An interpretable and interactive deep learning algorithm for a clinically applicable retinal fundus diagnosis system by modelling finding-disease relationship
Abstract The identification of abnormal findings manifested in retinal fundus images and diagnosis of ophthalmic diseases are essential to the management of potentially vision-threatening eye conditions. Recently, deep learning-based computer-aided diagnosis systems (CADs) have demonstrated their po...
Main Authors: | , , , , , , , , |
---|---|
Format: | Article |
Language: | English |
Published: |
Nature Portfolio
2023-04-01
|
Series: | Scientific Reports |
Online Access: | https://doi.org/10.1038/s41598-023-32518-3 |
_version_ | 1797845978147979264 |
---|---|
author | Jaemin Son Joo Young Shin Seo Taek Kong Jeonghyuk Park Gitaek Kwon Hoon Dong Kim Kyu Hyung Park Kyu-Hwan Jung Sang Jun Park |
author_facet | Jaemin Son Joo Young Shin Seo Taek Kong Jeonghyuk Park Gitaek Kwon Hoon Dong Kim Kyu Hyung Park Kyu-Hwan Jung Sang Jun Park |
author_sort | Jaemin Son |
collection | DOAJ |
description | Abstract The identification of abnormal findings manifested in retinal fundus images and diagnosis of ophthalmic diseases are essential to the management of potentially vision-threatening eye conditions. Recently, deep learning-based computer-aided diagnosis systems (CADs) have demonstrated their potential to reduce reading time and discrepancy amongst readers. However, the obscure reasoning of deep neural networks (DNNs) has been the leading cause to reluctance in its clinical use as CAD systems. Here, we present a novel architectural and algorithmic design of DNNs to comprehensively identify 15 abnormal retinal findings and diagnose 8 major ophthalmic diseases from macula-centered fundus images with the accuracy comparable to experts. We then define a notion of counterfactual attribution ratio (CAR) which luminates the system’s diagnostic reasoning, representing how each abnormal finding contributed to its diagnostic prediction. By using CAR, we show that both quantitative and qualitative interpretation and interactive adjustment of the CAD result can be achieved. A comparison of the model’s CAR with experts’ finding-disease diagnosis correlation confirms that the proposed model identifies the relationship between findings and diseases similarly as ophthalmologists do. |
first_indexed | 2024-04-09T17:48:41Z |
format | Article |
id | doaj.art-db40dd2c407843a68edf21b13b516105 |
institution | Directory Open Access Journal |
issn | 2045-2322 |
language | English |
last_indexed | 2024-04-09T17:48:41Z |
publishDate | 2023-04-01 |
publisher | Nature Portfolio |
record_format | Article |
series | Scientific Reports |
spelling | doaj.art-db40dd2c407843a68edf21b13b5161052023-04-16T11:13:28ZengNature PortfolioScientific Reports2045-23222023-04-0113111310.1038/s41598-023-32518-3An interpretable and interactive deep learning algorithm for a clinically applicable retinal fundus diagnosis system by modelling finding-disease relationshipJaemin Son0Joo Young Shin1Seo Taek Kong2Jeonghyuk Park3Gitaek Kwon4Hoon Dong Kim5Kyu Hyung Park6Kyu-Hwan Jung7Sang Jun Park8VUNO Inc.Department of Ophthalmology, Seoul Metropolitan Government Seoul National University Boramae Medical CenterVUNO Inc.VUNO Inc.VUNO Inc.Department of Ophthalmology, College of Medicine, Soonchunhyang UniversityDepartment of Ophthalmology, Seoul National University College of Medicine, Seoul National University Bundang HospitalDepartment of Medical Device Research and Management, Samsung Advanced Institute for Health Sciences and Technology, Sungkyunkwan UniversityDepartment of Ophthalmology, Seoul National University College of Medicine, Seoul National University Bundang HospitalAbstract The identification of abnormal findings manifested in retinal fundus images and diagnosis of ophthalmic diseases are essential to the management of potentially vision-threatening eye conditions. Recently, deep learning-based computer-aided diagnosis systems (CADs) have demonstrated their potential to reduce reading time and discrepancy amongst readers. However, the obscure reasoning of deep neural networks (DNNs) has been the leading cause to reluctance in its clinical use as CAD systems. Here, we present a novel architectural and algorithmic design of DNNs to comprehensively identify 15 abnormal retinal findings and diagnose 8 major ophthalmic diseases from macula-centered fundus images with the accuracy comparable to experts. We then define a notion of counterfactual attribution ratio (CAR) which luminates the system’s diagnostic reasoning, representing how each abnormal finding contributed to its diagnostic prediction. By using CAR, we show that both quantitative and qualitative interpretation and interactive adjustment of the CAD result can be achieved. A comparison of the model’s CAR with experts’ finding-disease diagnosis correlation confirms that the proposed model identifies the relationship between findings and diseases similarly as ophthalmologists do.https://doi.org/10.1038/s41598-023-32518-3 |
