The problem SheCure solves

We have developed an all-in-one women health platform, which leverages AI to detect and diagnose women specific health issues and thereby help raise awareness for its treatment and management.

We included the following features:
• Breast Cancer Detection tool : Convolutional Neural Networks to detect the presence of Metastatic tissue and Invasive Ductal Carcinoma Breast’s Histopathological Images.
• Predict Cervical Cancer and preliminary test results using a Voting Ensemble and Extreme Random Trees models predict presence of and preliminary test results according to user’s demographic information, habits, and historic medical records
Using the UCI's Machine Learning repository's Cervical cancer (Risk Factors) Data Set.
• Dedicated Dashboard for Patient and Doctor in both mobile and web application
• Tracking of previous Prescription & Medical History
• Menstrual Cycle Tracker, with dietary and workout suggestions.
• Responsive UI
• Booking Lab/Appointment
• Video Conferencing for expert consultancy

Fear of the stigma associated with reproductive health services has always been one of the reasons why youth and unmarried individuals avoid making use of such services. This stigma imposes a great deal of mental stress, fear, and depression on patients and causes delays in the diagnosis and treatment of their conditions.

A research highlights that there is an urgent need to encourage more public awareness around women's health matters to assure women that they will be taken seriously, and that they should never feel ashamed about their health concerns whether that be their periods, reproductive health, menopause or something else.

Through this app, we aim at solving this barrier for women to connect with doctor and try to reduce the errors and hassle in the diagnosis of the ailments.

Challenges we ran into

Most web browsers don't support the tiff image format, which is contained in the dataset. While preprocessing the same for our web application we converted the images to .png format such that the model is trained with data i.e. similar to the expected input..

Because Tensorflowjs is a new technology, web apps bulit using it may not work in some browsers. The user will see a message saying the "Ai is loading..." but that message will never go away because the app is actually frozen. It's better to advise users to use the latest version of Chrome.

The web app for this project uses the Javascript language for the most part. We also used Javascript to feed the images to the model. The challenge is that Javascript is very fast whereas the model isn't fast enough to keep up. This difference in speed can lead to incorrect predictions. We used async/await to fix this.

Due to bias towards negative values in medical data, accuracy can be a bad metric as it gives an impression of a good performing model but in fact is biased, and won't perform wwll with other metrics. We fixed that by using Average Precision Score Weighted as the primary metric. We also calculated accuracy post that which was satisfactory.

Integrating the Machine Learning models in the web application.