Heartflow Inc v. Cleerly Inc

I. Executive Summary and Procedural Information

• Parties & Counsel:
  • Plaintiff: Heartflow, Inc. (Delaware)
  • Defendant: Cleerly, Inc. (Delaware)
  • Plaintiff's Counsel: Mayer Brown LLP
• Case Identification: 2:26-cv-00292, E.D. Tex., 04/13/2026
• Venue Allegations: Plaintiff alleges venue is proper in the Eastern District of Texas because Defendant Cleerly operates regular and established places of business within the district, specifically through "Cleerly heart scan imaging locations" that offer its services to patients and are extensions of its business operations.
• Core Dispute: Plaintiff alleges that Defendant's AI-powered cardiac imaging analysis software infringes six patents related to non-invasive cardiovascular diagnostic technology.
• Technical Context: The technology at issue involves using artificial intelligence and machine learning to analyze coronary computed tomography angiography (CCTA) scans to create 3D models of a patient's heart, assess plaque, and determine the functional significance of arterial blockages.
• Key Procedural History: The complaint alleges that Defendant Cleerly was founded by Dr. James K. Min, a former consultant to Plaintiff Heartflow who was bound by confidentiality and invention assignment obligations. Plaintiff alleges Dr. Min used its confidential information to develop a competing enterprise. Plaintiff also notes that it received FDA de novo clearance for its foundational product in 2014, establishing it as a novel technology. The complaint further alleges that Cleerly was put on notice of its infringement through pre-suit correspondence and that Cleerly's own patents repeatedly cite Heartflow's patents-in-suit.

Case Timeline

Date	Event
2010-06-16	Dr. Min signs Non-Disclosure Agreement with Heartflow's predecessor Compl. ¶14
2012-09-12	Earliest Priority Date for '569 Patent and '425 Patent Compl. ¶142 Compl. ¶279
2012-12-05	Dr. Min signs Consulting Agreement with invention assignment clause Compl. ¶14
2013-12-18	Earliest Priority Date for '303 Patent Compl. ¶25
2014-11-26	FDA grants de novo clearance for Heartflow's FFRct Analysis platform Compl. ¶5
2016-07-19	Dr. Min incorporates Cleerly, Inc. Compl. ¶42
2017-03-28	Issue Date for U.S. Patent No. 9,607,386 Compl. ¶248
2017-05-09	Earliest Priority Date for '813 Patent Compl. ¶106 "'813 Patent", p. 1
2017-05-24	Dr. Min notifies Heartflow of his termination of the Consulting Agreement Compl. ¶45
2017-09-26	Issue Date for U.S. Patent No. 9,770,303 Compl. ¶176
2017-12-12	Issue Date for U.S. Patent No. 9,839,399 Compl. ¶213
2019-01-01	Cleerly allegedly seeks FDA approval (approximate date) Compl. ¶48
2020-06-26	Cleerly's counsel responds to Heartflow's letter, denying development of competing products Compl. ¶58
2021-05-25	Issue Date for U.S. Patent No. 11,013,425 Compl. ¶279
2022-03-29	Issue Date for U.S. Patent No. 11,288,813 Compl. ¶106
2022-07-12	Issue Date for U.S. Patent No. 11,382,569 Compl. ¶142
2026-04-13	Complaint Filing Date Compl. p. 1

II. Technology and Patent(s)-in-Suit Analysis

U.S. Patent No. 11,288,813 - "Systems and methods for anatomic structure segmentation in image analysis"

• Patent Identification: U.S. Patent No. 11,288,813, issued March 29, 2022 Compl. ¶106

