Organizing follow-up care for patients with macular retinal pathologies using artificial intelligence systems

Full Text

BACKGROUND

The National Artificial Intelligence Development Strategy adopted in the Russian Federation determined the artificial intelligence (AI) to be used in increasing the efficiency of organizations.¹ This can be achieved by automating routine (repetitive) production processes and operations.

In accordance with current guidelines, if risk factors for disease progression are present in patients with diabetes mellitus-induced retinal pathology and age-related macular degeneration, regular follow-ups by an ophthalmologist are recommended to monitor changes and start adequate treatment if necessary [1, 2]. Retinal examination using a computer analyzer, i.e., structural optical coherence tomography (OCT), is included in the standard of care for primary specialized medical care of patients with diabetic retinopathy and age-related macular degeneration and is a recurring service.² The frequency of doctor visits is determined individually depending on the planned management strategies and clinical signs. According to domestic studies, the minimum number of OCT examinations of the macular retina performed in medical institutions during follow-ups in patients with age-related macular degeneration and diabetic macular edema is 1,629,429 per year.

Foreign analytical studies have shown the willingness of patients to switch to controlled self-medications during follow-up (up to 54% of respondents according to 2019 data, STADA Health Report 2020) and to receive part of the medical services using telemedicine technologies, i.e., without face-to-face visits to a healthcare professional [3–5]. In a study by the Russian Public Opinion Research Center, 48% of Russians are open to follow-up and treatment using telemedicine services [6]. Thus, organizational and clinical prerequisites have been established for the provision of medical services using AI, including the use of telemedicine technologies, for patients who require repeated diagnostic appointments as part of the follow-up for posterior eye segment pathology.

The Order of the Ministry of Health of the Russian Federation dated November 12, 2012, No. 902n, On Approval of the Procedure for the Provision of Medical Care to the Adult Population for Diseases of the Eye, Appendages, and Orbit refers to equipping medical consultative and diagnostic departments of outpatients clinics with optical coherence tomographs. However, patients with retinal pathology most often receive follow-up care in a medical ophthalmology center once they have started treatment, and this reduces the availability of treatment for patients with newly diagnosed pathology, which must be treated the earliest time possible. Existing technologies need changing and intensifying, including through the use of AI technologies.

AIM

To develop the methodological basis for an organizational system for the follow-up of patients with posterior eye segment pathology using AI-based clinical decision support systems (CDSS).

MATERIALS AND METHODS

Content analysis was performed to assess the existing regulatory framework. To create a structured directory, sociological and expert methods were used. Artificial neural networks (ANNs) were trained using segmentation and classification methods.

Study design

The existing regulatory framework was assessed based on an analysis of the Constitution of the Russian Federation, federal laws, sublegislative acts, and judicial practice. The methodology for a structured medical document describing an OCT image was created using an expert method: i.e., 100 ophthalmologists with the appropriate educational level, including advanced professional education, engaged in specialized medical care for patients with posterior segment pathology were included in a survey. The structured medical document was used as the basis for ANN predictors. As a result of annotation, 60,000 medical images were described. For each image, a.json file describes the presence of features after mapping. To analyze and classify binary features in an image, neural networks with the DenseNet121 architecture without pretrained weights were trained. The Mask R-CNN architecture was used for the segmentation task.

Eligibility criteria

A sign was added to the significant factors that indicated disease deterioration when the agreement between respondents reached 70%. A sign was added to the CDSS output when its accuracy (mean sensitivity and specificity) reached a value of ≥0.7.

Study conditions

A database was constructed based on the anonymized data from clinical studies (retinal examinations using OCT) conducted in the Orenburg and Tambov branches of the S.N. Fyodorov Eye Microsurgery Complex of the Ministry of Health of the Russian Federation. The ANN was trained at the Research Institute of Digital Intelligent Technologies of the Orenburg State University (Russia).

Study duration

Clinical studies (retinal examinations using OCT) were performed between 2015 and 2023. The structured binary classifier directory for describing the macular retina was created in 2022, and the ANNs were trained in 2022–2023.

Medical intervention

For the database, retinal examinations involved OCT scans of the macular retina.

Primary outcome

The methodological basis for the use of AI in the regular follow-up of patients with posterior segment pathology was developed and evaluated.

