Rethinking Retinal Image Quality: Treating Quality Threshold as a Tunable Hyperparameter

Fabian Yii; Raman Dutt; Tom MacGillivray; Baljean Dhillon; Miguel Bernabeu; Niall Strang

doi:10.1007/978-3-031-16525-2_8

Rethinking Retinal Image Quality: Treating Quality Threshold as a Tunable Hyperparameter

Fabian Yii^*, Raman Dutt, Tom MacGillivray, Baljean Dhillon, Miguel Bernabeu, Niall Strang

^*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract / Description of output

Assuming the robustness of a deep learning model to suboptimal images is a key consideration, we asked if there was any value in including training images of poor quality. In particular, should we treat the (quality) threshold at which a training image is either included or excluded as a tunable hyperparameter? To that end, we systematically examined the effect of including training images of varying quality on the test performance of a DL model in classifying the severity of diabetic retinopathy. We found that there was a unique combination of (categorical) quality labels or a Goldilocks (continuous) quality score that gave rise to optimal test performance on either high-quality or suboptimal images. The model trained exclusively on high-quality images yielded worse performance in all test scenarios than that trained on the optimally tuned training set which included images with some level of degradation.

Original language	English
Title of host publication	Ophthalmic Medical Image Analysis
Subtitle of host publication	9th International Workshop, OMIA 2022, Held in Conjunction with MICCAI 2022, Singapore, Singapore, September 22, 2022, Proceedings
Editors	Bhavna Antony, Huazhu Fu, Cecilia S. Lee, Tom MacGillivray, Yanwu Xu, Yalin Zheng
Publisher	Springer
Pages	73-83
Number of pages	11
ISBN (Electronic)	9783031165252
ISBN (Print)	9783031165245
DOIs	https://doi.org/10.1007/978-3-031-16525-2_8
Publication status	Published - 15 Sept 2022
Event	9th International Workshop on Ophthalmic Medical Image Analysis, OMIA 2022, held in conjunction with the 25th International Conference on Medical Imaging and Computer-Assisted Intervention, MICCAI 2022 - Singapore, Singapore Duration: 22 Sept 2022 → 22 Sept 2022

Publication series

Name	Lecture Notes in Computer Science
Publisher	Springer
Volume	13576
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	9th International Workshop on Ophthalmic Medical Image Analysis, OMIA 2022, held in conjunction with the 25th International Conference on Medical Imaging and Computer-Assisted Intervention, MICCAI 2022
Country/Territory	Singapore
City	Singapore
Period	22/09/22 → 22/09/22

Keywords / Materials (for Non-textual outputs)

Deep learning
Image quality
Tunable hyperparameter

Access to Document

10.1007/978-3-031-16525-2_8

camera_ready_manuscriptAccepted author manuscript, 2.87 MB

Cite this

Yii, F., Dutt, R., MacGillivray, T., Dhillon, B., Bernabeu, M., & Strang, N. (2022). Rethinking Retinal Image Quality: Treating Quality Threshold as a Tunable Hyperparameter. In B. Antony, H. Fu, C. S. Lee, T. MacGillivray, Y. Xu, & Y. Zheng (Eds.), Ophthalmic Medical Image Analysis: 9th International Workshop, OMIA 2022, Held in Conjunction with MICCAI 2022, Singapore, Singapore, September 22, 2022, Proceedings (pp. 73-83). (Lecture Notes in Computer Science ; Vol. 13576). Springer. https://doi.org/10.1007/978-3-031-16525-2_8

Yii, Fabian ; Dutt, Raman ; MacGillivray, Tom et al. / Rethinking Retinal Image Quality : Treating Quality Threshold as a Tunable Hyperparameter. Ophthalmic Medical Image Analysis: 9th International Workshop, OMIA 2022, Held in Conjunction with MICCAI 2022, Singapore, Singapore, September 22, 2022, Proceedings. editor / Bhavna Antony ; Huazhu Fu ; Cecilia S. Lee ; Tom MacGillivray ; Yanwu Xu ; Yalin Zheng. Springer, 2022. pp. 73-83 (Lecture Notes in Computer Science ).