The Invention Explained

• Problem Addressed: The patent describes that prior art methods for automatically segmenting anatomical structures from medical images, particularly using convolutional neural networks (CNNs), were often limited in accuracy to the level of a single pixel or voxel Compl. ¶111 '813 Patent, col. 1:29-34 This "quantization error" could reduce the precision of the resulting anatomical model, and these methods could not easily incorporate known structural assumptions (e.g., that a blood vessel is a single connected component without holes), leading to "spurious components and holes in the segmented objects" Compl. ¶111 '813 Patent, col. 1:48-53
• The Patented Solution: The invention solves this problem with a method that uses a trained CNN to achieve sub-pixel or sub-voxel accuracy Compl. ¶112 After determining a centerline of the anatomical structure (e.g., a blood vessel), the method extracts a series of 2D cross-sectional frames along that centerline Compl. ¶113 The trained CNN then performs a regression of pixel or voxel intensity values along multiple radial directions outward from the center point of each frame to predict the precise locations of the structure's boundary (Compl. ¶108, Compl. ¶113; '813 Patent, Compl. ¶abstract; '813 Patent, Compl. ¶¶col. 3:35-47).
• Technical Importance: This approach allows for the creation of more geometrically accurate 3D models of structures like coronary arteries, which is a critical prerequisite for reliable downstream analysis, such as calculating blood flow or assessing plaque burden.

Key Claims at a Glance

• The complaint asserts at least Claims 1, 3, 4, 6-8, 10-13, 15, and 17-19 Compl. ¶115 Independent Claim 1 includes the following essential elements:
  • receiving first image data of an anatomic structure of a patient;
  • obtaining an estimate of a boundary of the anatomic structure;
  • determining a centerline of the anatomical structure;
  • extracting a plurality of frames from the image data, where each frame is a plane orthogonal to the centerline;
  • determining, in each frame, a center point of the anatomic structure; and
  • generating, using a trained CNN, predictions of a plurality of locations of the boundary by performing a regression of intensity values of pixels/voxels along a plurality of radial angular directions out from each center point.
• The complaint reserves the right to assert additional claims Compl. ¶116

U.S. Patent No. 11,382,569 - "Systems and methods for estimating blood flow characteristics from vessel geometry and physiology"

• Patent Identification: U.S. Patent No. 11,382,569, issued July 12, 2022 Compl. ¶142

The Invention Explained

• Problem Addressed: The patent's background explains that while functional assessment of blood flow (e.g., Fractional Flow Reserve or FFR) is critical for treatment planning, conventional methods required invasive catheterization, which poses risks to the patient and high costs "'569 Patent", col. 1:25-36 Non-invasive computational fluid dynamics (CFD) simulations were an alternative but demanded "a substantial computational burden" "'569 Patent", col. 1:56-2:2
• The Patented Solution: The patent discloses using a machine learning algorithm to rapidly estimate blood flow characteristics non-invasively Compl. ¶146 The method involves training the algorithm on a large dataset comprising, for numerous individuals, their specific anatomical models and known blood flow characteristic values (e.g., from invasive FFR or CFD) (Compl. ¶146; Compl. ¶147, Compl. ¶abstract). The trained model learns the associations between anatomical features (e.g., vessel cross-section, diseased length) and the resulting blood flow. This trained model can then be executed on a new patient's anatomical data to quickly predict blood flow characteristics without performing a full, computationally expensive CFD simulation Compl. ¶148
• Technical Importance: This machine learning-based approach provides a computationally inexpensive and non-invasive method for functional assessment of coronary artery disease, aiming to combine the diagnostic power of invasive FFR with the safety of non-invasive imaging.

