Development and validation of Deep Learning Model to diagnose Coronary Artery Disease from Computed Tomography Coronary Angiography Images

Protocol

**1. Title of the project:**Development and validation of Deep Learning Model to diagnose Coronary Artery Disease from Computed Tomography Coronary Angiography Images

2. Type of Study:

Retrospective study

3. Aims & objectives (hypotheses if applicable): Build and validate Deep Learning Model to diagnose Coronary Artery Disease (CAD) from Computed Tomography Coronary Angiography (CTCA) Images

4. Justification for study (whether of national significance with rationale): CAD is a leading cause of mortality worldwide and in recent years there has been a trend of the younger demographic increasingly suffering from CAD, especially myocardial infarction. Diagnosis of CAD heavily relies on the use of invasive angiography in our country. Though non-invasive imaging tests like CTCA are available which can be used in a proportion of patients suspected of CAD, the lack of availability of expert radiologists/cardiologists who can report CTCA images has limited the widespread utilization of this test. Addressing this challenge necessitates the creation of advanced, real-time cardiac screening, diagnosis, and prognosis systems. Through this study, we aim to develop one such system: an accurate deep learning-based artificial intelligence (AI) model to diagnose CAD from CTCA images in a time-efficient manner. Such a model can be scaled to diagnostic centres in tier 2 and 3 cities potentially improving CAD diagnostic pathways in resource-limited settings and potentially reducing morbidity and mortality through early diagnosis. The proposal aligns with the Ayushman Bharath Digital mission to create a digital health solution ecosystem.5. Departments involved:

Department of Computer Science and Engineering, Manipal

Department of Cardiology, Kasturba Medical College, Manipal

Department of Radiodiagnosis and Imaging, Kasturba Medical College, Manipal

Cardiovascular medical imaging applications, Philips Innovation Campus Bengaluru

Department of Data Science, Prasanna School of Public Health, Manipal

6. Study period: 1 year

Sample size: Time-bound. All the CTCA images available in the last 10 years from hospital records and PACS will be utilized.

8. Materials and methods:

a) Inclusion and exclusion criteria: Inclusion criteria: Adult patients (> 18 years of age) who have undergone CTCA in the last 10 years for diagnosis of CAD.Exclusion Criteria:Patients with poor quality CTCA images. b) Biological materials required(type - blood, tissue etc and quantity): Yes ☐    No   ☒

i) Biological material: Nilii) Biosafety Measures: NA

c) Statistical methods: Mean and standard deviation will be used to summarize continuous normally distributed data, and median (interquartile range) will be used to summarize non-normal data. Frequency and percentages will be used for nominal/categorical data. The steps involved in machine learning are detailed in the methods section. We will use the temporal validation method: training data will be from patients in the first 75-80% duration of the study and test data will be from the final 20-25% duration (e.g. CTCA images from the first 7-8 years of data collection period will be part of training data and the last 2-3 years will be part of test data). Using this method, we ensure that the test data is truly unseen during the training phase. After models are trained using the training data, they will be evaluated on the unseen test data against the ground truths (labels) using the confusion matrix, accuracy, f1-score, and ROC analysis.

d) Tools used: Nil

9. Detailed description of procedure/processes:

The research team from the Manipal Academy of Higher Education (MAHE) is collaborating with Indian Institute of Technology, Madras (IIT Madras) for building the AI deep learning model. A Memorandum of Understanding (MoU) will be signed to share data and expertise for effective collaboration.

After obtaining the necessary permissions and ethical clearance, CTCA images from hospital records and PACS will be accessed and copied to local storage identified for the study after appropriate de-identification of images. Relevant clinical and lab data including the CTCA reports will also be recorded.

All data, especially the image data will be de-identified, stored securely and accessed only by the research team and data scientists authorized by the research team for model building. Data will be divided into training and test data using a temporal method (as described in statistical analysis). Training data will be further divided into train and validation subsets for training the models. Appropriate data preprocessing steps will then be applied to prepare the data for ML tasks. Several candidate model architectures will be tried and the one giving the best accuracy along with good cross-validation scores (indicating robustness) will be chosen as the final model. Evaluation of the models for classification tasks will be done on unseen test data using standard methods like confusion matrix, accuracy score, f1-score, and ROC analysis.