@inproceedings{871e731303a84e3d8e7e71820e35160f,

title = "Rethinking Retinal Image Quality: Treating Quality Threshold as a Tunable Hyperparameter",

abstract = "Assuming the robustness of a deep learning model to suboptimal images is a key consideration, we asked if there was any value in including training images of poor quality. In particular, should we treat the (quality) threshold at which a training image is either included or excluded as a tunable hyperparameter? To that end, we systematically examined the effect of including training images of varying quality on the test performance of a DL model in classifying the severity of diabetic retinopathy. We found that there was a unique combination of (categorical) quality labels or a Goldilocks (continuous) quality score that gave rise to optimal test performance on either high-quality or suboptimal images. The model trained exclusively on high-quality images yielded worse performance in all test scenarios than that trained on the optimally tuned training set which included images with some level of degradation.",

keywords = "Deep learning, Image quality, Tunable hyperparameter",

author = "Fabian Yii and Raman Dutt and Tom MacGillivray and Baljean Dhillon and Miguel Bernabeu and Niall Strang",

note = "Funding Information: F. Yii and R. Dutt–Contributed equally to this work. F. Yii is supported by the Medical Research Council [grant number MR/N013166/1]. Publisher Copyright: {\textcopyright} 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.; 9th International Workshop on Ophthalmic Medical Image Analysis, OMIA 2022, held in conjunction with the 25th International Conference on Medical Imaging and Computer-Assisted Intervention, MICCAI 2022 ; Conference date: 22-09-2022 Through 22-09-2022",

year = "2022",

month = sep,

day = "15",

doi = "10.1007/978-3-031-16525-2_8",

language = "English",

isbn = "9783031165245",

series = "Lecture Notes in Computer Science ",

publisher = "Springer",

pages = "73--83",

editor = "Bhavna Antony and Huazhu Fu and Lee, {Cecilia S.} and Tom MacGillivray and Yanwu Xu and Yalin Zheng",

booktitle = "Ophthalmic Medical Image Analysis",

address = "United Kingdom",

}

Yii, F, Dutt, R, MacGillivray, T , Dhillon, B , Bernabeu, M & Strang, N 2022, Rethinking Retinal Image Quality: Treating Quality Threshold as a Tunable Hyperparameter. in B Antony, H Fu, CS Lee, T MacGillivray, Y Xu & Y Zheng (eds), Ophthalmic Medical Image Analysis: 9th International Workshop, OMIA 2022, Held in Conjunction with MICCAI 2022, Singapore, Singapore, September 22, 2022, Proceedings. Lecture Notes in Computer Science , vol. 13576, Springer, pp. 73-83, 9th International Workshop on Ophthalmic Medical Image Analysis, OMIA 2022, held in conjunction with the 25th International Conference on Medical Imaging and Computer-Assisted Intervention, MICCAI 2022, Singapore, Singapore, 22/09/22. https://doi.org/10.1007/978-3-031-16525-2_8

Rethinking Retinal Image Quality: Treating Quality Threshold as a Tunable Hyperparameter. / Yii, Fabian; Dutt, Raman; MacGillivray, Tom et al.
Ophthalmic Medical Image Analysis: 9th International Workshop, OMIA 2022, Held in Conjunction with MICCAI 2022, Singapore, Singapore, September 22, 2022, Proceedings. ed. / Bhavna Antony; Huazhu Fu; Cecilia S. Lee; Tom MacGillivray; Yanwu Xu; Yalin Zheng. Springer, 2022. p. 73-83 (Lecture Notes in Computer Science ; Vol. 13576).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

TY - GEN

T1 - Rethinking Retinal Image Quality

T2 - 9th International Workshop on Ophthalmic Medical Image Analysis, OMIA 2022, held in conjunction with the 25th International Conference on Medical Imaging and Computer-Assisted Intervention, MICCAI 2022

AU - Yii, Fabian

AU - Dutt, Raman

AU - MacGillivray, Tom

AU - Dhillon, Baljean

AU - Bernabeu, Miguel

AU - Strang, Niall

N1 - Funding Information: F. Yii and R. Dutt–Contributed equally to this work. F. Yii is supported by the Medical Research Council [grant number MR/N013166/1]. Publisher Copyright: © 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.