Key Claims at a Glance

• The complaint asserts at least Claims 1, 2, 4-9, 11-13, and 15-17 Compl. ¶150 Independent Claim 1 includes the following essential elements:
  • acquiring, for a plurality of individuals, an individual-specific geometric model and values of a blood flow characteristic;
  • training a machine learning algorithm using this data to generate learned associations between anatomical features (e.g., vascular cross-section area, diseased length) and the blood flow characteristic;
  • acquiring, for a new patient, images of their vascular system;
  • generating a geometric model for the new patient from the images; and
  • executing the trained machine learning algorithm to non-invasively determine values of the blood flow characteristic for the new patient using the learned associations.
• The complaint reserves the right to assert additional claims Compl. ¶151

U.S. Patent No. 9,770,303 - "Systems and methods for predicting coronary plaque vulnerability from patient-specific anatomic image data"

• Patent Identification: U.S. Patent No. 9,770,303, issued September 26, 2017 Compl. ¶176
• Technology Synopsis: The patent addresses the need for improved methods of predicting coronary plaque vulnerability from patient-specific image data Compl. ¶179 The disclosed solution involves a method that combines multiple types of analysis (image characteristics, geometrical, CFD, structural mechanics) on a patient's anatomical model to determine plaque vulnerability, and then integrates this with numerical descriptions of cardiac risk factors derived from a larger population of individuals to predict a patient-specific cardiac risk Compl. ¶180 Compl. ¶181
• Asserted Claims: At least Claims 1, 3-5, 7, 9, 11-13, 15, 17, 19, and 20 Compl. ¶183
• Accused Features: The complaint alleges that Cleerly's Plaque Analysis product infringes by performing various analyses (image characteristics, geometrical) to determine plaque vulnerability and predict cardiac risk (Compl. ¶188; Compl. ¶189; Compl. ¶190, Compl. ¶196).

U.S. Patent No. 9,839,399 - "Systems and Methods for Numerically Evaluating Vasculature"

• Patent Identification: U.S. Patent No. 9,839,399, issued December 12, 2017 Compl. ¶213
• Technology Synopsis: The patent addresses the need for a non-invasive, unified pipeline to automatically score the complexity and extent of disease in a patient's coronary vasculature Compl. ¶216 Compl. ¶217 The solution involves creating a 3D model from patient imaging, analyzing multiple predefined arterial characteristics, generating numerical measurements and assigning point values for them, and executing an algorithm to provide a cardiovascular score Compl. ¶215
• Asserted Claims: At least Claims 2-6, 8-14, and 17-18 Compl. ¶220
• Accused Features: The complaint alleges that Cleerly's Plaque Analysis and ISCHEMIA products infringe by creating a 3D model, analyzing multiple characteristics (e.g., plaque volume, composition, stenosis), and generating numerical scores presented to the user Compl. ¶220 Compl. ¶224 Compl. ¶225

U.S. Patent No. 9,607,386 - "Systems and methods for correction of artificial deformation in anatomic modeling"

• Patent Identification: U.S. Patent No. 9,607,386, issued March 28, 2017 Compl. ¶248
• Technology Synopsis: The patent addresses the problem of artificial deformations in anatomical models derived from medical images (e.g., from motion artifacts or myocardial bridging), which can compromise medical assessments Compl. ¶250 The invention provides a method to correct these artifacts by identifying the deformed portion, estimating what the local area of a non-deformed anatomy would be, and modifying the model accordingly Compl. ¶251
• Asserted Claims: At least Claims 2-4, 7, 8, 10-12, 15, 16, and 18-20 Compl. ¶253
• Accused Features: The complaint alleges Cleerly's platform infringes by identifying and accounting for artificial deformations, such as excluding stented segments from analysis or using interpolation to estimate a non-deformed vessel radius at a site of stenosis (Compl. ¶258; Compl. ¶259, Compl. ¶262; Compl. ¶263; Compl. ¶264).