10. Outcome measures: Accuracy of deep learning AI image classification model in diagnosing CAD.

11. Potential risks and benefits: Minimal risk. This is a retrospective study involving patient records from the past. The image data will be de-identified and stored securely so that only the research team and data scientists involved in the study have access to the data.

By developing an accurate deep learning model to diagnose CAD from non-invasive CTCA, invasive procedures like conventional angiography can be minimized or avoided. Further, real-time (few minutes) diagnostic capabilities will help to scale to secondary care settings which are resource-limited thus filling a gap in care at these places. Integration of these AI systems into clinical workflow will improve decision-making and efficiency resulting in improved patient outcomes.

12. Ethical considerations and methods to address issues: IEC clearance will be obtained before accessing data for this study. All data, especially the image data will be de-identified, stored securely and accessed only by the research team and data scientists authorized by the research team for model building ensuring confidentiality, privacy, and security. Necessary MoU will be signed for collaboration with external teams outside MAHE (i.e., Indian Institute of Technology, Madras) for deep learning tasks and data sharing.

13. Budget (give details) and proposed funding source:

Source: AI Centers of Excellence Grant from Ministry of Education, Government of India.

This is a joint grant between MAHE and IIT, Madras. Total amount: Rs. 2 crores of which about Rs. 81.8 lakh is marked for MAHE for this project.

Sl. No

Details

Justification

Funding agency

INR

|1.

Capital equipment (Laptop, Workstation, iPAD)

To monitor the AI model development work with high end GPU, workstation for clinicians and for project management

Ministry of Education, GOI

9,28,000

|2.

Subscriptions (Amazon AWS, Google cloud storage)

Cloud based GPU computing and to run existing DL models, CTCA image data is huge in TB to store

Ministry of Education, GOI

9,50,000

|3.

Manpower (Structured salary, Consultancy charges, Trainings)

Research Scientist technical, Research Scientist medical, Supporting staff, Annotating CT images, charges for trainings

Ministry of Education, GOI

38,42,000

|4.

Travel (Meetings, Data collection, project demonstration)

Project discussion with other radiology centers, travel for Image annotation training and any other technical training, for image collection from other centers

Ministry of Education, GOI

11,90,000

|5.

Consumables (Stationary items, Printer, Promotion and publicity, 3D printing material)

Essential items requirement for work and for data collection, for printing the segmented arteries for subjective assessment

Ministry of Education, GOI

4,15,000

|6.

Contingency<15% of (1, 2, 3) (Validation, annotation, auditing, institute overhead)

If more annotation is required with increase in samples, For quality assurance of the work, 5% of the budget

Ministry of Education, GOI

8,58,000

|Total

81,83,000

14. Review of literature (within 1000 words):

In India, NCDs, like CAD (coronary artery disease), constitute a major public health challenge, accounting for upto 27% of total deaths (1). Despite awareness among patients and clinicians, a significant proportion of CAD remains underdiagnosed until very late. Of particular concern in India is the early age of onset, rapid progression of CAD and high mortality rate (2). In the current clinical practice, there is a heavy reliance on coronary angiography for the diagnosis of CAD, which is an invasive procedure and involves significant costs (3). There is a need for reliable, affordable, scalable, non-invasive solutions that can replace conventional coronary angiography for some of the indications (3).

The CT Coronary Angiogram (CTCA), a test for diagnosis of coronary artery disease, is grossly underutilized compared to conventional coronary angiogram due to a lack of wide availability of expert radiologists, reporting delays, and poor reporting quality (4). However, the test has nearly 96% diagnostic accuracy in specific scenarios (5). Further, information from clinical data and other test results have not been adequately utilized to improve diagnostic and prognostic systems. Traditional frameworks centered around radiologists/cardiologists have been unable to cater to the requirements of the population, demanding innovative Artificial Intelligence (AI) approaches (6). AI approaches using deep learning (DL) have shown promise in automatic stenosis detection with minimal involvement of expert radiologists or cardiologists (7).

We aim to develop an AI-based real-time CAD diagnostic and prognostic system soon after image acquisition. The scientific result will be a clinically validated AI model/tool that is scalable to primary and secondary settings. In Phase 1, we focus on developing the POC for CT image-based diagnosis of CAD. In Phase 2, on top of POC, we focus on a robust AI model with multi-modality data such as ECG, Echo, Calcium scoring, and stress tests to diagnose and predict future cardiovascular events. This proposal seeks ethical approval for Phase I (Proof of Concept) which involves building a non-invasive CTCA image-based deep learning model to diagnose the CAD with radiomics feature extraction and analysis.