Subgroup analysis

The follow-up methodology was based on an assessment of the medical care organization and clinical data of patients diagnosed with age-related macular degeneration (ICD-10 code: H 35.3) or diabetic macular edema (ICD-10 code: H 35.8).

Outcome measures

Sensitivity and specificity are statistical parameters of a diagnostic test used to identify patients with pathology and healthy individuals derived from type I and II errors in the binary classification.

Statistical analysis

For the sample size calculation, the minimum volume of the sample study was determined considering the validity criterion and maximum error (product of the confidence interval and specified accuracy).

Statistical procedures included descriptive statistics, calculation of the mean values and relative values, and mathematical modeling. The statistical significance of differences in the studied data according to qualitative characteristics was analyzed through mathematical calculations and subsequently evaluated using Pearson’s chi-squared test. Quantitative variables were described during the preliminary assessment for compliance with Gauss’s law (normal probability distribution).

RESULTS

Study objects

The study focused on organizing ophthalmological care for the adult population as part of their follow-up for posterior segment pathology. In the content analysis, 22 records were analyzed. The database included 60,000 records of retinal examinations using OCT for which a medical image (scan) and a result interpretation in accordance with the developed binary classifier are available.

Primary outcome results

In accordance with Federal Law dated November 21, 2011, No. 323-FZ “On the Fundamentals of Public Healthcare of Citizens in the Russian Federation,” medical care in the Russian Federation was organized and provided:

1) In accordance with the regulations on the organization of medical care by type of medical care

2) In accordance with the procedures for the provision of medical care approved by the authorized federal executive body and mandatory for execution in the territory of the Russian Federation by all medical organizations

3) Based on guidelines

4) Considering the standards of medical care approved by the authorized federal executive body.³

In addition, Paragraph 15 of Article 2 of this law defines “attending physician,” who is a doctor responsible for organizing and directly providing medical care to a patient during follow-up and treatment. In the provision of primary specialized medical care, certain functions of the attending physician may be delegated to nursing staff in accordance with the approved procedure; however, the provision of primary specialized medical care, including medical services specified in the professional standard of an ophthalmologist, should not be delegated.^{4 5}

Results of imaging studies cannot be interpreted by nursing staff independently when providing medical services. This is only possible with the use of registered medical devices.

CDSSs require registration as a medical device for use by nursing personnel. When used by doctors, they can remain within the legal field of information services because the interpretation of results, diagnosis, and recommendations, including those in follow-up, remain within the competence of the attending physician, regardless of the legal status of the software.

Using the expert method, 123 binary features that allow describing the structure of healthy and abnormal macular retina were selected and characterized. The features were grouped into sections: general, vitreoretinal and retinal interface, retinal contour, retinal thickness, retinal structures, and choroid. In the survey of experts with appropriate education and professional experience, 75% of the doctors classified only 26 features out of 123 as predictors of worsening conditions in patients with age-related macular degeneration or diabetic macular edema (Table 1).

 

Table 1. Signs of clinical deterioration

Sign	Section
Moderate macular thickness increase (≤500 μm)	Macular thickness
Severe macular thickness increase (>500 μm)
Focal intraretinal edema identified	Retinal structure
Diffuse intraretinal edema identified
Cystic intraretinal edema identified
Abnormality: small (<50 μm) multiple cysts
Abnormality: medium (50–150 μm) multiple cysts
Abnormality: large (>150 μm) multiple cysts
Serous neuroepithelial detachment visualized
Slit-like neuroepithelial detachment visualized
Serous RPE detachment visualized
Dome-shaped RPE detachment
Flat wave-like RPE detachment visualized
Flat RPE detachment visualized
“Tabletop” RPE detachment visualized
Hemorrhagic RPE detachment visualized
Fibrovascular RPE detachment visualized
Double-layer sign visualized
Highly reflective opacity shadowing deeper layers (subretinal hemorrhage)
Subretinal fluid visualized
Hyperreflective focus visualized in inner retinal layers
Hyperreflective focus visualized in outer retinal layers
Hyperreflective focus visualized above RPE
Hyperreflective focus visualized within RPE
Hyperreflective focus visualized under RPE
Hyperreflective focus shadowing the underlying layers visualized

Note: RPE, retinal pigment epithelium.

 

The selected features directly or indirectly indicated the onset of a pathological process, including neovascularization, which requires specialized and high-tech medical care in tertiary medical institutions (e.g., federal medical organizations). In addition, the features were consistent with the guidelines and current scientific literature describing medical care provision and interpretation of the results of imaging diagnostic studies [7, 8].