U.S. Patent No. 11,013,425 - "Systems and methods for analyzing and processing digital images to estimate vessel characteristics"

• Patent Identification: U.S. Patent No. 11,013,425, issued May 25, 2021 Compl. ¶279
• Technology Synopsis: This patent is related to the '569 Patent and is also directed to using machine learning to predict physiologic values like FFR from image-derived arterial geometry Compl. ¶281 It describes workflows that create feature vectors from anatomical and physiological parameters, train machine-learning models on this data, and then compute physiologic index values for new patients Compl. ¶281
• Asserted Claims: At least Claims 5-10, 15-17, and 20 Compl. ¶283
• Accused Features: The complaint alleges Cleerly's Plaque Analysis and ISCHEMIA products infringe by receiving CCTA data, extracting patient-specific geometry, deriving quantitative features (plaque volume, stenosis, etc.), and using a trained machine-learning model to compute a per-vessel ischemia index Compl. ¶287 Compl. ¶288 Compl. ¶292

III. The Accused Instrumentality

• Product Identification: The accused instrumentalities are Defendant Cleerly's AI-powered cardiac imaging analysis software platform and its associated products, including "Cleerly Plaque Analysis," "Cleerly ISCHEMIA," and "Cleerly COMPARE" (Compl. ¶68; Compl. ¶69; Compl. ¶70).
• Functionality and Market Context: The complaint alleges that Cleerly's platform is a "comprehensive, AI-driven platform that quantifies and assesses coronary artery disease (CAD)" Compl. ¶115 It functions by receiving coronary CT angiography (CCTA) image data of a patient's coronary arteries (Compl. ¶119). Its AI and machine learning algorithms then generate a 3D model of the arteries, segment the lumen and vessel walls, and perform a range of quantitative analyses (Compl. ¶¶120-121, Compl. ¶150). These analyses include measuring plaque volume and composition, evaluating stenosis severity, and determining the likelihood of ischemia Compl. ¶115 The platform is described as a direct competitor to Heartflow's products Compl. ¶70 Compl. ¶310 A screenshot of Cleerly's user interface displays new "Lesion areas" with details on segment, stenosis, and exclusion markers Compl. p. 41

IV. Analysis of Infringement Allegations

'813 Patent Infringement Allegations

Claim Element (from Independent Claim 1)	Alleged Infringing Functionality	Complaint Citation	Patent Citation
receiving first image data of an anatomic structure of a patient;	Cleerly's platform receives coronary computed tomography angiography (CCTA) image data of a patient's coronary arteries.	¶119	col. 3:37-38
obtaining an estimate of a boundary of the anatomic structure;	Cleerly's algorithm performs automated "vessel wall and lumen segmentation," which involves obtaining an estimate of the coronary vessel lumen and wall boundaries.	¶120	col. 3:38-40
determining a centerline of the anatomical structure;	Cleerly's algorithm performs centerline extraction as a core processing step, tracking and connecting seed points to form a coronary tree.	¶121	col. 3:40-41
extracting a plurality of frames from the received first image data, each successive frame in the plurality of frames defining a respective plane orthogonal to the centerline...	Cleerly's segmentation process generates cross-sectional representations ("straightened multiplanar reformation") along the extracted centerline, which constitutes extracting frames orthogonal to the centerline.	¶¶122-123	col. 3:41-47
determining, in each frame, a center point of the anatomic structure based on the intersection of the centerline and the respective plane;	The intersection of the centerline with each extracted orthogonal plane inherently defines a center point, which is necessarily used as a reference for determining lumen and vessel wall contours.	¶124	col. 3:47-52
generating, using a trained convolutional neural network (CNN), predictions of a plurality of locations of the boundary...wherein the trained CNN is configured to perform a regression of intensity values of the pixels or voxels...along each of a plurality of radial angular directions out from each center point...	Cleerly uses trained CNNs (e.g., "VGG-19 network, 3D U-Net") to perform "lumen wall evaluation and vessel contour determination." This process is alleged to be functionally equivalent to performing a regression of intensity values radiating outward from the center point. A provided figure shows automated plaque contouring by Cleerly's deep learning system on a cross-section of an artery (Compl. p. 45).	¶125; ¶127	col. 3:52-67

• Identified Points of Contention:
  • Scope Questions: The central infringement question may turn on the construction of "perform a regression of intensity values." The complaint alleges Cleerly's CNN-based determination of vessel contours is "functionally equivalent" to this claimed step Compl. ¶127 This suggests a potential dispute over whether Cleerly's method, which is described in an exhibit as predicting contours via "attraction points radially" Compl. Ex. 49, p. 49, meets the specific "regression" limitation as defined in the patent.
  • Technical Questions: A key technical question will be what evidence demonstrates that Cleerly's CNNs are in fact "configured to perform a regression" as claimed, rather than a different machine learning task like classification or direct coordinate prediction. The distinction between these technical approaches could be a focal point of the infringement analysis.

