Federal Circuit Finds AI-Based Dental Monitoring Patent Claims Abstract Under Section 101

PatentNext Takeaway: The Federal Circuit’s decision in Dental Monitoring SAS v. Align Technology, Inc., No. 2024-2270 (Fed. Cir. July 7, 2026), provides another important warning for patent applicants seeking protection for artificial intelligence (AI) and machine-learning technologies. Applying a generic AI or “deep learning” system to a specialized medical or dental problem—even where the AI is trained using a large, field-specific dataset—may not be enough to establish patent eligibility under 35 U.S.C. § 101.

The decision reinforces the Federal Circuit’s recent holding in Recentive Analytics, Inc. v. Fox Corp. that using conventional machine learning in a new field of use does not, standing alone, constitute a technological improvement. Patent practitioners should therefore draft AI-related claims to capture the specific technological mechanism or improvement that produces the claimed result, rather than merely claiming the collection of data, analysis of that data using AI, and generation of a result. Particularly important, Dental Monitoring demonstrates that purported improvements described in the specification or argued during litigation may carry little weight when those improvements are not actually required by the claims.

****

The Federal Circuit recently affirmed a Northern District of California decision finding several AI-based dental monitoring patent claims invalid under 35 U.S.C. § 101. In Dental Monitoring SAS v. Align Technology, Inc., a three-judge Federal Circuit panel held that claims directed to analyzing dental images using a “deep learning device” were directed to abstract ideas and did not contain an inventive concept sufficient to render them patent eligible. The July 7, 2026 decision is nonprecedential, but its reasoning closely follows the Federal Circuit’s precedential 2025 decision in Recentive Analytics, Inc. v. Fox Corp. and provides an important example of how that case may be applied to AI inventions in the medical technology space.

The decision is also noteworthy because the claimed inventions were not merely generic business applications of AI. They concerned orthodontic treatment and dental-image analysis—areas involving physical patients, dental appliances, imaging devices, and clinical assessment. Nevertheless, those real-world aspects were insufficient because, in the Federal Circuit’s view, the claims ultimately used conventional machine learning to perform information collection and analysis.

Background: AI-Based Orthodontic Monitoring

Dental Monitoring owns U.S. Patent Nos. 11,049,248 (the “’248 patent”) and 10,755,409 (the “’409 patent”). Both patents concern analysis of images of a patient’s dental arch.

The ’248 patent generally concerns assessing the shape or fit of an orthodontic aligner using a “deep learning device,” while the ’409 patent concerns acquiring and analyzing images of a dental arch using a deep learning device. The patents describe a deep learning device as a machine-learning device that can be trained to recognize patterns in images.

Dental Monitoring sued Align Technology in 2022, alleging that Align’s Invisalign Virtual Care AI platform infringed claims of the patents. The district court established what it called a “patent showdown,” requiring each party to select a representative claim for discovery and cross-motions for summary judgment. Dental Monitoring selected claim 14 of the ’248 patent, while Align selected claim 12 of the ’409 patent. The parties stipulated that the rulings concerning those claims would also apply to certain additional claims.

The resulting appeal addressed claims 1 and 14 of the ’248 patent and claims 1, 7, and 12 of the ’409 patent.

The ’248 Patent: Assessing Aligner Fit Using Deep Learning

Claim 1 of the ’248 patent generally required acquiring an image of an orthodontic aligner being worn by a patient using a cellphone and analyzing the image using a trained deep learning device to determine an attribute relating to the separation between a tooth and the aligner.

Dependent claim 14 provided additional details concerning the AI training process. Among other things, the claim required a learning base containing more than 1,000 dental-arch images, training a deep learning device using that learning base, submitting an analysis image to the trained device, determining probabilities concerning tooth zones and tooth attributes, and using those probabilities to determine the amplitude of separation between a tooth and the aligner.

The ’409 Patent: Evaluating Dental Images and Providing Feedback

Claim 1 of the ’409 patent generally required acquiring an image of a patient’s dental arch, analyzing the image with a trained deep learning device, determining an image-attribute value, comparing that value with a setpoint, and sending an informational message based on the comparison so that the operator could determine whether a new image should be acquired.

Claim 7 similarly required creation of a learning base containing more than 1,000 historical dental-arch images and training a deep learning device using that data. Claim 12 further required that the information message be sent by the image-acquisition apparatus itself.

Align challenged the claims under § 101. The district court agreed, finding them patent ineligible under the Supreme Court’s two-step Alice framework. Contemporary reporting on the decision likewise highlighted the Federal Circuit’s conclusion that using a specifically trained deep-learning device did not supply the technological improvement necessary to save the claims.

