“Elina 1.0” - Team NeoHealth

(The project has been renamed, from X-TPA to Elina 1.0)

1. Project Overview

a.1. Team members:

1. Trịnh Lâm Khải - Hanoi Amsterdam High School for Gifted Students • Email: trinhlamkhai@gmail.com

2. Lưu Trung Đức - Hanoi Academy

3. Đoàn Thị Minh Khuê - Hanoi Amsterdam High School for Gifted Students

a.2. Mentor: • MSc.Nguyen Minh Huy • Email: huy.nm211231m@sis.hust.edu.vn

• Affiliation: School of electrical and electronic engineering - Hanoi University of Science and Technology

1. Project Overview

Respiratory infections are rising in cases, with respiratory pandemics emerging continuously in recent times: SARS (2003), MERS (2012), Covid-19 (2019 - now). In contrast, the number of qualified physicians for interpreting chest radiographs is insufficient for the increasing demand from patients. To solve this problem, “Elina 1.0” works as a chest radiograph interpretation assistant and manages patients’ records. “Elina 1.0” includes a mobile App with AI technology and a bracelet that uses Barcode/QRcode and health sensors, aiding doctors with managing patients’ records and carrying out treatment guidelines. Thus, “Elina 1.0” hopes to reduce pressure on the medical force, therefore creating a healthier community.

1. Concept

c.1. Problem

Due to worsening air pollution and outbreaks of respiratory epidemics (COVID-19, SARS,MERS), cases of lung-related issues, such as bronchitis, pneumonia, pulmonary embolism…, are higher than ever. It is estimated about 1.76 million people each year die from lung cancer globally, making lung cancer the most dangerous cancer. In Vietnam in particular, on average, there are 15,000 people every year suffering from lung diseases caused by environmental pollution, metastasis, or transmission. During the last outbreak of Covid-19, in Vietnam, it is estimated that for every 37 patients, there is only 1 doctor. The medical force is under tremendous pressure, doctors and nurses have to work overtime due to increased workload, which causes delays, interruptions in disease detection, treatment for patients, and reduces efficiency in managing information of patients. For these reasons, our team NeoHealth wishes to propose a solution called “Chest radiograph interpretation and patients' records management assistant", which we named “Elina 1.0”. The system will help doctors diagnose the patient's condition via chest X-ray imaging, effectively managing the information of doctors - patients through mobile’s app and smart bracelet.

c.2. Current tasks

Due to high cost and insufficient physicians capable of interpreting chest X-ray scans, patients have to wait very long for lung diagnosis, and treatments are not immediate. In Vietnam in particular, our team NeoHealth found that there is no software to support doctors in reading X-ray results that is ready to use and combine patients’ management. The system that the team is building on the basis of a mobile app and smart bracelet device will help effectively manage doctor-patient information, while reducing costs and time for diagnosis and treatment of diseases.

c.3. Solution

Our team NeoHealth propose a solution consists of 3 parts:

• Mobile App: store patients’ records and send uploaded X-ray scans to AI

• AI algorithms (artificial intelligence) in image processing on chest X-ray images to point out anomalies related to lungs

• Bracelet with QR/Barcode to scan patient’s profile and give health parameters (%SpO2, PRbpm, %PI..)

1. Plan

d.1. Approach

1. Expected used technologies

Figure 1: Overview Flowchart

1. AI algorithm

Figure 2: Block diagram of diagnostic models of disease

1. Data: 2 main datasets:

• National Institutes of Health Chest X-Ray Dataset (NIH): 112,120 chest x-ray images

• SIIM-FISABIO-RSNA COVID-19 Detection (SIIM): 6,334 chest x-ray images

To prepare for the process of training and evaluating the model, we will divide the data into three sets: Train/Val/Test in the ratio: 80:10:10.

1. Model training

• Using the Dense Convolution Network (DenseNet) neural network architecture as the core of the model for diagnosing the disease status of Covid-19 patients..