'569 Patent Infringement Allegations

Claim Element (from Independent Claim 1)	Alleged Infringing Functionality	Complaint Citation	Patent Citation
acquiring, by a processor, for each of a plurality of individuals, an individual-specific geometric model... and values of a blood flow characteristic...	Cleerly's technology is described as being "based on over 10 million images from over 40,000 patients" and was developed using data from the CREDENCE and PACIFIC trial cohorts, which included geometric models and invasive FFR (a blood flow characteristic) measurements.	¶154; ¶155; ¶157	col. 1:63-67
training, by the processor, a machine learning algorithm using [the acquired data]... [to] generate learned associations by relating features identified from... anatomic data and the blood flow characteristic...	Cleerly's ISCHEMIA model was trained to predict invasive FFR by relating "quantitative atherosclerotic and vascular morphology features to invasive FFR values" from the trial cohorts, which constitutes generating learned associations.	¶156; ¶157	col. 2:1-12
...the individual-specific anatomical data includes a vascular cross section area, a diseased length, and one or more boundary conditions...	The quantitative variables used by Cleerly's algorithm include stenosis diameter percentages, plaque volumes, and vessel dimensions, which correspond to the claimed anatomic data features like cross-sectional area and diseased length.	¶156	col. 2:12-16
acquiring, by the processor, for a patient different from the plurality of individuals, one or more images of patient-specific anatomic data...	Cleerly's platform acquires and analyzes a new patient's CCTA images, who is necessarily different from the individuals in the training cohorts.	¶158	col. 2:17-21
generating, by the processor, a geometric model of an image region comprising the lesion of interest...	Cleerly's AI generates a 3D model of the patient's coronary arteries from the CCTA images, identifying lumen, vessel walls, and stenoses.	¶159	col. 2:22-26
executing, by the processor, the trained machine learning algorithm for at least a point of the geometric model... to non-invasively determine values of the blood flow characteristic... using the learned associations...	Cleerly's ISCHEMIA product uses the trained machine learning model to determine the likelihood of ischemia (a value of a blood flow characteristic) from the CCTA data alone. A visual in the complaint shows the "Clinical workflow" where Cleerly ISCHEMIA is used for this purpose (Compl. p. 47).	¶160; ¶161	col. 2:27-34

• Identified Points of Contention:
  • Scope Questions: The defendant may argue that the term "machine learning algorithm" is indefinite or that their specific implementation (a "random forest machine-learned algorithm" (Compl. ¶156)) falls outside the scope of the claims, especially if the patent's specification is argued to disclose a narrower set of algorithms.
  • Technical Questions: A key evidentiary question for the plaintiff will be demonstrating that Cleerly's training process and algorithm execution map directly onto the specific sequence of steps recited in Claim 1. While the complaint makes strong narrative allegations, the case may require detailed discovery into the proprietary workings of Cleerly's training and prediction models.

