[00:00:00] Speaker 03:
Our first case is dental monitoring versus aligned technology, 2024-2270. Mr. Perry.

[00:00:10] Speaker 00:
Thank you, Your Honor. May it please the Court.

[00:00:14] Speaker 00:
Claim 14 of the 248 patent discloses a method of determining the amplitude of separation between orthodontic aligners and teeth based on a cell phone image. This invention marked a paradigm shift from a qualitative assessment, which human orthodontists performed previously, to quantitative assessments, including the extent and scale of the mean and maximal spacings between the teeth and the aligners.

[00:00:45] Speaker 03:
But it basically involves taking a picture and comparing a distance between a tooth and an aligner and then using a cell phone.

[00:00:58] Speaker 00:
That is part of it, Your Honor, and it goes on to additional claims from that, including assigning the value to the images in the historical data set of more than 1,000 images.

[00:01:09] Speaker 00:
something that didn't exist at the time. Teeth are very difficult to image, as Dr. Mongin testified in his declaration.

[00:01:16] Speaker 00:
There is the creation of the learning base in the dependent claims, the assignment of a value, a numerical value to the separation, including preferably the mean and maximum, which, of course, is required by this claim, Claim 14, which refers to the amplitude of separation. That's the difference between a mean and a maximum. No human dentist had ever calculated the mean or maximum or amplitude according to the record in this case.

[00:01:41] Speaker 03:
Calculations seem rather abstract.

[00:01:45] Speaker 00:
Your Honor, the abstraction would be take a picture of teeth and see if the aligners fit. That's what the district court said the claim recites.

[00:01:52] Speaker 00:
actual claims, the actual claim language has expanded in the spec.

[00:01:56] Speaker 02:
But when the orthodontist looks at it and says, oh, it doesn't quite fit, and doesn't the orthodontist say, adjust it by, and then the next word is a number?

[00:02:08] Speaker 00:
No. No, Your Honor. If the aligner doesn't quite fit, the orthodontist either continues treatment with that aligner set until the tooth moves into place, or the orders a new set of aligners based on a computer model of the teeth, which would be a separate process. And this is disclosed in the specification. For example, the average patient has 20 aligners assigned initially, but actually ends up with approximately 46 aligners because teeth don't move in the same way that they're expected.

[00:02:39] Speaker 02:
You started your argument by saying there was a shift from qualitative to quantitative. Yes, Your Honor. You returned to that, which triggered... I'm trying to understand how when either by adjustment or by creating something new, numbers don't enter into the picture.

[00:02:55] Speaker 00:
So, Your Honor, there is no evidence, and in fact it's not true, that dentists measured the space between teeth and aligners. Dentists used either the rocking method, which is to put a finger on the aligner and just see if it fits, or the eyeball method. And Dr. Cusnoto testified to this, as did Dr. Valli.

[00:03:13] Speaker 00:
That was not quantitative. That was qualitative. It was subjective. The same orthodontist on a Monday and on a Friday would assess it differently.

[00:03:20] Speaker 02:
But when you've gone beyond what the thumb or the eyeball teaches you to the next stage, I need to change something or make something. Don't numbers have to come into that?

[00:03:31] Speaker 00:
Yes, Your Honor, but that's what the model does. The underlying technology has a 3D model of the teeth that's taken by an imaging machine.

[00:03:40] Speaker 00:
And if the adjustment is not proceeding as planned, then the model needs to be changed. That's what the aligners are created from. That is disclosed elsewhere in the spec. There actually are separate patents related to the treatment method that follows from that. But the quantitative piece is not present until this invention. No dentist had measured the amount of separation, the extent or scale of separation, to use the language of the specification, before this patent. And the advantages, they're disclosed in the patent. There are two. One, it's particularly reliable because the same dentist may assess a space differently on a Monday or a Friday or between myself and Mr. Mitchell because you're looking at different things.

[00:04:21] Speaker 00:
Second, it lessens the need for in-person dental visits. And this is explicitly disclosed as the problem that is sought to be solved and the solution for the problem in the specification for the 248 patent. Because without a quantitative assessment, The dentist cannot decide whether to see a patient or to advance to the next aligner, whereas a quantitative assessment permits that. A qualitative assessment requires the in-person visit to the dentist. And that's the major advantage of this patent, which is part of a suite of patents that relate to remote provision of orthodontic care.

