Rondevoo Tech LLC v. Oxford Instruments Inc

I. Executive Summary and Procedural Information

• Parties & Counsel:
  • Plaintiff: Rondevoo Technologies, LLC (California)
  • Defendant: Oxford Instruments, Inc. (Delaware)
  • Plaintiff’s Counsel: Stamoulis & Weinblatt LLC
• Case Identification: Rondevoo Technologies, LLC v. Oxford Instruments, Inc., 1:19-cv-01554, D. Del., 08/20/2019
• Venue Allegations: Venue is alleged to be proper in the District of Delaware because Defendant is a Delaware corporation and resides in the district.
• Core Dispute: Plaintiff alleges that Defendant’s "Imaris for Tracking" image analysis software infringes three patents related to generating special-purpose image analysis algorithms through an iterative, user-driven training process.
• Technical Context: The patents relate to machine-learning systems for image analysis, where an algorithm is "trained" using expert user feedback to recognize and quantify specific features, a process critical in fields like microscopy and medical research.
• Key Procedural History: The complaint does not mention any prior litigation, Inter Partes Review (IPR) proceedings, or licensing history related to the patents-in-suit.

Case Timeline

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

U.S. Patent No. 7,088,854 - "Method and apparatus for generating special-purpose image analysis algorithms"

(Issued August 8, 2006; the "’854 Patent")

The Invention Explained

• Problem Addressed: The patent's background section describes the difficulty of using computer systems to accurately classify and count objects, or "entities," within complex images, particularly when distinguishing characteristics are subtle and variable. It notes that manual analysis is tedious, inconsistent, and impractical for large-scale studies ('854 Patent, col. 1:36-47, col. 2:3-15).
• The Patented Solution: The invention provides a system where an "evolving algorithm" learns to identify entities based on an expert user's input and judgment. This process of user feedback refines the algorithm, which can then be saved as a "product algorithm" to automatically classify entities in new images, thereby capturing and automating the expert's knowledge ('854 Patent, Abstract; col. 6:5-35). The feedback loop is illustrated in the process flow of Figure 2 ('854 Patent, Fig. 2).
• Technical Importance: The technology aimed to standardize and automate quantitative image analysis, a significant bottleneck in scientific disciplines that rely on interpreting complex visual data like histological sections ('854 Patent, col. 3:1-7).

Key Claims at a Glance

• The complaint asserts independent claim 1 (Compl. ¶13).
• The essential elements of Claim 1 are:
  • A computer program product comprising a computer usable medium with computer readable program code.
  • The code is configured to obtain at least one image with chromatic data points.
  • The code is configured to generate an "evolving algorithm" that partitions the data points into an entity identified in accordance with a "user's judgment."
  • The code is configured to store an instance of the evolving algorithm as a "product algorithm" that enables automatic classification of entities in a second image based on the user's judgment.

U.S. Patent No. 7,254,266 - "Method and apparatus for generating special-purpose image analysis algorithms"

(Issued August 7, 2007; the "’266 Patent")

The Invention Explained

• Problem Addressed: As a divisional of the application leading to the ’854 Patent, this patent addresses the same problem: the need for a reproducible and efficient method of quantifying features in complex images that overcomes the limitations of manual expert analysis (’266 Patent, col. 1:26-47).
• The Patented Solution: The invention claims a specific method for automating expert quantification. The core of the method is a multi-step "training mode" where a user provides iterative feedback. The system presents identified entities to a user, obtains feedback on their accuracy, executes the algorithm using that feedback, and repeats the process before storing the approved "evolving algorithm" as a distributable "product algorithm" for a second user to apply (’266 Patent, Abstract; col. 4:32-55).
• Technical Importance: This patent formalizes the interactive training process itself as a method, detailing the specific steps of presentation, feedback, execution, and approval required to create a robust, user-trained analysis tool (’266 Patent, col. 3:1-7).

Key Claims at a Glance

• The complaint asserts independent claim 1 (Compl. ¶18).
• The essential elements of Claim 1 are:
  • A method for automating expert quantification of image data.
  • Obtaining a product algorithm configured to recognize an entity via a "training mode" that uses "iterative input" from a first user to an "evolving algorithm."
  • The training mode comprises the steps of:
    • presenting a first set of entities to the user for feedback;
    • obtaining the feedback;
    • executing the evolving algorithm with the feedback;
    • presenting a second set of entities for feedback;
    • obtaining approval and storing the evolving algorithm as a product algorithm.
  • Providing the product algorithm to a second user for application against a second set of image data.

U.S. Patent No. 8,687,879 - "Method and apparatus for generating special-purpose image analysis algorithms"

(Issued April 1, 2014; the "’879 Patent")

Technology Synopsis

Continuing the same patent family, the ’879 Patent claims a "non-transitory computer program product" for automating image quantification. The technology involves generating a "locked evolving algorithm" through the same type of iterative user-feedback training mode, where a user's input on the accuracy of identified entities is used to refine and then store a product algorithm for subsequent use (’879 Patent, Abstract). The "non-transitory" and "locked" language appears tailored to address patent eligibility standards prevalent at the time of prosecution.

Asserted Claims

Independent claim 1 (Compl. ¶23).

Accused Features

The complaint alleges infringement by the Accused System's process of generating an algorithm based on user manual annotation of objects, which allegedly trains the system (Compl. ¶44).

III. The Accused Instrumentality

Product Identification

The "Imaris for Tracking" system (the "Accused System") (Compl. ¶25).

