About the WebApp

This web app has just been fully updated and migrated to the latest cutting edge version of tensorflow 2.0 (python) and tensorflowjs (for javascript ver 1.0). The last version would need to load a much larger AI/ML model with more shards resulting in longer initial load times and slower prediction rates. The current version predicts faster and loads faster with an even smaller AI/ML model foot print. The integration of keras as a higher level API has made writing native tensorflow amazingly straight forward.

This web app runs an AI/ML algorithm completely in the browser and can use graphics acceleration for faster inference and prediction.

It uses the camera (On phone or desktop) to scan images and tries to recognize Galeazzi versus Monteggia fractures (not certified for medical use yet). No images or video are uploaded to any cloud server. All computations are processed directly on your machine.

Currently known to run on: The latest firefox, opera and chrome on desktop/laptop, Samsung galaxy S8, Mi6, OnePlus5T (on android 8.0 oreo), iphoneX, iphoneSE, iphoneC (running at least safari 11). Hardware limitations may limit the speed of the prediction or drain phone batteries quickly as AI/ML may prove demanding on certain types of hardware. The current version's initial loading time can reach less than 10-30 seconds or more depending on the internet speed (initial loading times for the older version ranged from 1-2 minutes) as the AI/ML model is deployed. Subsequent reloads will be faster as the model is cached (unless the browser is closed and restarted).

Camera Scanner

Press the round button on the lower right to start scanning.

Focus the camera on a Galeazzi or Monteggia fracture.

Real time predictions and their respective probabilites are on the upper left.

An error log module will list errors on the lower left if there are any. Otherwise it will say that web cam captures and the model predicts.

Final prediction will be at the bottom bar revealing a link to a google image search

To hide real time predictions/statistics and error log...press the button below

Current work in progress by:

Dr. Erwin John T. Carpio (Radiologist, FrontEnd Web Developer, AI Enthusiast).

I tried my hand at an AI/ML project so that I might hopefully encourage my fellow radiologists in my country or anywhere in the world to more actively participate in the growth and development of AI/ML for radiology. I hope to prove to them that the technology is not beyond their reach especially with amazing libraries like tensorflow which have made it easier and more approachable for professionals like myself to explore the amazing world and possible applications of AI/ML in my domain of expertise.

After Thoughts

Would a DICOM scanner be intersting?

This is currently a "proof of concept" (powered by tensorflow 2.0) to help inform and spread the word on the possibilities of AI/ML applications in medicine and radiology. If you have suggestions or want to collaborate on a project we can always discuss. You can find me on:

Some Resources:

Tensorflow (python)

Tensorflow.js

keras.io