[00:04:56] Speaker 00:
And the key invention here... The quantification of the spacing is a new thing. It didn't exist in the world before this application was filed before 2017. And it wasn't possible without the technological tools that are described in the district court's terms with great particularity in this patent.

[00:05:19] Speaker 00:
And to get back to the abstraction question, Judge Lurie, again, if it were take a picture and make it an assessment, that would be, you know, closer to the other side of the line. But this one is the dependent claims, at least, the creation of the learning base, the training of the learning base on the historical images, the assignment of a value to each tooth, or at least a tooth, in each image based on that spacing. So again, the quantification comes on the input side so that it's available on the output side. That also did not exist in the universe.

[00:05:49] Speaker 00:
There was no off-the-shelf product that could create this. In 2017, no orthodontist could walk into the store and order up an assessment method that quantified this space. This invention taught the world how to do it. This invention, which was created at tremendous expense, this is undisputed by our clients, was the first to do it. In concluding otherwise, Your Honors, the district court basically committed three errors that ripple through this court's 101 jurisprudence.

[00:06:20] Speaker 00:
The first one, respectfully, was to overgeneralize the claims. to say that these are collect, analyze, and determine claims, when in fact these claims are written, particularly the dependent claims, much more specifically.

[00:06:36] Speaker 00:
The images have to have an aligner in the service position, that is on the teeth. They have to be taken with a cell phone. They have to have its values assigned to them. And most importantly, independent claim 14, the amplitude of separations.

[00:06:52] Speaker 00:
quantification of the spacing must be made using the method. That is far different than simply look at teeth and see if there's a space. That is a diagnostic method to determine whether or not there is sufficient space to allow, for example, moving on to the next aligner or to require coming into the dentist for a new 3B model to be taken.

[00:07:14] Speaker 00:
And The second error the court made was related to that, which was really failing to recognize that this is a diagnostic method. The court looked at this as a do-it-on-a-computer claim, automating things that humans could already do.

[00:07:29] Speaker 00:
The court said that in so many words. The record, Judge Alsop said, reflects that orthodontists had long looked at spaces. Well, that's true, but it's irrelevant. These are not alphabetization claims or sorting claims or counting claims. They're not speeding up things that dentists could do. They are doing things that no human dentist could do then and no human dentist can do now. Without the addition of this technological innovation, no orthodontist can do what is claimed in the method.

[00:08:01] Speaker 03:
One can say the same thing with respect to all computerized processes because they work so fast.

[00:08:07] Speaker 00:
Your Honor, I completely agree that there's a difference between the speed of a process and the existence of a process. So if the question is adding up a string of numbers to get to the end, which is what a calculator or a computer does, that is speeding up something that I can do with a pencil and paper.

[00:08:27] Speaker 00:
But if the question is determining the amplitude, the difference between the mean and maximal separation between a clear aligner and a single tooth based on a cell phone image, that is something that I and no human orthodontist can do with a pencil and paper, a protractor, a ruler, or anything else. It cannot be done without the technological advancement and innovation described and claimed in these patents. And that's the difference between a speeding up claim which is recentive or electric power or one of those cases, and a new thing in the world claim, which is cardio net or micro and those kinds of cases.

[00:09:03] Speaker 00:
And I would submit that that is one of the dividing lines in this court's one-on-one cases. When we have computer-implemented inventions, it's whether the computer is simply automating, routinizing, or speeding up, adding efficiency on the one hand, ineligible, or is the computer necessary to take a leap forward Beyond what a human could ever do with unlimited time, No human could do what this invention does without technology. That's an invention. That's what the patent system is designed to protect is innovations that harness new technologies to solve problems that cannot be dissolved by human beings with their minds and pens and papers.

[00:09:41] Speaker 02:
You talked a couple of times about means and averages.

[00:09:48] Speaker 02:
Two questions, I guess. Well, I don't know. Where does that come from? Is that in the claim?

