichida  2 , Yukio Miki  3 , Joji Kawabe  1
Affiliations

Investigation of the effectiveness of no-reference metric in image evaluation in nuclear medicine

Shigeaki Higashiyama et al. PLoS One. .

Abstract

Background: In nuclear medicine, normalized mean square error (NMSE) is widely used for image quality evaluation and machine adjustment. However, evaluating clinical images in nuclear medicine using NMSE necessitates acquiring a reference image, which is time consuming and impractical. Therefore, it is necessary to explore no-reference metrics, such as perception-based image quality evaluator (PIQE) and natural image quality evaluator (NIQE), as alternatives for evaluating the quality of clinical images used in nuclear medicine.

Purpose: To examine whether no-reference metrics can be applied to image quality evaluations for clinical images in nuclear medicine.

Methods: Images of the Hoffman Brain Phantom containing 18F-fluoro-2-deoxy-D-glucose (FDG) were obtained using Biograph Vision (Siemens Co., Ltd). From the collected images, 14 images with varying pixel counts and acquisition times were created. Sixteen images were visually evaluated by five image experts and ranked accordingly. Image quality was assessed using NMSE, PIQE, and NIQE, and rankings were calculated based on these scores.

Results: The Spearman's significance test revealed a strong correlation between image quality evaluations using PIQE and visual evaluations by specialists (p<0.0001). PIQE demonstrated comparable performance to image experts in evaluating image quality, suggesting its potential for clinical image quality assessment in nuclear medicine.

Conclusions: PIQE offers a viable method for evaluating image quality in nuclear medicine, presenting a promising alternative to traditional visual inspection methods.

PubMed Disclaimer

Conflict of interest statement

The authors have declared that no competing interests exist.

Figures

Fig 1
Fig 1. A Hoffman 3D brain phantom with 26 MBq of 18F-fluoro-2-deoxy-D-glucose (FDG) with a 440-pixel matrix were obtained over 1800 s.
One slice of the axial image that depicts the frontal and temporal lobes, bilateral lateral ventricles, and basal ganglia was selected from the acquired brain phantom images. Images with different collection times (120, 180, 300, 360, 450, 600, and 900 s) were prepared with 326x188 pixels and RGB with 239K. A total of eight image types with different collection times and color scale are shown.
Fig 2
Fig 2. 880-pixel images corresponding to each of the eight imaging times and color scale are shown.
Images with different collection times were prepared with 326x188 pixels and RGB with 239K.
Fig 3
Fig 3. An image presented to evaluators.
A 440-pixel at 180 s image and a 880-pixel at 900 s image are displayed on the left and right sides, respectively. An image presented to evaluators. On the left is a 440-pixel at 180 s image, and on the right is a 880-pixel at 900 s image. This image corresponds to square number 29 shown in Fig 2.
Fig 4
Fig 4. Evaluation sheet for the images (out of 240 images) presented to the evaluator.
When the evaluator records the score, the cell in Fig 2 is blank, and the score is entered for the left and right images compared using the pairwise method. The rating is the total score in the leftmost column and bottom row.
Fig 5
Fig 5. Block diagram of the proposed method.
The input image was subjected to a preprocessing step. A block-level analysis was performed to identify the distortion, and each distorted block was assigned a score based on the distortion type. The block-level scores were pooled to determine the overall image quality.
Fig 6
Fig 6. Chart of region of interest (ROI) set to evaluate uniformity.
The arrow indicates the location of the ROI in the frontal lobe, the double arrow indicates the location of the ROI in the temporal lobe, and the arrowhead indicates the location of the ROI in the occipital lobe. Referring to previous literature, the size of the ROI was set to 5 mm in diameter.
Fig 7
Fig 7. Correlation between the visual assessment and PIQE rankings.
PIQE: Perception-based image quality evaluator. Spearman’s significant difference test between the visual assessment and PIQE rankings revealed a rs of 0.9559 (p < 0.0001), indicating a strong correlation.
Fig 8
Fig 8. Significant difference examined in the ranking of PIQE.
The difference between the top and bottom ranks from 1st to 16th. The difference between 1st and 4th place was group 1, the difference between 5th and 12th place was group 2, and the difference between 13th and 16th place was group 3. There was no significant difference among the three groups.
Fig 9
Fig 9. Correlation between the visual assessment and NIQE rankings.
NIQE: natural image quality evaluator. Spearman’s significant difference test between the visual assessment and NIQE rankings revealed a rs of 0.2324 (p 0.3865), indicating no strong correlation between the two methods.
Fig 10
Fig 10. Graphical display of the results of Table 9.
Fig 11
Fig 11. Graphical display of the results of Table 10.
Fig 12
Fig 12. Graphical display of the results of Table 11.
See this image and copyright information in PMC

References

    1. Ueda D., Katayama Y., Yamamoto A., Ichinose T., Arima H., Watanabe Y., et al.. Deep Learning–based angiogram generation model for cerebral angiography without misregistration artifacts. Radiology. 2021; 299(3): 675–681. doi: 10.1148/radiol.2021203692 - DOI - PubMed
    1. Zhu X., Cheng Y., Peng J., Wang R., Le M., & Liu X. Super-Resolved Image Perceptual Quality Improvement via Multi-Feature Discriminators. Computer Vision and Pattern Recognition. 2019. doi: 10.1117/1.JEI.29.1.013017 - DOI
    1. Blau Y., Mechrez R., Timofte R., Michaeli T., & Zelnik-Manor L. The 2018 PIRM Challenge on Perceptual Image Super-resolution. arXiv:1809.07517v3 [cs.CV]. 2019.
    1. Blau Y., & Michaeli T. The Perception-Distortion Tradeoff. arXiv:1711.06077v4 [cs.CV]. 2020. doi: 10.1109/CVPR.2018.00652 - DOI
    1. Yu Z., Rahman M. A., Schindler T. H., Gropler R. J., Laforest R., Wahl R. L., et al.. Need for Objective Task-based Evaluation of Deep Learning-Based Denoising Methods: A Study in the Context of Myocardial Perfusion SPECT. Journal of Nuclear Medicine. 2020; 61(supplement 1): 575. doi: 10.48550/arXiv.2303.02110 - DOI - PMC - PubMed

MeSH terms

Substances

LinkOut - more resources