Neo

The motivation and drive are through the roof in a hack, they are precisely what helped us win the second prize at Practo Sandbox 2017. We worked on the Patient love track — A tech solution that enables patients with newer ways to benefit out of the Practo platform. Here is the tale of the solution that we built in 30 hours.

In villages and remote areas, access to sophisticated health care is scarce. Usually there is one physician for a village. We wanted to equip the doctors with tools that would help them diagnose better and also a convenient way for users on the Practo platform to perform a quick diagnosis by themselves. So we decided to build an app that could detect skin cancer and provide with a decentralised access to their records.

There were certain challenges on this path. Storing patient records which are usually images on the blockchain is not very efficient since there is a limit on the data that can stored in each block. Also, we needed a deep learning model that did not overfit to the training data and had the least false negatives. Over the course of 30 hours, we built the first version of Neo. A production grade prototype overcoming numerous hurdles.

Just like any other hack, all of us were confused as to what to build. Having attended hacks before, we had the foresight to ensure that all of us were clear on what we were going to make. So, we spent quite a lot of time re-iterating through multiple ideas and finalising what exactly we wanted to build.

We sketched out a road map to ensure that everyone is clear on their tasks. Building a working prototype was our utmost priority. We decided to train a very basic deep learning model that was fast to train in order to check the quality and diversity of the data. On the blockchain front, we planned on building the InterPlanetary File System (IPFS) that could store all the records of the patient. The design and UX was to be designed for the Android app.

Around 1:00 PM we were almost done building all the components of the app. All the API endpoints were ready. Most of them were also integrated into the app as well. Funnily enough we faced some issues while working with Kotlin. So finally at around 3:00 PM, we finished building and demoed for the first review.

On the initial page the user has an option to capture or pick a photo from the gallery. For the purpose of the hack, we put three sample skin cancer images that could be used for testing. Once the user takes a picture it is sent to our server for inference. The server then sends a JSON response based on which the user will see one of the two screens below.

The user can then upload these records onto the blockchain and get a hash key in response. This hash key (Left in above image) could be used by other doctors to retrieve the patient’s records.

Apart from skin cancer, we also wanted to integrate a Pneumonia detection module that could detect the areas in which Pneumonia was present in an X-Ray. The architecture was ready but we could not train the entire model due to infrastructure and resource constraints.

All of us had a great learning experience. Kudos to Practo for organizing an amazing hack!

If you have any feedback or questions we would love to hear them.

We will be putting up our codebase for all the components on GitHub soon.

Related posts:

Darico es un token digital basado en el valor del oro, Bitcoin y Ethereum. Aprende cómo funciona hoy en nuestra revisión. Darico, que se encuentra en línea en Darico.io, se representa a sí mismo como…

It is inconvenient to use mobile phone ringtones in some public places, usually we will set the mobile phone to vibrate mode, so that we will not miss some important calls or messages. But if the…