The Federal Circuit’s Section 101 Analysis

Under Alice Corp. v. CLS Bank International, courts first determine whether a claim is directed to a patent-ineligible concept such as an abstract idea. If so, the court considers whether additional claim elements provide an “inventive concept” sufficient to transform the abstract idea into patent-eligible subject matter.

The Federal Circuit found the Dental Monitoring claims deficient at both steps.

Alice Step One: Collecting and Analyzing Dental Data Remained Abstract

At Alice step one, the Federal Circuit characterized claim 14 of the ’248 patent as directed to the abstract idea of collecting and analyzing image information using a deep learning device.

Similarly, the court characterized the ’409 patent claims as involving image acquisition, analysis using a deep learning device, comparison against a setpoint, and transmission of the resulting information.

The Federal Circuit concluded that both fell within the familiar category of claims involving collection of information, analysis of information, and presentation of the resulting information—claims the court has repeatedly treated as abstract under cases such as Electric Power Group, LLC v. Alstom S.A.

Merely Adding a “Deep Learning Device” Was Not Enough

Central to the court’s analysis was its conclusion that use of a “deep learning device” did not materially alter the § 101 analysis.

The patent specifications described the deep learning device broadly and indicated that it could employ known neural-network technologies designed for image classification or object detection. Accordingly, the Federal Circuit viewed the claims as applying generic machine learning to a new data environment—dental-arch image analysis.

That reasoning closely follows the Federal Circuit’s decision in Recentive Analytics, Inc. v. Fox Corp., where the court held that applying existing machine-learning techniques to a new field of use does not make an otherwise abstract process patent eligible. As PatentNext previously discussed, Recentive emphasized that claims should identify a specific technological improvement rather than merely instructing that an existing process be performed using machine learning. See PatentNext: Federal Circuit Rejects AI Claims Lacking Technical Detail.

A Specialized Training Dataset Did Not Save the Claims

One particularly important aspect of Dental Monitoring concerns AI model training.

The claims did not merely say “use AI.” Certain claims required the deep learning device to be trained using a specific learning base containing more than 1,000 dental-arch images.

That detail was still insufficient.

The Federal Circuit explained that training a machine-learning system on a particular subset of relevant data is inherent in the nature of machine learning. Accordingly, restricting the training data to dental images did not itself constitute a technological improvement.

This aspect of the opinion has potentially broad implications for AI patent drafting. Simply claiming:

train a machine-learning model using a domain-specific training dataset

may provide little protection against a § 101 challenge if the claimed training process otherwise employs conventional machine-learning technology.

The stronger eligibility position is generally to claim what is technically different about the training process, model architecture, data representation, feature generation, inference procedure, or resulting computer functionality.

The Claimed Technological Improvement Must Actually Be in the Claim

Dental Monitoring argued that its technology provided a more precise and objective method for quantitatively measuring separation between individual teeth and an orthodontic aligner.

The Federal Circuit rejected the argument for an important reason: the asserted technological improvement was not required by the claim language.

The court observed that nothing in claim 14 required the particular quantitative assessment Dental Monitoring relied upon in its argument. Instead, the claimed determination could encompass something an orthodontist was already capable of determining visually. The use of machine learning might make that determination faster or more efficient, but the Federal Circuit held that improved speed and efficiency resulting from generic machine learning did not make the claim patent eligible.

This may be one of the most important drafting lessons from the case.

A patent specification may describe an impressive technical advance. But if the independent claims abstract away the very mechanism responsible for that advance, the patentee may later be unable to rely on the unclaimed technical details to establish eligibility.

Alice Step Two: No Inventive Concept in Generic Deep Learning

The claims fared no better under Alice step two.

The Federal Circuit concluded that the “deep learning device” was conventional because the specifications themselves indicated that it could be implemented using known and commercially available neural-network technologies.

And because the device was being used to perform the very abstract process identified at Alice step one—capturing, analyzing, and presenting information—it did not constitute an additional inventive concept capable of transforming the abstract idea into patent-eligible subject matter.

Dental Monitoring argued that the specialized training procedure provided the inventive concept. Again, the Federal Circuit disagreed. Training a machine-learning model using appropriate training data was inherent in conventional machine learning and did not make the underlying AI technology non-generic.

The court also rejected the argument that the overall use of deep learning for orthodontic treatment may have been unconventional at the time of invention. Under Alice step two, the relevant question is not simply whether the claimed combination as a whole was new or unconventional. Rather, there must be something that transforms the abstract idea into “significantly more.”

The court ultimately affirmed the district court’s judgment that claims 1, 7, and 12 of the ’409 patent and claims 1 and 14 of the ’248 patent were patent ineligible under § 101. Because § 101 resolved the dispute, the court did not need to reach Align’s separate § 112 invalidity arguments.

Dental Monitoring Extends the Recentive Lesson to Medical AI

Although Dental Monitoring is nonprecedential, its reliance on Recentive is significant.