spellingShingle | Jaemin Son Joo Young Shin Seo Taek Kong Jeonghyuk Park Gitaek Kwon Hoon Dong Kim Kyu Hyung Park Kyu-Hwan Jung Sang Jun Park An interpretable and interactive deep learning algorithm for a clinically applicable retinal fundus diagnosis system by modelling finding-disease relationship Scientific Reports |
title | An interpretable and interactive deep learning algorithm for a clinically applicable retinal fundus diagnosis system by modelling finding-disease relationship |
title_full | An interpretable and interactive deep learning algorithm for a clinically applicable retinal fundus diagnosis system by modelling finding-disease relationship |
title_fullStr | An interpretable and interactive deep learning algorithm for a clinically applicable retinal fundus diagnosis system by modelling finding-disease relationship |
title_full_unstemmed | An interpretable and interactive deep learning algorithm for a clinically applicable retinal fundus diagnosis system by modelling finding-disease relationship |
title_short | An interpretable and interactive deep learning algorithm for a clinically applicable retinal fundus diagnosis system by modelling finding-disease relationship |
title_sort | interpretable and interactive deep learning algorithm for a clinically applicable retinal fundus diagnosis system by modelling finding disease relationship |
url | https://doi.org/10.1038/s41598-023-32518-3 |
work_keys_str_mv | AT jaeminson aninterpretableandinteractivedeeplearningalgorithmforaclinicallyapplicableretinalfundusdiagnosissystembymodellingfindingdiseaserelationship AT jooyoungshin aninterpretableandinteractivedeeplearningalgorithmforaclinicallyapplicableretinalfundusdiagnosissystembymodellingfindingdiseaserelationship AT seotaekkong aninterpretableandinteractivedeeplearningalgorithmforaclinicallyapplicableretinalfundusdiagnosissystembymodellingfindingdiseaserelationship AT jeonghyukpark aninterpretableandinteractivedeeplearningalgorithmforaclinicallyapplicableretinalfundusdiagnosissystembymodellingfindingdiseaserelationship AT gitaekkwon aninterpretableandinteractivedeeplearningalgorithmforaclinicallyapplicableretinalfundusdiagnosissystembymodellingfindingdiseaserelationship AT hoondongkim aninterpretableandinteractivedeeplearningalgorithmforaclinicallyapplicableretinalfundusdiagnosissystembymodellingfindingdiseaserelationship AT kyuhyungpark aninterpretableandinteractivedeeplearningalgorithmforaclinicallyapplicableretinalfundusdiagnosissystembymodellingfindingdiseaserelationship AT kyuhwanjung aninterpretableandinteractivedeeplearningalgorithmforaclinicallyapplicableretinalfundusdiagnosissystembymodellingfindingdiseaserelationship AT sangjunpark aninterpretableandinteractivedeeplearningalgorithmforaclinicallyapplicableretinalfundusdiagnosissystembymodellingfindingdiseaserelationship AT jaeminson interpretableandinteractivedeeplearningalgorithmforaclinicallyapplicableretinalfundusdiagnosissystembymodellingfindingdiseaserelationship AT jooyoungshin interpretableandinteractivedeeplearningalgorithmforaclinicallyapplicableretinalfundusdiagnosissystembymodellingfindingdiseaserelationship AT seotaekkong interpretableandinteractivedeeplearningalgorithmforaclinicallyapplicableretinalfundusdiagnosissystembymodellingfindingdiseaserelationship AT jeonghyukpark interpretableandinteractivedeeplearningalgorithmforaclinicallyapplicableretinalfundusdiagnosissystembymodellingfindingdiseaserelationship AT gitaekkwon interpretableandinteractivedeeplearningalgorithmforaclinicallyapplicableretinalfundusdiagnosissystembymodellingfindingdiseaserelationship AT hoondongkim interpretableandinteractivedeeplearningalgorithmforaclinicallyapplicableretinalfundusdiagnosissystembymodellingfindingdiseaserelationship AT kyuhyungpark interpretableandinteractivedeeplearningalgorithmforaclinicallyapplicableretinalfundusdiagnosissystembymodellingfindingdiseaserelationship AT kyuhwanjung interpretableandinteractivedeeplearningalgorithmforaclinicallyapplicableretinalfundusdiagnosissystembymodellingfindingdiseaserelationship AT sangjunpark interpretableandinteractivedeeplearningalgorithmforaclinicallyapplicableretinalfundusdiagnosissystembymodellingfindingdiseaserelationship |