Functionality and Market Context

The complaint describes the Accused System as a solution that "enables image analysis based on product algorithms" (Compl. ¶25). Its alleged function is to generate analyses for complex data such as "3D and 4D microscopic images" (Compl. ¶28, ¶33). The core accused functionality is a training process where the system "generat[es] an algorithm based on user manual annotation of objects of interest thereby training the convolutional neural network (CNN)" (Compl. ¶34). This suggests the product is marketed to the scientific and research communities for advanced image analysis.

No probative visual evidence provided in complaint.

IV. Analysis of Infringement Allegations

’854 Patent Infringement Allegations

Identified Points of Contention

• Scope Questions: A central question may be whether the term "user's judgment" can be construed to cover the act of a user providing initial training data (e.g., "manual annotation of objects") for a machine learning model, as the complaint alleges (Compl. ¶34). An alternative interpretation could require a more direct, evaluative feedback step where the user passes judgment on the algorithm's output, as depicted in the patent's figures (’854 Patent, Fig. 2, step 204).
• Technical Questions: The complaint alleges the Accused System trains a "convolutional neural network (CNN)" (Compl. ¶34). The ’854 Patent specification, however, primarily describes its "evolving algorithm" in terms of Bayesian classifiers and Fourier shape descriptors (’854 Patent, col. 16:60-65, col. 24:6-8). This raises the question of whether a CNN is technically equivalent to the "evolving algorithm" contemplated and disclosed in the patent.

’266 Patent Infringement Allegations

Identified Points of Contention

• Scope Questions: Claim 1 recites a highly specific, multi-step iterative process (present first set, get feedback, execute, present second set, get approval). The complaint's allegations group several of these distinct steps into a general description of training a CNN (Compl. ¶36-37, ¶39). A key point of contention will be whether the accused training workflow actually performs each of these claimed steps in sequence, or if it employs a different, more streamlined process that may not map to the claim's requirements.
• Technical Questions: As with the ’854 Patent, a technical dispute may arise over whether the accused CNN constitutes the "evolving algorithm" of the claim, given the different underlying technologies disclosed in the patent specification.

V. Key Claim Terms for Construction

The Term: "evolving algorithm"

(Asserted in ’854 Claim 1 and ’266 Claim 1)

Context and Importance

This term is the technological heart of the asserted claims. Its construction will likely determine whether the claims can cover the accused Convolutional Neural Network (CNN) technology, which is not explicitly mentioned in the patents. Practitioners may focus on this term because the outcome could depend on whether it is interpreted broadly to mean any algorithm that improves with user data, or narrowly to mean the specific types of algorithms disclosed in the specification.

Intrinsic Evidence for Interpretation

• Evidence for a Broader Interpretation: The specification states that the system may utilize "a set of evolving algorithms (e.g., Bayes' Theorem, a neural network, or any other image classification algorithm)" (’854 Patent, col. 6:8-10). The explicit mention of "a neural network" and the catch-all "or any other image classification algorithm" may support an interpretation broad enough to include a CNN.
• Evidence for a Narrower Interpretation: The detailed description of the invention heavily emphasizes specific implementations, such as a Bayesian classifier for assigning pixels to classes and blobs to entities (’854 Patent, col. 16:60-65, col. 18:58-62). A defendant may argue that these detailed embodiments limit the scope of "evolving algorithm" to these or closely related statistical methods.

The Term: "user's judgment"

(Asserted in ’854 Claim 1)

Context and Importance

Infringement of the ’854 Patent is premised on the allegation that a user providing manual annotations to train a CNN constitutes an exercise of "user's judgment" that identifies an entity. The case may turn on whether "judgment" is construed as the proactive creation of training data or a reactive confirmation/correction of the system's own attempts.

Intrinsic Evidence for Interpretation

• Evidence for a Broader Interpretation: The patent summary states the system is trained "by a user skilled at the identification of a particular object," suggesting the user's skill (i.e., judgment) is the primary input for training (’854 Patent, col. 3:42-45). This may support the view that any input embodying that skill, including annotation, meets the limitation.
• Evidence for a Narrower Interpretation: The detailed process flow describes presenting an already-identified entity "to the user for verification" (’854 Patent, Fig. 2, step 204). The specification also describes the user making a "judgment about the correctness of the classification" (’854 Patent, col. 10:20-22). This language suggests "judgment" is a responsive act of verifying or correcting the algorithm's output, not the initial act of creating training data from a raw image.

VI. Other Allegations

Willful Infringement

The complaint alleges that the Defendant has knowledge of the patents-in-suit "at least as of the service of the present Complaint" (Compl. ¶53). This allegation forms the basis for a claim of post-suit willful infringement, for which the Plaintiff seeks enhanced damages in its prayer for relief (Compl. p. 14, ¶f).

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

• A core issue will be one of technological scope: can the term "evolving algorithm," disclosed in patents from the early 2000s that primarily describe Bayesian statistical methods, be construed to cover the accused product's modern "convolutional neural network (CNN)" technology? This raises a fundamental question of whether the patent's language is broad enough to capture subsequent, more advanced implementations of machine learning.
• A second key issue will be one of procedural and functional equivalence: does the accused system's workflow, allegedly based on a user manually annotating training data, perform the specific, multi-step sequence of presentation, feedback, execution, and approval recited in Claim 1 of the ’266 Patent? The court will likely need to compare the precise operational steps of the accused product against the detailed method claimed in the patent.