This work aligns with 3, 4 and 9 SDG goals. The work will follow the consensus recommendations CLAIM, CLEAR, FAIR, DISHA [MeiTy] and GDPR [EU] guidelines for data privacy and security (8–14).

 15. References:

1.         Kalra A, Jose AP, Prabhakaran P, Kumar A, Agrawal A, Roy A, et al. The burgeoning cardiovascular disease epidemic in Indians – perspectives on contextual factors and potential solutions. Lancet Reg Health - Southeast Asia [Internet]. 2023 May 1 [cited 2024 Apr 14];12. Available from: https://www.thelancet.com/journals/lansea/article/PIIS2772-3682(23)00016-1/fulltext

2.         Sreeniwas Kumar A, Sinha N. Cardiovascular disease in India: A 360 degree overview. Med J Armed Forces India. 2020 Jan;76(1):1–3.

3.         Knaapen P. Computed Tomography to Replace Invasive Coronary Angiography? Circ Cardiovasc Imaging. 2019 Feb;12(2):e008710.

4.         Parsons IT, Bannister C, Badelek J, Ingram M, Wood E, Horton A, et al. The HASTE Protocol: a standardised CT Coronary Angiography service operated from a District General Hospital. Open Heart. 2018 Jul 11;5(2):e000817.

5.         Bittencourt MS, Hulten EA, Veeranna V, Blankstein R. Coronary Computed Tomography Angiography in the Evaluation of Chest Pain of Suspected Cardiac Origin. Circulation. 2016 May 17;133(20):1963–8.

6.         Baeßler B, Götz M, Antoniades C, Heidenreich JF, Leiner T, Beer M. Artificial intelligence in coronary computed tomography angiography: Demands and solutions from a clinical perspective. Front Cardiovasc Med. 2023 Feb 16;10:1120361.

7.         Paul JF, Rohnean A, Giroussens H, Pressat-Laffouilhere T, Wong T. Evaluation of a deep learning model on coronary CT angiography for automatic stenosis detection. Diagn Interv Imaging. 2022 Jun;103(6):316–23.

8.         Mongan J, Moy L, Kahn CE. Checklist for Artificial Intelligence in Medical Imaging (CLAIM): A                     Guide for Authors and Reviewers. Radiol Artif Intell. 2020 Mar;2(2):e200029.

9.         Wiggins WF, Magudia K, Schmidt TMS, O’Connor SD, Carr CD, Kohli MD, et al. Imaging AI in Practice: A Demonstration of Future Workflow Using Integration Standards. Radiol Artif Intell. 2021 Nov;3(6):e210152.

10.       van Timmeren JE, Cester D, Tanadini-Lang S, Alkadhi H, Baessler B. Radiomics in medical imaging—“how-to” guide and critical reflection. Insights Imaging. 2020 Aug 12;11(1):91.

11.       Kocak B, Baessler B, Bakas S, Cuocolo R, Fedorov A, Maier-Hein L, et al. CheckList for EvaluAtion of Radiomics research (CLEAR): a step-by-step reporting guideline for authors and reviewers endorsed by ESR and EuSoMII. Insights Imaging. 2023 May 4;14(1):75.

12.       Vallières M, Zwanenburg A, Badic B, Cheze Le Rest C, Visvikis D, Hatt M. Responsible Radiomics Research for Faster Clinical Translation. J Nucl Med. 2018 Feb;59(2):189–93.

13.       Wilkinson MD, Dumontier M, Aalbersberg IjJ, Appleton G, Axton M, Baak A, et al. The FAIR Guiding Principles for scientific data management and stewardship. Sci Data. 2016 Mar 15;3(1):160018.

14.       European Parliament, Council of the European Union. Regulation (EU) 2016/679 of the European Parliament and of the Council [Internet]. OJ L 119, 4.5.2016, p. 1–88; 2016 [cited 2023 Apr 13]. Available from: https://data.europa.eu/eli/reg/2016/679/oj

15.       (MeiTY) Data Transfer of Digital Health Records [online]: https://www.pib.gov.in/Pressreleaseshare.aspx?PRID=1578929 (Accessed: 04.01.2024).