Recentive involved machine learning applied to television scheduling and network mapping. Dental Monitoring involved AI applied to dental images and orthodontic treatment. Yet the Federal Circuit applied essentially the same principle:

Changing the field in which conventional machine learning operates does not itself create patent-eligible technology.

That principle is particularly important for medical-device, diagnostic, imaging, and digital-health inventions. The fact that an AI system analyzes medically significant data, assists a clinician, or interacts with physical medical equipment does not automatically establish eligibility.

Instead, applicants should ask a more fundamental question: What technological mechanism makes this AI system different from simply applying conventional machine learning to medical data?

Practitioner Tips: Drafting AI Patent Claims to Better Withstand Section 101

The Dental Monitoring decision provides several practical lessons for drafting AI and machine-learning patent applications.

1. Claim the Technological Improvement—Not Merely the Use of AI

Avoid claims that can be summarized as:

collect data → apply AI → determine information → report the result.

Identify the particular technological feature responsible for the improvement and place that feature in the claim. See PatentNext: How to Patent Software Inventions: Show an “Improvement.” Depending on the invention, that could include a particular model architecture, preprocessing operation, feature representation, inference technique, sensor-processing technique, data structure, feedback mechanism, or computational arrangement.

2. Do Not Rely on a Specialized Training Dataset Alone

A claim requiring hundreds, thousands, or millions of domain-specific training examples may still describe conventional machine learning.

Instead, consider whether the invention includes a nonconventional manner of generating, labeling, organizing, transforming, weighting, or using the training data and claim that mechanism where supported.

3. Put the Asserted Technical Advantage Into the Claim

If eligibility depends on improved image resolution, reduced computational requirements, a particular quantitative measurement, improved signal quality, better sensor operation, increased prediction accuracy through a particular technical technique, or another technological benefit, the claims should contain the limitations responsible for that benefit.

Dental Monitoring illustrates the danger of relying during litigation on an improvement that the claim does not actually require.

4. Explain the Technical “How” in the Specification

The specification should not merely state that AI achieves a better result. Explain how the claimed architecture or process produces that result.

Where appropriate, describe model inputs and outputs, preprocessing steps, feature extraction, model architecture, training operations, loss functions, data transformations, interactions among system components, and alternative technical implementations.

Those disclosures may provide support for narrower claims if broader claims later face a § 101 challenge.

5. Be Careful About Characterizing the AI Components as Generic

The Federal Circuit relied in part on the patents’ own disclosures to conclude that the claimed deep learning devices were conventional.

Specifications understandably disclose commercially available or conventional implementations to satisfy enablement and provide breadth. But practitioners should also clearly identify which portions of the implementation constitute the applicant’s technical contribution and how those components differ from conventional implementations.

6. Do Not Depend Solely on Speed or Efficiency

Automating a task previously performed by humans—even if AI performs it substantially faster or more accurately—may not establish eligibility when the improvement results simply from using conventional computer or machine-learning technology.

Instead, connect improvements in speed, accuracy, or efficiency to the specific technical mechanism that produces those improvements.

7. Avoid Claims That Merely Automate Human Observation or Judgment

Claims may face increased scrutiny where a court can characterize the claimed process as a human activity performed using a computer.

For medical AI inventions, consider whether the system performs a technical operation that a clinician could not practically or technically perform unaided—for example, extracting a particular machine-generated representation from raw sensor data, modifying image-acquisition parameters in response to computed characteristics, or carrying out a specifically claimed computational transformation.

8. Use Dependent Claims as Section 101 Fallback Positions

Draft dependent claims that progressively add the concrete technical implementation.

For an AI imaging invention, dependent claims might specify particular preprocessing techniques, model architectures, feature-generation processes, training mechanisms, sensor-control operations, or other technical steps.

Those limitations can provide useful fallback positions if a broader independent claim is later characterized as merely applying generic AI to an abstract process.

Conclusion

Dental Monitoring SAS v. Align Technology reinforces an increasingly clear message from the Federal Circuit regarding AI patents: AI terminology does not itself create patent eligibility.

A claim does not necessarily become technological simply because it recites a neural network, a deep learning device, a large training dataset, or application of machine learning to an important real-world field such as medicine or dentistry.

For patent practitioners, the drafting objective should therefore be to identify and claim the technological innovation beneath the AI label. What changed technically? How does the claimed system accomplish something differently? What particular architecture, processing technique, or interaction among components produces the improvement?

Those questions should be answered not only in the specification, but—where possible—in the claims themselves.

As Recentive and now Dental Monitoring demonstrate, the difference between claiming “use AI to solve this problem” and claiming a particular technological solution implemented using AI may determine whether an AI patent survives § 101.