1. Programming the app on Android/IOS operating systems

• Interface design: designed on Figma

Figure 4: Demo App’s User Interface

• Server

1. Smart Bracelet

Hardware block diagram for smart bracelet

Our smart bracelet would be using 3D printing plastic material that is eco-friendly to the environment, the diameter of our product is likely to be 27cm. In addition, the product would be powered by a 5V DC battery.

Firstly, the bracelet would measure the user's data such as heart rate, SPO2, and body temperature using the MAX30102 and MLX96014 sensors. This data would first be transmitted to the ESP8266 before delivered to Server 1(Express) to store the user's health database. Furthermore, from these stored data, Server 1 would send back these data to ESP8266 to be displayed on our OLED 0.91 inch monitor attached to the bracelet.

Next, the bracelet would manage and store data through the use of QR code. Each bracelet would contain a unique QR code to constantly update and store user health information received from Server 1 for doctors and experts to easily access.

• Testing:

We will assess the accuracy and response time of the image recognition results of the neural network model after training and, in addition, checking the quality and stability of the Server system, the transmission of information between the Server and the mobile application system. Next, Test the UI part of the mobile application; delay when switching between interface windows on the application and the accuracy of the health measuring bracelet

• Applications

Product testing with doctors and patients at Bach Mai hospital, Central Hospital for Tropical Diseases, Viet Duc Hospital, 108 Military Central Hospital

d.2. Resource:

• Our team will receive resources from knowledge from subjects from school such as Mathematics, Physics, Biology and IT assistance from teachers in such subjects; search engines like Google and references from past papers and research from Github,Youtube; counseling from technical experts involved in software, hardware issues.

d.3. Goals

1. Main Products

   1. Mobile App (completed)

   2. AI interpret chest X-ray scans and point out health anomalies (completed)

   3. Bracelet (done by 31/7/2022 - Person in charge: Lưu Trung Đức)

2. Work allocation:

   1. Lưu Trung Đức: In charge of Smart Bracelet

   2. Đoàn Thị Minh Khuê: Design demo, do background researches, in charge of Mobile App

   3. Trịnh Lâm Khải: do background researches, In charge of AI developing

3. Product operational specifications

Product operational specifications include the accuracy of the AI algorithm; the range of measurement values and error of measured health indicators. The App gives a diagnosis of the lung patient's condition with an accuracy of about 90%. The operation parameters of the bracelet include: body temperature (measured from 25 - 42°C, error about 0.2 - 0.5°C), heart rate (measured from 60 - 150 beats/minute), SpO2 (range of 0 - 100%)

1. Inspection and Evaluation

Firstly, the project will be tested with volunteers. After the initial testing, the team will analyze the product tolerances and make reasonable adjustments to prepare for the next test. More specifically, the team will check the user interface section of the mobile application; latency when switching between the interface windows on the application, the quality and stability of the Server system, the transmission of information between the Server and the mobile application system, and finally the accuracy of the bracelet.

Once the product functions have been run steadily and reached the allowable error, the product will be put into inspection at the Hanoi Department of Science and Technology. From the use, comments and suggestions of users, the authors will complete the product before making the market.

d.4. Risks

1. Insufficient data: The team uses 2 datasets consisting of over 112,000 images. This is a large number of lung X-ray scans, however, our team will continue collecting data from local hospitals to strengthen our database.

2. AI’s accuracy does not meet requirements: To tackle this issue, our team will increase data collecting and test numerous neuron models to select the most accurate and efficient model.

3. Data updated from the bracelet are delayed from real time.: Although this issue does not affect “Elina”’s performance, conducting more research to improve the bracelet’s components will likely help the bracelet transmit real-time data.

d.5.Timeline:

1. Table 1: Project milestones and criteria

WORKTIMEMEMBER1Brainstorming Idea, research documents (Completed)1/3/2022 - 15/3/2022All 3 members, Mentor 2Learn about the methods of testing, screening and diagnosis of patients with lung diseases and health monitoring devices already available. Expecting features of the system. (Completed)19/3/2022 - 21/3/2022All three members, Mentor3Design and build the system, including three main parts:

• Processing X-ray using AI algorithm.

• Apply on Android/IOS

• Storage and processing data server.

