Image search software helps detect cancer

Muisc

>> The Visuvi as I mentioned is a visual search engine with patent pending technology around content based image analysis. What that means is that unlike Google where you type in text to initiate a search query we allow user to upload an image and we then analyze that image to determine the content to deliver relevant results. So in this example, what would you type into Google to try and describe or find the artist who painted this?

>> Serrot. Laughter

>> Well that's cheating.

>> Laughter I just happened to know the answer.

>> But if you didn't know the answer right it becomes a little bit more challenging; truly a situation where an image is worth a 1000 words.

>> Right.

>> So with the Visuvi if you click to the next slide you can see you can take a capture of the image through any means a mobile phone, a PDA, computer to initiate a search query, our image analyzes the content of the image to deliver relevant results. The unique thing here is that there is no text, there's no meta tags involved. It's entirely computer generated image analysis. So we click to the next one. Something a little more practical and not to scare you but but being men there's a 1 in 6 chance that we could be diagnosed with prostate cancer. So this is actually a prostate biopsy. If you click to the next. One of the things that we're trying to do or where we're seeing use case in our technology is to improve the diagnosis for cancer. In prostate cancer for example the most effective means of diagnosis is a $6000 procedure, the majority of which is a human being, a pathologist that has to sit down and actually review this information. And so our technology can do a couple of things, one of which is is improve the qualitative and quantitative information around that diagnosis but more importantly is to reduce that cost. So in reducing that cost we can provide them the information but also eventually improve patient care.

>> So the way this is built you've got this Java EEE application in the back end that essentially has a hash table of of a lot of images.

>> Yeah so right now we have about a 150 million images. We're indexing at a rate of about 400 images per second, so roughly a billion a month. We're a young company, been around for less than a year. So yeah the index that we have is pretty substantial.

>> Right and the front end that you've built is that looks to me like a FXF.>> That's correct. This is as a matter of fact this is a Java FX app that we've developed and proud to say that again this very talented team put this together in a week. So what we're looking at here this is an actual biopsy slide, prostate biopsy and we can magnify this looking at the resolution here to get into some more granular detail and the use case is fairly seemingly fairly simple right? So the pathologist might see something that's unique and they want to be able to identify similar patient cases and then of course the respective diagnosis. So with our technology and again using Java, a pathologist could highlight that respective item, capture that with with this Java FX app and then use that image to initiate a search query to find related patient cases. Again this is live. This is all happening in real time.

>> Right so you just searched millions of biopsies to find similar features.

>> 90,000 images in this case in 0.3 seconds.

>> Yeah and so what you end up with is images that are kind of similar with patient outcomes, patient histories.

>> Yeah so we can click on the patient outcome here and look at the diagnosis and in this case this was cancer, unfortunately.

>> Yep. That's that's that's pretty cool and I suspect that there are more than a few people in this room who will live an extra few years just because of you guys.

Music

==== Transcribed by Automatic Sync Technologies ====