[00:09:53] Speaker 00:
Your Honor, the claim is an amplitude.

[00:09:57] Speaker 02:
Value for an attribute.

[00:09:59] Speaker 00:
So the claim has two quantifications. First is the value, which must be assigned to the historical images in the learning base and then assessed in the analysis image. And a value according to the specification may, for all of the claims, remember this is one subset of a broader invention, may include things like tooth shape, tooth color, type of tooth, molar, canine, whatever. For the 248 invention, the pertinent part of the specification, which begins on column 27.

[00:10:33] Speaker 02:
Start with the claim. I guess I want to focus on the... Yes, Your Honor. Just on the claim. You know, I don't see things here about mean and average.

[00:10:42] Speaker 00:
So...

[00:10:43] Speaker 00:
There's two different issues, value and mean and average.

[00:10:47] Speaker 00:
The value is what is assigned to the tooth attribute. That is a numerical number assigned by a human being, preferably an orthodontist, according to the claim, to the image.

[00:10:58] Speaker 02:
And those values can be arbitrary, you know, 7 for molars and 2 for incisors or… They can be something… 174 for broken teeth.

[00:11:08] Speaker 00:
Correct, Your Honor. The values themselves are not set, but they must be established for this claim in at least 1,000 images for at least one tooth show a value of the separation between the tooth and the aligner because the images are taken with the aligner in service position, that is, in the mouth. Right.

[00:11:25] Speaker 00:
The second half of your question, and this is Clause 4 of Claim 14, and this is really the kicker of this invention, if you will, Judge Toronto, a determination as a function of said probability, and this is a deep learning machine, so it is taking the probability based on those assessed values, of an amplitude of said separation.

[00:11:48] Speaker 00:
Amplitude is an ordinary English word, which means the difference between between the mean and maximal amount of something. And if we go to the specification, we find in column 28, starting at line 48, that the tooth attribute, that is the separation, is chosen from the group formed by a maximal separation and a mean separation and then the measurement of the difference, which is an amplitude. And that's described in column 28.

[00:12:17] Speaker 00:
And in column 30...

[00:12:20] Speaker 00:
The specification links up the two points of your question, Judge Toronto, starting at line 42. The value of the tooth attribute, that is the separation for this claim, provides an informational item relating to the shape of the aligner, and then the value can be a measurement of the separation, for example, a measurement of the maximum separation or the mean separation for the tooth.

[00:12:47] Speaker 02:
And just so the mean and maximum are measures applied to the at least 1,000 examples?

[00:12:58] Speaker 03:
Yes.

[00:12:59] Speaker 02:
Okay. They're not mouth specific.

[00:13:01] Speaker 03:
They are.

[00:13:01] Speaker 02:
In my mouth, you're not measuring a mean or an average of distance. I don't even know what those words mean.

[00:13:08] Speaker 00:
Not correct, Your Honor. For each of the at least 1,000 images.

[00:13:13] Speaker 00:
at least one tooth must be assigned a value of the mean and maximal separation between that tooth and the aligner for the historical images. For the analysis image, the deep learning device then assesses, based on that learning base, the same thing, the mean and maximal value, which is...

[00:13:35] Speaker 00:
this is line 50 of column 30, the extent of the separation of the aligner or tooth represented in that analysis tooth zone. That's the method. In other words, the historical database has a mean and a maximal for a tooth in every one or all teeth, but at least one tooth in every image. Then the deep learning device can use that data set to evaluate the extent of the separation for the particular tooth being imaged in the patient under question. Does that make sense? Does that answer your question?

[00:14:03] Speaker 02:
I think so.

[00:14:05] Speaker 00:
Thank you, Your Honor.

[00:14:06] Speaker 02:
The separation between tooth 18 and 17, there's a statistical measure in the database, and then you compare that to what's in my mouth.

[00:14:13] Speaker 00:
Correct, Your Honor.

[00:14:15] Speaker 00:
If I may reserve the balance of my time.

[00:14:16] Speaker 03:
Counselor, you're well into your rebuttal time. We'll give you three minutes back. Thank you, sir. Mr. Bagatelle.