(Completed)1/4/2022 - 2/6/2022Trịnh Lâm Khải, Đoàn Thị Minh Khuê, Mentor 4Complete project summary report3/6/2022 - 15/6/2022All three members, Mentor5Design and build the smart bracelet 16/6/2022 - 31/7/2022Lưu Trung Đức, Mentor6Testing the product1/8/2022 - 15/8/2022All three members, Mentor 7Adjusting, finishing the product. 16/08/2022 - 31/08/2022All three members. Mentor 8Auditing. 1/09/2022 - 15/09/2022All three members, Mentor9Complete the final research summary report, participating in the final round at VinUni. 16/09/2022 - 25/09/2022All three members, Mentor

1. Aiding: Aid received will be allocated to Server renting, and materials for Smart Bracelet

d.6.Project Budget:

Smart Bracelet’s accessoriesCostSourceESP826692.000 VNDKit RF thu phát Wifi ESP8266 NodeMCU Lua CP2102 – Hshop.vnHeartbeat/SPO2 sensor(MAX30102)65.000 VNDCảm biến nhịp tim và oxy trong máu MAX30102 – Hshop.vnBody temperature sensor(MLX90614)345.000 VNDCảm biến nhiệt GY-906-DCC MLX90614 Medical Accuracy Non-Contact IR The – Hshop.vnOled display 0.91 inch55.000 VNDMàn hình Oled 0.91 inch giao tiếp I2712C – Hshop.vn5V DC battery5.000 VND LED5.000 VND 3D printing plastic150.000 VND Server Local server on personal computerFreePersonal computerServer rental from third-party4.000.000 VND/ 6 months Total cost4.717.000 VND

1. Team’s personal information.

- Lưu Trung Đức: 12th grader with background in Engineering

Able to code in Arduino, Python, Java

GPA: 9.2; SAT: 1460

- Trịnh Lâm Khải: 11th grader with a background in Programming, Web Design.

Able to code in Python, Javascript, C++.

GPA:9.5; SAT 1440; IELTS: 7.5

- Đoàn Thị Minh Khuê: 11th grader with background in Competitive Programming, Web Design, Graphic Design. Fluent in C++, can use Java, Python, Figma. Proficient in Adobe Photoshop and Adobe Illustrator. GPA: 9.3; IELTS 7.5; SAT 1520.

The main reason all three members were interested in the research was our passion for engineering science, computers, and furthermore our desire to apply technology to other fields, especially medicine.

Skills to learn: Product Engineering, Machine learning, Server and API concepts, Expo framework, App interface design, Analyzing X-ray images...

f. References

[1] VnExpress, 2020, Nearly 15,000 Vietnamese suffer invasive pulmonary lung infection annually https://e.vnexpress.net/news/news/nearly-15-000-vietnamese-suffer-invasive-pulmonary-lung-infection-annually-4057904.html

[1] Trần Mạnh, Đình Nam "Các phương pháp xét nghiệm và chẩn đoán bệnh COVID-19", http://baochinhphu.vn, 2020.

[2] M. Abadi et al., “TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems,” 2016, [Online]. Available: http://arxiv.org/abs/1603.04467..

[3] F. Florencio, T. Valença, E. D. Moreno, and M. Colaço Junior, “Performance analysis of deep learning libraries: Tensor flow and PyTorch,” J. Comput. Sci., vol. 15, no. 6, pp. 785–799, 2019, doi: 10.3844/jcssp.2019.785.799.

[4] R. A. Hughes, “Geoscience data and derived spatial information: Societal impacts and benefits, and relevance to geological surveys and agencies,” Spec. Pap. Geol. Soc. Am., vol. 482, pp. 35–40, 2011, doi: 10.1130/2011.2482(04).

[5] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, “You only look once: Unified, real-time object detection,” Proc. IEEE Comput. Soc. Conf. Comput. Vis. Pattern Recognit., vol. 2016-Decem, pp. 779–788, 2016, doi: 10.1109/CVPR.2016.91.

[6] P. Mishra, PyTorch Recipes. 2019.