The binary classifier was divided into features suitable for use in ANN training by classification and segmentation methods. The ANN training based on 60,000 medical images made it possible to create a CDSS for follow-up in patients with posterior segment pathology. The accuracy of ANN models was assessed using the balanced accuracy metric to correctly account for the heterogeneous distribution of classes in the data. The mean balanced accuracy across features was 81%.

Secondary outcome results

The developed version of the CDSS is in the open-access system: http://retinadeepai.site/. The system recognizes objects to be analyzed as predictors for each of the developed features and allows for the configuration of the analysis of various maps and modes, including those obtained using different devices. The recommended analysis mode is RetinaMap. The service has an ergonomic, intuitive interface that makes it possible to upload a report attached to the medical records. The service is not a medical device, but an information service; it does not store or process personal patient data. The service does not make a diagnosis or offer prompts for diagnosis. It describes the structure of the normal and abnormal retina, taking on part of the routine processes of an ophthalmologist during patient follow-ups.

Adverse events

None identified.

DISCUSSION

Currently, several AI-based systems are being developed globally for the early diagnosis of fundus pathology. For example, certain systems detect early signs of diabetic retinopathy using fundus images obtained from stationary and portable fundus cameras: the IDx-DR system (LumineticsCore, USA) and the Retina.AI platform (Digital Vision Solutions LLC, Russia).

In addition, several technological developments aimed at automated analysis of data obtained using medical devices.

• The Retina.AI platform is aimed at analyzing OCT images of the retina and identifying one of the specified syndrome complexes: subretinal fluid, intraretinal cysts, retinal pigment epithelial detachment, subretinal hyperreflective material, epiretinal membrane, retinal drusen, full-thickness macular hole, lamellar macular hole, and vitreomacular traction (https://www.screenretina.com/). Thus, the project is aimed at diagnosing conditions and assessing the course of the process. The project is not registered as a medical device.
• The Altris AI platform (USA) automates the selection of abnormal OCT scans and detection of >70 disorders and pathological signs, including epiretinal fibrosis, intraretinal cystoid fluid, pseudocysts, diffuse edema, fibrovascular detachment of the retinal pigment epithelium, and subretinal hyperreflective material (https://www.altris.ai/). The project aims to diagnose conditions (diseases) and train medical personnel. The project is registered as a medical device in the country of origin (US Food and Drug Administration).

The product we are developing contains a larger number of assessed features (123) and aims not at diagnosis but at changing the organization of the diagnostic and treatment process, i.e., delegating routine work from doctors to nursing staff, reducing research time, and increasing the availability of medical services.

It requires changing the professional standard of nursing staff by including the use of AI-based information services as part of patient follow-up. In addition, in the Order of the Ministry of Health of Russia dated March 15, 2023 No. 168n, “On Approval of the Procedure for Regular Follow-up of Adults,” the regulatory framework must be updated for the regular follow-up of patients with posterior segment pathology to include this pathology.

Primary outcome summary

The methodological basis of the organizational technology for regular follow-up of patients with posterior segment pathology has been developed. A structural classifier for describing the retina has been prepared, and a CDSS has been developed for use in the regular follow-up of this patient population.

Primary outcome discussion

The Order of the Ministry of Health of the Russian Federation dated September 7, 2020, No. 947n, “On Approval of the Procedure for Organizing a Document Flow System in the Field of Healthcare to Manage Medical Electronic Records” defines the procedure for using electronic medical records. To date, several structured electronic medical documents have been developed and approved and are currently used in Russia. The Center for the Development of Structured Electronic Medical Documents is part of the Central Research Institute for Organization and Informatization of Healthcare of the Ministry of Health of the Russian Federation. It ensures the implementation of the tasks of the Ministry of Health of Russia to improve the procedure for organizing document flow in healthcare by developing, updating, and modernizing guidelines for the implementation of structured electronic health records. The developed directory can become the basis for a medical protocol for an imaging study of the retina using OCT.

The accumulation of structured information will contribute to the improvement of the developed CDSS. Using AI-based CDSSs, changing the regulatory framework will help speed up the implementation of the organizational technology for regular follow-up of patients with posterior segment pathology.

Study limitations

The results of the study are intended for follow-up in patients with posterior segment pathology. The results cannot be used in the primary diagnosis of eye, ocular adnexa, and orbit pathologies.