[00:14:37] Speaker 01:
May it please the Court, Dan Bagatelle with Jorge Santana for Align.

[00:14:41] Speaker 01:
The District Court correctly held that the claims were directed to an abstract idea and were not saved by any inventive concept. Mr. Perry focused on Claim 14, so I'll start with that. That claim is directed to collecting and analyzing data regarding aligner fit. That's an abstract idea under electric power and many other cases of this court. It does not refer to a method of treatment or a use of the analysis. It just calls for acquiring and analyzing the data and outputting a number.

[00:15:12] Speaker 01:
That's all abstract. What they argued below, and even in their opening brief, was that the novelty here was using some sort of a deep learning device, using a learning base of 1,000 images.

[00:15:26] Speaker 01:
All that was old, however.

[00:15:29] Speaker 01:
The computer was old. The algorithm was old. The camera was old.

[00:15:34] Speaker 01:
The algorithms that were straight out of column 16 could be a Google algorithm, Microsoft algorithm. All that was completely new. All they're saying is that it's something new in this field.

[00:15:44] Speaker 03:
Now we hear... Mr. Perry says it's not just the fact that the computer is faster.

[00:15:50] Speaker 01:
Well... Mr. Perry is now focusing on what he calls the amplitude of separation. This is a relatively new argument. There was no construction of amplitude of separation. Amplitude just refers to some difference between some things. So basically, you're returning a number, respecting the value. But what did orthodontists do historically? You can look at the picture on page 5 of the Red Brief. They were looking at it to see if it was less than a millimeter. That's what they were doing historically. How is the annotated database created?

[00:16:22] Speaker 01:
According to the patent, take a look at columns 29, 30. That's created by an orthodontist or some other expert in the field assigning those values. Of course it was done by humans and could be done by humans. That's how the database is created.

[00:16:38] Speaker 01:
So all you're doing is adding new data to the old algorithm and the old – you know, image acquisition machine, which is simply a cell phone, old hardware, they're simply doing it on a computer, just the same as in Recentive and Ronsolier.

[00:16:59] Speaker 01:
The idea that there's something new about quantitatively superior, yes, computers are faster, yes, computers can be a little bit more consistent. That's inherent in the nature of computers.

[00:17:09] Speaker 01:
But to get where they're going, they had to create a learning base. And to create the learning base, they had to do it manually.

[00:17:15] Speaker 01:
You can see that throughout the patent. If you do a search on the word manual, you'll see it numerous times. That's how the learning base is created. So at its core, the claims call for applying existing hardware, existing deep learning image analysis algorithms to just another field. And that is not patent eligible at step one or step two, even if they were the first to do it in a field.

[00:17:39] Speaker 01:
in Mr. Perry's brief, he's focused on some policy arguments saying that this is going to be a disaster and that the district court has created some sort of patent-free zones for AI. That's simply not the case. We never argued for that. In fact, we argued to the contrary at the summary judgment hearing, and the district court never held that.

[00:17:58] Speaker 01:
We are not suggesting that you cannot have any patents that relate to AI. If you use AI to develop a better camera or to develop a better aligner, or to develop a better method of treatment, that may be patent eligible. If you improve the AI by making algorithms that are uniquely tailored to a particular field, you may be able to patent that if you claim it in a non-abstract way. But what you can't do is simply apply existing generic AI to a new set of data.

[00:18:28] Speaker 01:
Not only has this court held so in Resentive and Ron Solaire, but it would be terrible public policy because you'd have a gold rush. Everyone would rush into the field trying to get a blocking patent and prevent anybody else from using AI in the field. And that would be a policy disaster.

[00:18:45] Speaker 01:
The court has any further questions regarding the 101 issue?

[00:18:49] Speaker 03:
What's the test for a software patent being patent eligible?

[00:18:58] Speaker 01:
Well, I don't think we have something from the court saying that it is not categorically ineligible. I remember Enfish very well.

[00:19:06] Speaker 01:
So that is not the test, simply that it's software. But it has to be something that's improving the machine, improving the computing analysis. It could be a new database.

[00:19:15] Speaker 03:
And is that done by something other than software in order to make it eligible? Or can it be done through software?