PY - 2022/9/15

Y1 - 2022/9/15

N2 - Assuming the robustness of a deep learning model to suboptimal images is a key consideration, we asked if there was any value in including training images of poor quality. In particular, should we treat the (quality) threshold at which a training image is either included or excluded as a tunable hyperparameter? To that end, we systematically examined the effect of including training images of varying quality on the test performance of a DL model in classifying the severity of diabetic retinopathy. We found that there was a unique combination of (categorical) quality labels or a Goldilocks (continuous) quality score that gave rise to optimal test performance on either high-quality or suboptimal images. The model trained exclusively on high-quality images yielded worse performance in all test scenarios than that trained on the optimally tuned training set which included images with some level of degradation.

AB - Assuming the robustness of a deep learning model to suboptimal images is a key consideration, we asked if there was any value in including training images of poor quality. In particular, should we treat the (quality) threshold at which a training image is either included or excluded as a tunable hyperparameter? To that end, we systematically examined the effect of including training images of varying quality on the test performance of a DL model in classifying the severity of diabetic retinopathy. We found that there was a unique combination of (categorical) quality labels or a Goldilocks (continuous) quality score that gave rise to optimal test performance on either high-quality or suboptimal images. The model trained exclusively on high-quality images yielded worse performance in all test scenarios than that trained on the optimally tuned training set which included images with some level of degradation.

KW - Deep learning

KW - Image quality

KW - Tunable hyperparameter

UR - http://www.scopus.com/inward/record.url?scp=85138778121&partnerID=8YFLogxK

U2 - 10.1007/978-3-031-16525-2_8

DO - 10.1007/978-3-031-16525-2_8

M3 - Conference contribution

AN - SCOPUS:85138778121

SN - 9783031165245

T3 - Lecture Notes in Computer Science

SP - 73

EP - 83

BT - Ophthalmic Medical Image Analysis

A2 - Antony, Bhavna

A2 - Fu, Huazhu

A2 - Lee, Cecilia S.

A2 - MacGillivray, Tom

A2 - Xu, Yanwu

A2 - Zheng, Yalin

PB - Springer

Y2 - 22 September 2022 through 22 September 2022

ER -

Yii F, Dutt R, MacGillivray T , Dhillon B , Bernabeu M, Strang N. Rethinking Retinal Image Quality: Treating Quality Threshold as a Tunable Hyperparameter. In Antony B, Fu H, Lee CS, MacGillivray T, Xu Y, Zheng Y, editors, Ophthalmic Medical Image Analysis: 9th International Workshop, OMIA 2022, Held in Conjunction with MICCAI 2022, Singapore, Singapore, September 22, 2022, Proceedings. Springer. 2022. p. 73-83. (Lecture Notes in Computer Science ). doi: 10.1007/978-3-031-16525-2_8

Rethinking Retinal Image Quality: Treating Quality Threshold as a Tunable Hyperparameter

Abstract / Description of output

Publication series

Conference

Keywords / Materials (for Non-textual outputs)

Access to Document

Other files and links

Fingerprint

Image Analysis Core

Cite this

Rethinking Retinal Image Quality: Treating Quality Threshold as a Tunable Hyperparameter

Abstract / Description of output

Publication series

Conference

Keywords / Materials (for Non-textual outputs)

Access to Document

Other files and links

Fingerprint

Equipment

Image Analysis Core

Cite this