CONCLUSION

In the legal framework of the Russian Federation, some of the physician’s medical functions could not be delegated to nursing staff such as when interpreting the results of a medical examination. For example, in eye, ocular adnexa, and orbital disorders, the professional standard and regulatory framework must be amended. However, no uniform rules have been established for creating a structured electronic medical record containing the results of interpreting a retinal examination using OCT. The proposed classifier made it possible to create and train (based on 60,000 medical images) a CDSS, which, as an information service not intended for establishing a diagnosis, allows for the reorganization of the follow-up process. The primary service of the developed CDSS is patient routing. If signs of deterioration are observed in the clinical presentation, referral to a medical ophthalmology center is warranted to assess the disease course and provide specialized, including high-tech medical care.

ADDITIONAL INFO

Funding source. This study was not supported by any external sources of funding.

Competing interests. The authors declare that they have no competing interests.

Authors’ contribution. All authors made a substantial contribution to the conception of the work, acquisition, analysis, interpretation of data for the work, drafting and revising the work, final approval of the version to be published and agree to be accountable for all aspects of the work. A.D. Chuprov, I.P. Bolodurina — development of the concept, approval of the final version of the manuscript; A.O. Lositskiy — development of methodology, database collection, writing and editing the text of the article; A.Yu. Zhigalov — development of methodology, conducting research.

 

¹ Decree of the President of the Russian Federation dated October 10, 2019 No. 490 “On the Development of Artificial Intelligence in the Russian Federation” (together with the “National Artificial Intelligence Development Strategy for the period until 2030”). Access: https://www.consultant.ru/document/cons_doc_LAW_335184/1f32224a00901db9cf44793e9a5e35567a4212c7/ Access date: November 22, 2023.

² Order of the Ministry of Health of the Russian Federation dated December 24, 2012, No. 1492n “On Approval of the Standard of Primary Health Care for Diabetic Retinopathy and Diabetic Macular Edema.” Access: https://base.garant.ru/70344052/53f89421bbdaf741eb2d1ecc4ddb4c33/ Access date: November 22, 2023.

³ Federal Law “On the Fundamentals of Public Healthcare of Citizens in the Russian Federation” dated November 21, 2011, No. 323-FZ. Access: https://www.consultant.ru/document/cons_doc_LAW_121895/ Access date: November 22, 2023.

⁴ Order of the Ministry of Health and Social Development of the Russian Federation dated March 23, 2012, No. 252n “On Approval of the Procedure for Assigning Certain Functions of the Attending Physician to a Paramedic or Midwife by the Head of a Medical Organization when Organizing the Provision of Primary Health Care and Emergency Medical Care for the Direct Provision of Medical Care to the Patient During Follow-up and Treatment, Including the Prescription and Administration of Medications, Including Narcotic Drugs and Psychotropic Drugs.” Access: https://base.garant.ru/70170588/ Access date: November 22, 2023.

⁵ Order of the Ministry of Health and Social Development of the Russian Federation dated June 5, 2017, No. 470n “On Approval of the Professional Standard for Ophthalmologists.” Access: https://docs.cntd.ru/document/436744741 Access date: November 22, 2023.

About the authors

Aleksandr D. Chuprov

The S. Fyodorov Eye Microsurgery Federal State Institution

Email: nauka@ofmntk.ru
ORCID iD: 0000-0001-7011-4220

MD, Dr. Sci. (Medicine), Professor

Russian Federation, Orenburg

Irina P. Bolodurina

Orenburg State University; Orenburg State Medical University

Email: prmat@mail.osu.ru
ORCID iD: 0000-0003-0096-2587
SPIN-code: 4848-0669

Dr. Sci. (Engineering), Professor

Russian Federation, Orenburg; Orenburg

Aleksandr O. Lositskiy

The S. Fyodorov Eye Microsurgery Federal State Institution

Email: eyedoct@yandex.ru
ORCID iD: 0000-0002-8716-6438
SPIN-code: 2044-3410

MD, Cand. Sci. (Medicine)

Russian Federation, Orenburg

Artur Yu. Zhigalov

Orenburg State University

Author for correspondence.
Email: prmat@mail.osu.ru
ORCID iD: 0000-0003-3208-1629
SPIN-code: 4692-9037
Russian Federation, Orenburg

References

Supplementary files