Empowering Medicine with Responsible AI Solutions: A Journey through the Medical-Themed Hackathon
Published Jul 19 2023 12:00 PM 2,425 Views
Iron Contributor

Introduction

The world of medicine has always been at the forefront of innovation, constantly striving to improve patient care and outcomes. With the rise of Artificial Intelligence (AI), a new era of possibilities has emerged, offering groundbreaking solutions to complex medical challenges. In this context, a medical-themed hackathon took center stage, emphasizing the importance of responsible AI solutions and rapid prototyping using low-code platforms like Microsoft Power Platform.

This hackathon solution example aims to tackle a pressing issue in healthcare: accurately diagnosing pneumonia, a common and potentially life-threatening lung infection. The challenge lies in distinguishing pneumonia from other lung conditions, such as lung cancer, which can lead to misdiagnoses and delayed treatments. Leveraging the power of AI and low-code solutions, the hackathon's team developed a cutting-edge lung pneumonia detection system that revolutionizes patient care and transforms the way medical professionals operate.


Scenario

Pneumonia, affecting people of all ages, poses a significant health risk, especially to vulnerable populations. Accurate and timely diagnosis is crucial for effective treatment. However, the overlap of symptoms between pneumonia and lung cancer can create diagnostic challenges. The hackathon sought to address this by devising an AI-powered solution capable of swiftly and accurately identifying pneumonia cases, ensuring patients receive timely and appropriate care.

The Solution

The innovative solution developed during the hackathon harnessed the capabilities of Custom Vision, Power Apps, and Azure. These powerful tools provided the foundation for building a comprehensive and user-friendly system for healthcare providers. 

  1. Custom Vision: Using supervised machine learning, the team trained a model to recognize pneumonia features in chest X-rays and medical images. The model's accuracy was improved by training it on a vast dataset of known pneumonia cases and normal images.

  2. Power Apps: To empower healthcare providers with a seamless experience, a user-friendly interface was designed. Providers could effortlessly upload medical images from their devices and instantly receive predictions from the Custom Vision model. Early detection of pneumonia cases would facilitate prompt and appropriate treatment.

  3. Azure Map Air Condition: To further enhance patient care, the team integrated Azure's capabilities to monitor air quality and identify regions with high levels of pollution. This information could help reduce the risk of respiratory diseases caused by air pollution.

By leveraging these AI technologies, healthcare providers can improve the accuracy and speed of diagnosing pneumonia and lung cancer, ultimately leading to better health outcomes for patients

Technical Overview

Picture1.jpg

Dataflows :
Power Apps, Compute vision, Static Web app and Azure Map air condition.

Power Apps: we can analyze data and effortlessly navigate through a range of options, including scanning and uploading images of lungs, viewing air condition maps displaying regions with optimal air quality for those seeking pneumonia relief.

Computer vision will be employed to train and identify images with similar cases, distinguishing between pneumonia and normal.

Static Web App: will be used to host the website for direct visualization of the air condition map

custom vision.png


On this screen, you will find the images that we have uploaded for labeling and training. We have uploaded a total of 2079 images for training, each with a tag to indicate its significance. These images include examples of lung bacteria, lung virus, and normal lung.

Performance.png

 


In this training section, we will review the precision, recall, and average precision (AP) metrics in detail, by training the images we've uploaded and labeled.


Precision,
calculated as TP/(TP+FP), represents the proportion of lung pneumonia images that were correctly identified as positive out of all images that were predicted to have lung pneumonia.
Recall, calculated as TP/(TP+FN), represents the proportion of images predicted to have lung pneumonia that were correctly identified as positive out of all images that actually had lung pneumonia.
A.P., or average precision, is calculated by plotting the precision-recall (PR) curve and measuring the area under the curve (AUC). This metric is used to balance between precision and recall to avoid over- or underfitting.

 

Quick Test.png

We can now test our model by uploading an image and predicting an outcome. On the left side of the screen, you can view the details about the uploaded image in the prediction section. This includes the probabilities of lung virus, lung bacteria, and normal lung. We can use this information to analyze and evaluate the accuracy of our model.
deploiment.png

In this section, we can use the URL link and application key for our Power Apps application. These tools allow us to integrate the Custom Vision service into our app, providing a seamless user experience for our users.

Health care Home.png


This screen displays the welcome app, which includes a button to analyze a picture.
Test apps.png

On this screen, we can view an uploaded and trained image, along with its classification results. The image has been classified as 70.42% lung virus, followed by 26.81% lung bacteria, which indicates pneumonia, and 2.77% normal lung. On the left side of the section, we have the options to check the air condition, scan for the process of scanning and analyzing the image and return to the home screen. 
link for public access : Current Air Quality - Azure Maps Web SDK Samples (green-bush-080959510.3.azurestaticapps.net)

Picture2.png

This screen displays the current air quality for different regions, along with the levels of particulate matter. This information can help you understand the natural air conditions in various areas.

 

Conclusion


In conclusion, the development of AI technologies has paved the way for innovative solutions that address real-world problems. In the field of medicine, we have developed a cutting-edge solution that harnesses the power of Custom Vision, Power Apps, and Azure to accurately diagnose pneumonia cases. By training a machine learning model to recognize the features of pneumonia in chest X-rays and other medical images, healthcare providers can quickly and accurately identify cases of pneumonia and begin treatment as soon as possible. Additionally, by monitoring air quality with Azure Map Air condition, the risk of respiratory diseases caused by air pollution can be reduced. Ultimately, the integration of AI technologies in healthcare can lead to better health outcomes for patients and revolutionize the field of medicine. As the world continues to embrace AI, it is paramount to prioritize responsible AI solutions. The hackathon scenario exemplifies the collaboration between technology and medicine, emphasizing the need to leverage AI's power ethically and responsibly for better health outcomes and a brighter future in healthcare.

Co-Authors
Version history
Last update:
‎Jul 17 2023 06:28 AM
Updated by: