Lecture: How to Use Automated Interpretation for Fundus Images

Cybersight Consult now supports Artificial Intelligence (AI) grading of colour fundus images attached to consultation cases.  The system is capable of detecting and visualising glaucoma, macular disease and diabetic retinopathy in less than a minute, and at no cost.

In this video, Dr. Daniel Neely provides a step-by-step demonstration of how to attach fundus images to a Cybersight Consult case for automated interpretation.

Lecturer: Dr. Daniel Neely, Professor of Ophthalmology | Indiana University School of Medicine, Indianapolis, USA

Transcript

DR NEELY: Greetings. I’m Dr. Daniel Neely with Orbis International. I’m excited to tell that you Orbis/Cybersight has partnered with the artificial intelligence company Visulytix. Through this partnership, we have now incorporated an artificial intelligence fundus photo interpretation feature into the Cybersight consult programming. In this short video, I’ll show you how to take advantage of this exciting new feature. The first thing you do is go to your home screen. Just as you normally would, with Cybersight. And you will be submitting a new patient consultation case, because you’ll want a patient photo, fundus photos, to be submitted. So we’re starting a patient case. And then we have our initial information here. And you see the list of subspecialty consultations to be submitted. Keep in mind that the AI software, which is interpretation of fundus photos and optic disc photos, will only appear when you’re under certain subspecialties. Currently those three subspecialties are the categories of retinovitreous, pediatric retina, and glaucoma. Any time you’re in one of these subspecialties, the AI software interpretation option will appear. For the purposes of this demonstration, I will go to retina, for this demo case. Selecting myself as the mentor. So that it comes to me. And we’ll put in a brief patient information here. We’ll have 50-year-old female. And let’s say that she’s a glaucoma suspect, with a history of hypertension, non-insulin-dependent diabetes, complains of decreased vision, and we’ve noticed cotton wool spots upon examination of her right eye. For visual acuity, I’ll just put in some made-up numbers here. Let’s make her right eye 6/60, and I’ll make the left eye normal, 6/6. All right. She’s a glaucoma suspect. So I’ll go ahead and add in some intraocular pressures here. I’ll put 25 on the right and I’ll put 25 on the left. And then scrolling down, medications… She’s a non-insulin-dependent diabetic, so let’s assume she’s on an oral hypoglycemic. Cup to disc ratio I will also add in. Let’s give her a moderate cup to disc ratio on both eyes of 0.5, and the posterior segment — I’m gonna make the right side abnormal, because I have a set of right eye fundus photographs. And I’ll put in the obvious cotton wool spots. And increased cup to disc ratio. We’ll leave the left eye normal. All right. So our diagnosis — we have glaucoma suspect. With cotton wool spots. Non-insulin-dependent diabetes. Treatment, none so far. And our question will be: Does this patient have diabetic retinopathy? And are they a candidate for laser or Avastin? All right? And normally this is where we would be attaching our standard photographs. And if you hold down the command button, you can select multiples. Upload those. But because this is a retina or glaucoma case, or in this case, I’ve combined both, with two different sets of patients’ funds photographs, we can also look here. Run automated interpretation. Okay? This is the Visulytix artificial intelligence interpretation. And in this case, I’ll put yes. And I will add a couple more photographs here. Same ones that I submitted earlier. And you can see those are uploaded. All right. So I have my standard photographs. I could have also put visual fields up there, et cetera. Down here, for the Visulytix AI software, you’re restricted to fundus photographs. So we want jpeg, bitmap, png, gif files. All of those can be submitted. But these need to be photographs of the retina. Photographs of the optic nerves. Now, why would we use the automated interpretation? Because normally our turnaround time target for a mentor to get back to you is going to be 24 hours or less. While you’re waiting on that detailed feedback from the mentor, you’ll be able to get almost immediate feedback from the artificial intelligence, and the AI system will begin to answer some of your questions. Which the volunteer faculty member will then further expand upon. So we’ve selected our AI interpretation images, and then we will go ahead and submit this. All right. So now the mentor receiving your case will be notified, and will get back to you within 24 hours, to send you a response. But in the mean time, within just a few minutes, you’ve received this notification that the automated interpretation results are available now. And you simply follow your email link that you received, and you can get the report almost immediately. So here’s