[00:19:23] Speaker 01:
I think there's not a bright line rule saying that software is patent ineligible. There has to be something But if you have to improve the technology... I'm sorry, in fact, Enfish was such a case.

[00:19:35] Speaker 02:
Enfish said... It was one of the basic functions of the computer that the court said, by doing a database different... Self-referential. And that's all software, right?

[00:19:45] Speaker 01:
Yes, a self-referential database. And if you come up with a new AI algorithm... that is uniquely tailored to something, that conceivably could be patent eligible. It depends on how you claim it. You can't just say, wouldn't it be nice to do something a little better? But if you come up with a particular way, we're going to use a little bit of this method, a little bit of that method, and combine it in this way, it works especially well in this particular field. That conceivably could be patent eligible. That's not what we have here. We have completely generic AI, pre-existing AI, pre-existing deep learning devices.

[00:20:19] Speaker 01:
There's absolutely nothing there. All Mr. Perry has said today, and this is a new argument, that somehow creating a measurement of an amplitude of separation is unique. It's not unique.

[00:20:32] Speaker 01:
Orthodontists were measuring that before, and the algorithms are the same that you would use to determine cats from dogs or any other distinction between two different groups, or measuring the size of a mountain range or a geological formation. It's simply new data on an old machine. And this court has held, consistent with Alice, that that is not patent eligible.

[00:20:56] Speaker 01:
If the court would like, I can turn to our alternative ground to support the judgment under Section 112.

[00:21:04] Speaker 01:
Okay.

[00:21:05] Speaker 01:
Well, the basic problem here is that the claims recite a deep learning device. And the specification makes clear that the deep learning device need not be or even include a neural network. And the problem with that is that nobody knew in 2017 and nobody knows now what such an animal would be. Every deep learning device known to humankind contains a neural network. That is a problem under Section 112A because you need to describe the full scope of your invention.

[00:21:40] Speaker 01:
They haven't. They have only described conventional neural network-based deep learning devices. It's also a problem under 112b because it's indefinite. We don't know and could not know what that would be. The only thing they've ever pointed to is something called a support vector machine. It's not mentioned in the patent. It's considered a type of shallow learning. It basically will help you classify a cat from a dog once you've looked at a but it's not a deep learning device. And their expert, Dr. Mangan, admitted that outside the scope of the patent, nobody would call that a deep learning device.

[00:22:14] Speaker 01:
The patent does refer to something called an RCNN, a region-based convolutional neural network, and it's a hybrid. Basically, you have a neural network. And you use that to get all the features. And then you use something called the support vector machine to kind of distinguish a cat from a dog, that sort of last stage of classifying. So it's a hybrid. But it always includes a neural network.

[00:22:38] Speaker 01:
And nobody knows what a deep learning device without a neural network would be. The main point that Mr. Perry has made is to say somehow that's not before this court. Well, it's an alternative ground to support the judgment of invalidity. This court routinely considers alternative grounds for invalidity. Sometimes it's anticipation rather than 101 or 112 rather than 101. Numerous, numerous cases. The important point is we couldn't and didn't appeal because for two reasons. First of all...

[00:23:10] Speaker 01:
We didn't lose. We prevailed and got all the relief we wanted. We got all the claims that are now at issue invalidated. Second of all, the district court didn't reach the issue. So the case on which they rely, Kempke, is inapposite. We couldn't cross-appeal under Bailey v. Dark Container in the years of precedence that say you cannot appeal in this circumstance. So we didn't. But it is an alternative ground to support the judgment if you need to reach it. For the reasons I've explained, we are confident that the claims are valid under Section 1, and you should affirm that.

[00:23:41] Speaker 02:
This body of law is not as fresh in my mind as it might be. Do I understand this argument is an argument that the claims are sufficiently broad that they cover unknown things that might arise in the future?

[00:23:58] Speaker 01:
Is that your argument about why there's an insufficient – Well, they've taken the position both in this litigation and in paralytic litigation in the District of Delaware before Judge Bryson that the claims – are not limited to deep learning devices with a neural network. So it's their construction that deep learning devices include more than that.