V. Key Claim Terms for Construction

For U.S. Patent No. 11,288,813:

• The Term: "a regression of intensity values"
• Context and Importance: This term is central to the patent's asserted technological improvement over prior art CNNs, which were allegedly limited to classification tasks. The infringement case against Cleerly hinges on whether its accused method performs this specific mathematical operation, making its construction a critical point of dispute.
• Intrinsic Evidence for Interpretation:
  • Evidence for a Broader Interpretation: The claims and specification do not appear to limit "regression" to a specific mathematical formula, which may support an interpretation covering any machine learning process that outputs a continuous value (like a distance) based on input intensity values, rather than a discrete class label. The term itself is a standard technical term.
  • Evidence for a Narrower Interpretation: The patent describes a specific problem of overcoming pixel-level "quantization error" (Compl. ¶111; '813 Patent, col. 1:47). A defendant could argue that "regression" should be narrowly construed to mean only methods that directly address this by outputting a continuous, sub-pixel coordinate or distance, as opposed to other types of continuous outputs.

For U.S. Patent No. 11,382,569:

• The Term: "a machine learning algorithm"
• Context and Importance: This term's scope is fundamental to the infringement analysis. The complaint alleges Cleerly uses a "random forest machine-learned algorithm" (Compl. ¶156), which is one of many types of machine learning. The defendant may argue that the patent does not enable the full scope of "machine learning algorithms" or that the specification's examples implicitly limit the term's scope.
• Intrinsic Evidence for Interpretation:
  • Evidence for a Broader Interpretation: The term is used broadly in the abstract and claims, suggesting it is meant to encompass the general field of machine learning as understood by a person of ordinary skill in the art at the time of the invention ("'569 Patent", abstract).
  • Evidence for a Narrower Interpretation: The specification of a related patent provides examples such as "support vector machines (SVMs), multi-layer perceptrons (MLPs), and multivariate regression (MVR)" "'425 Patent", col. 6:55-57 Cleerly may argue that its "random forest" algorithm is technically distinct from these examples and that the patent's disclosure does not support a scope broad enough to cover it, potentially raising questions of enablement or written description for the full scope of the term.

VI. Other Allegations

• Indirect Infringement: The complaint alleges both induced and contributory infringement for all asserted patents. Inducement is alleged based on Cleerly's actions of instructing and encouraging its customers (e.g., hospitals and physicians) to use the accused products in an infringing manner through tutorials, training, and marketing materials (Compl. ¶¶134, 168). Contributory infringement is alleged on the basis that Cleerly provides a material component of the patented inventions (its software platform and algorithms) that is not a staple article of commerce and is especially made for use in an infringing manner (Compl. ¶¶135, 169).
• Willful Infringement: Willfulness is a central theme of the complaint. The allegations are based on both pre-suit and ongoing knowledge of the asserted patents. The complaint alleges Cleerly had pre-suit knowledge through its founder, Dr. Min, who was a former Heartflow consultant with access to its technology Compl. ¶1 Compl. ¶2; through its hiring of a former Heartflow executive Compl. ¶61 Compl. ¶62; through express notice from Heartflow's counsel in 2020 Compl. ¶57; and through repeated citations to Heartflow's patents on the face of Cleerly's own issued patents and patent applications Compl. ¶67

VII. Analyst's Conclusion: Key Questions for the Case

1. A core issue will be one of technical and definitional scope: For the '813 patent, can the claim term "regression of intensity values" be construed to read on Cleerly's method of predicting boundary locations using a CNN? This will likely involve a deep dive into the specific mathematical operations performed by Cleerly's proprietary algorithms versus the teachings of the patent.

2. A second key issue will be the strength of the willfulness claim: The complaint presents extensive factual allegations regarding Cleerly's alleged pre-suit knowledge through its founder's history with Heartflow, its patent citations, and direct correspondence. The court will need to weigh this evidence to determine whether Cleerly's alleged infringement, if found, was willful, which carries the potential for enhanced damages.

3. A third question will be one of proof of infringement for the machine learning patents (e.g., '569, '425): While the complaint draws strong parallels between the claimed methods and Cleerly's described business, proving that Cleerly's proprietary training and execution processes meet every step of the claims will likely require significant discovery into its "black box" algorithms. The dispute may focus on subtle differences in how the respective machine learning models are trained and executed.