our case with the glaucoma suspect. Non-insulin-dependent diabetes. Cotton wool spots under examination. And as we scroll down, we’ll now find that there has been attached a PDF report from the Visulytix software. I’ll select that, and that will open up. And we can see here that there are several condensed results up here at the top. Now, the first two are going to be image quality, image gradability. And we want these numbers to be quite high. Close to 1. Like they are. And you can see that on the second image, it had a little trouble with gradability. Now, what does that mean, if the gradability factor is yellow or red, instead of positive green? Well, it simply means that you want to be a little more cautious with interpreting those results, because there’s something that the system didn’t like about that photograph, and so the confidence interval is a little bit lower in that case. But you can see that one of the photographs has already been interpreted to have severe diabetic retinopathy, while the other one is normal. Now, keep in mind I submitted photographs from two different patients. And let’s just look at these now. This is image number one. Image number one was the one with the cotton wool spots. And this is being interpreted as having some significant diabetic retinopathy. So here are cotton wool spots. And you can see that the software program is picking up the exudates, in addition to the cotton wool spots. And it has highlighted those with boxes. Likewise, hemorrhages have been detected. And are being denoted as present. And you can see that they have been boxed here and here. Those same hemorrhages in the next section — there’s that one hemorrhage that was highlighted. There’s the other hemorrhage that we saw up above. But in this box, the software program is looking for microaneurysms, and you can see that some of these really small aneurysms that were less apparent in the other photographs have now been highlighted and graded with the Visulytix software. It also, on every patient, will analyze the optic disc. And while this one may be a 0.4 or 0.5, this is passing the grading system, as this particular set of photographs did not have any evidence of glaucomatous changes. We do go ahead and get a vertical cup to disc ratio on this diabetic patient, and here it’s being estimated at 0.47. So just under the threshold of 0.5, with which we consider glaucoma suspects, or 0.6, where it’s a definite abnormality. This patient, the foveal region was found to be normal. And so we have a relatively good macular anomaly score. Now, image two, image two was the glaucoma suspect. So this is a different image, different patient, that I added to our case, just to demonstrate the glaucoma findings. This patient has no visible evidence of retinopathy that we can see. But they do have this anomalous disc, which I would estimate to be somewhere around 0.5 to 0.6. So no diabetic retinopathy detected. Normal diabetic retinopathy score. Exudates, absent. None were detected visually or by the AI software. As was the case also with hemorrhages. And microaneurysms. Okay? So absolutely no evidence of diabetic retinopathy in this score. Then you get down to the disc anomaly score, and here we see that the disc anomaly score is quite high. Almost 1. It’s a 0.97. So we have a highly anomalous disc. With a score greater than 0.75, and so not only is this a glaucoma suspect, but this is also a significantly anomalous disc, and our suspicion for glaucomatous changes should be quite high. The cup to disc ratio is 0.58 by the AI software interpretation. And again, because we have no retinopathy here, we have a normal macula anomaly score. I do have another case, with a little more significant diabetic retinopathy I’d like to show you. Here’s a final example, where we can clearly see these blot-dot hemorrhages on this particular patient. And the exudates are being analyzed by the AI software. And being highlighted in each of these macular areas, in this macular area. The hemorrhages — much more numerous. And you can see them all being cataloged and highlighted here. As are the microaneurysms. So this new AI, artificial intelligence software, this is not a substitute for a formal interpretation and formal feedback from the Orbis volunteer faculty member. But it is a way for us to give you some immediate feedback and interpretation of images in certain cases. And again, this is currently limited to retinopathy, diabetic retinopathy, and glaucomatous changes of the optic nerve. So you will only see this appear when you’re submitting a glaucoma consult or a retina consult. But please do take advantage of this, and let us know if you have any questions, or if any problems come up. We’d like to hear about that as well, so that we can fine tune this, and continue down this exciting new path of artificial intelligence incorporation into Cybersight consult. Thank you for your time.

 

Download Recording

High Quality

Standard Quality

May 7, 2019

Thoughts? Please leave a comment...