[00:24:16] Speaker 02:
Your principle of written description is that what's problematic here is that the claims go beyond neural networks.

[00:24:28] Speaker 02:
nobody knows what there might be that is within the claim but not in a neural network. Do we have cases that say that's a problem when you, the challenger, can't even identify, because it's unknown, a yet uninvented thing that's within the claim and why that's true? I don't know. Well, no, I think... It's like a sort of frighteningly broad written description principle.

[00:24:58] Speaker 01:
No, but I think what you do have... The cases that I recall are cases like Lizard Tech and Rivera versus ITC.

[00:25:04] Speaker 02:
Yeah, but didn't they know... You could identify what was not... You could name, couldn't you, in those cases, what was...

[00:25:14] Speaker 01:
I'll give you another example in the pharmaceutical space.

[00:25:16] Speaker 02:
People claim... Was it about transforms or something?

[00:25:20] Speaker 01:
I believe so. There are also cases that involve, well, medical devices, but more importantly, I think in the pharmaceutical space, you have cases where people claim genuses.

[00:25:31] Speaker 01:
And... you know, you've claimed much more than you actually possess. And this court in a whole series of cases has held that that's a written description problem when you have claimed more than one of the school in the art would recognize that you have. If you claim a broad genus, and this is a broad genus of deep learning devices that supposedly is uniquely defined for this case because their own expert said, we don't know of any, he's never heard of a deep learning device that doesn't use a neural network. So I think it fits within that line of cases. But it's also indefinite. So what does it cover?

[00:26:02] Speaker 01:
One of us on the art would not know.

[00:26:05] Speaker 01:
So we think that's a second independent problem in addition to the 101. But I think the simpler solution is that this is squarely within the scope of Recentive and Ron Solaire because it's just bringing old technology to a new field.

[00:26:23] Speaker 01:
The Board has no further questions.

[00:26:24] Speaker 03:
Thank you, Mr. Bagatelle. Mr. Perry has up to three minutes for rebuttal if he needs it.

[00:26:33] Speaker 00:
Thank you, Your Honor. Two quick points.

[00:26:36] Speaker 00:
First, my friend Mr. Bagatelle said that using AI to develop a better method of treatment may be patent eligible. Of course it may be.

[00:26:45] Speaker 00:
Just as using AI to develop a better method of diagnosis may be patent eligible. That's Cardionet with AI layered on top, and that, we submit, is this case. Judge Lurie, you asked what's the software test. I think test is too strong a word, but the through line I would draw in the software cases that do come over to AI is if a human being can do it on a pencil and paper or in their head, then doing it with software with nothing else is almost certainly ineligible. But if a person can't do it and needs a computer or a quantum mechanics machine or an electron microscope or some other tool of technology to do it, then that thing might well be eligible.

[00:27:24] Speaker 00:
And that's the fuzzy dividing line, and it is fuzzy, but it is here we have humans can't do it. And let me end on this. My friend said that orthodontists measured separation before the patent, before the priority date in 2017. That is false. It is wrong. It is not supported. And we had a summary judgment record here, Your Honor. There is no evidence to support that. The evidence is to the contrary. I will point the court to Mr. Mongan's declaration, Appendix 819 to 20, Paragraph 20.

[00:27:54] Speaker 00:
And Dr. Cusnotos' declaration, Appendix 868 to Paragraph 43, and Dr. Cusnotos is particularly telling, it says no human orthodontist could do this before 2017. This is a technological advance that allows orthodontists to do for the first time something they never have been able to do in history. That, we submit, is an invention.

[00:28:15] Speaker 00:
That is patent eligible. There may be other issues.

[00:28:19] Speaker 00:
112, this court has no jurisdiction to review the 112 issue. This is a stipulated judgment after a patent showdown on a single issue. There is no jurisdiction. I'm happy to go into the merits, but he's wrong on that as well. But on 101, this is the other side of the field for Mercentov. This is a new thing that never existed in the universe before. The how of it is described in great detail and explained for a person with a skill in the art to do that innovative method. Thank you, Your Honor.

[00:28:45] Speaker 03:
Thank you, Coach. Both counsel, the case is submitted.