Inspiration
"With 8,579 reported cases and more than 800 estimated deaths, the Sri Lankan public health surveillance system documented the largest outbreak of leptospirosis in Sri Lankan history in 2020." (1)
"In Manila, The Philippines, the total economic value of preventing leptospirosis was estimated to be $ 124.97 million per annum, which was 1.13% of metro Manila’s GDP." (2)
An accurate heuristic for detecting rare diseases like leptospirosis, which require expensive and uncommon tests to diagnose with certainty, has the potential to radically transform healthcare outcomes in underserved communities dealing with pandemic outbreaks. With PCR tests costing anywhere from $100 to $300 (not counting the cost of shipping them to communities without proper infrastructure), a quick and inexpensive way to determine a patient's risk in having leptospirosis that can be performed almost entirely locally could save thousands of lives and avoidable expenses in already impoverished communities.
What it does
QuantumDx provides a secure and private way to locally triage patients based on symptoms and vitals that more easily accessible to track in remote communities. It utilizes quantum computing to probabilistically determine which diagnoses are most likely. This method works better than ML models on small datasets (which are matters when dealing with rare diseases with few samples) and performs better at minute-pattern detection, which helps differentiate similar diseases in areas where multiple pandemics are co-concurrent. Additionally, it has the ability to be set up and used locally (with our quantum model only requiring .csv data with patient vitals, symptoms, etc.) which is essential in dealing with deadly diseases that need quick triaging. Since all the encoding is done locally, too, patient data is never uploaded to the cloud and privacy is ensured.
How we built it
Utilizing Qiskit to create a ZZ Feature Map and a Hillbert Space, we were able to use quantum entanglement and gates to detect how feature's affect each other's quantum states. Unlike traditional vector embedding, quantum states work particularly well with detecting cooccurrences of features to help properly predict and classify diagnoses, especially within smaller (less than 100 samples) datasets.
We then created a frontend web app with Vercel and hosted our quantum model on Railway DB, which allows users to input data, which is turned into a quantum signature (similar to a vector embedding with ML models). This signature is irreversible to ensure patient privacy and gets sent to our cloud model, which returns the probabilities of a negative and positive diagnosis. All of this is displayed on our front-end, alongside some model comparisons, hyperparameters, and other general information.
Challenges we ran into
Finding publicly accessible data with patient information is obviously very difficult for privacy reasons. We eventually did find data that fit our use case, which we then used to build our quantum model (3). Additionally, learning how to do quantum modeling with said data was difficult: Qiskit is a very different language from traditional ones, we had to do a deep dive into physics and quantum computing concepts, and we spent plenty of time trying to simulate quantum computing in place of having a real one (which includes determine which features and dimensionality affected run times and model performance) .
Accomplishments that we're proud of
A working quantum model that can provide probabilistic insights into a patient's diagnosis given input data (vitals, symptoms, and personal traits).
Secure and private data management, where no patient info is uploaded directly to the web in a way that can reverse engineer their inputs.
Intuitive and responsive front-end, to provide for easy testing and a playground for users. It is live hosted, too, so everyone can check it out!
What we learned
Namely, Qiskit: a software stack for quantum computing and development. But we also learned Railway DB, some Vercel (used to host the website), and a lot of patience along the way fine-tuning our model.
What's next for Quantum DX
Using real-world data in real-world scenarios, as well as real quantum computing (as opposed to simulations). This should speed up the data-processing time immensely. Additionally, creating a more streamlined dashboard that prioritizes healthcare usage (and not necessarily the informative, fancy frontend we used for this hackathon) and efficiency would be ideal for practical usage.
Sources and Further Reading
Leptospirosis' Impact
https://pmc.ncbi.nlm.nih.gov/articles/PMC9117097/ (1)
https://pmc.ncbi.nlm.nih.gov/articles/PMC10482283/ (2)
Data
https://data.mendeley.com/datasets/zmxmp42g9p/4 (3)
Technology
Built With
- python
- qiskit
- quantum
- railway
- react
- typescript
- vercel
Log in or sign up for Devpost to join the conversation.