PTAB

IPR2025-01422

CaptION Health Inc v. University Of British Columbia

Log In
Key Events
Petition
petition

1. Case Identification

2. Patent Overview

  • Title: ULTRASONIC IMAGE ANALYSIS
  • Brief Description: The ’029 patent is directed to a computer-implemented method and system for analyzing a set of ultrasound images. The technology involves using neural networks to extract features from the images to automatically determine both an image property (e.g., a view category) and a quality assessment value.

3. Grounds for Unpatentability

Ground 1: Anticipation by Krishnan - Claims 1-3, 9, 11, 21-22, 27, 29-30 are anticipated by Krishnan under 35 U.S.C. §102.

  • Prior Art Relied Upon: Krishnan (Application # 2005/0251013).
  • Core Argument for this Ground:
    • Prior Art Mapping: Petitioner argued that Krishnan discloses every element of the challenged method and system claims. Krishnan describes a system for analyzing medical images (including ultrasound) that performs steps identical to those in independent claim 1: receiving a set of images, extracting features using an "Automatic Feature Analysis" module, determining an image property via a "view identification module," and determining a quality assessment via a "quality assessment module." Krishnan further discloses using classifiers, which it specifies can be "neural networks" (meeting limitations of claims 9 and 27), to perform these determinations based on the extracted features. Finally, Krishnan teaches associating the resulting view and quality data with the images and displaying them, anticipating the output and system claims.
    • Key Aspects: The petition emphasized a side-by-side comparison of flowcharts from the ’029 patent and Krishnan, asserting they depict strikingly similar processes.

Ground 2: Obviousness over Krishnan and Chen - Claims 3-8 and 23-26 are obvious over Krishnan in view of Chen under 35 U.S.C. §103.

  • Prior Art Relied Upon: Krishnan (Application # 2005/0251013) and Chen (a 2015 conference paper on fetal ultrasound plane detection).
  • Core Argument for this Ground:
    • Prior Art Mapping: Petitioner contended that Krishnan provides the foundational system for analyzing a set of ultrasound images to determine a view category. Chen was argued to supply the specific neural network architecture recited in the dependent claims. Specifically, Chen teaches using a first feature extracting neural network (a convolutional neural network or CNN) to derive spatial features from each image in a sequence (claims 4-6) and inputting those features into a second feature extracting neural network (a recurrent neural network or RNN) to analyze temporal features across the sequence (claims 7-8).
    • Motivation to Combine: A POSITA would combine Krishnan's general framework with Chen's specific architecture to improve performance. Chen explicitly teaches that its method of using both CNNs and RNNs to consider "spatio-temporal feature representations" provides "superiority" for analyzing ultrasound videos compared to methods considering only spatial features. This provides an express reason to modify Krishnan's system to incorporate Chen's more advanced techniques for handling sequential image data.
    • Expectation of Success: Petitioner asserted a high expectation of success because both references operate in the same field of ultrasound image analysis using machine learning. Chen's architecture was described as a "general framework" that could be readily integrated into Krishnan's system without undue experimentation.

Ground 3: Obviousness of Training Claims over Krishnan, Chen, and Wu - Claims 12-20 are obvious over Krishnan in view of Chen and Wu under 35 U.S.C. §103.

  • Prior Art Relied Upon: Krishnan (Application # 2005/0251013), Chen (a 2015 conference paper), and Wu (a 2017 journal article on image quality assessment).

  • Core Argument for this Ground:

    • Prior Art Mapping: Petitioner argued that independent claim 12 and its dependents recite a conventional method for training neural networks, not a novel technique. Krishnan discloses a system using trainable neural network classifiers for view identification and quality assessment. Chen and Wu were presented as teaching the specific, well-known methods for training these types of networks. Chen describes training a view-identification network using labeled training samples corresponding to standard views. Wu describes training a quality-assessment network using thousands of training images labeled by quality class. Together, the references were argued to teach every step of the claimed training method.
    • Motivation to Combine: A POSITA implementing Krishnan’s system would need to train its neural network classifiers to make them functional. It would be obvious to turn to standard, published techniques for doing so. Chen and Wu provide explicit, task-specific examples of how to train networks for the very functions described in Krishnan (view identification and quality assessment, respectively) using pre-labeled data.
    • Expectation of Success: Success would be expected because the combination merely applies standard, well-known training methodologies to the system disclosed in Krishnan. Both Chen and Wu demonstrate that their training schemes are effective and generalizable.

  • Additional Grounds: Petitioner asserted an alternative obviousness challenge for claims 9-10 and 27-28 based on Krishnan in view of Aase (Application # 2019/0076127), arguing Aase explicitly teaches using separate, dedicated neural networks for view assignment and quality assessment.

4. Key Claim Construction Positions

  • Petitioner argued that the means-plus-function terms in claim 30 do not require express construction beyond what the parties agreed to in parallel district court litigation.
  • The agreed-upon corresponding structures identified in the specification are:
    • "means for receiving signals...": a processor with I/O interface.
    • "means for deriving... feature representations": a processor and memory operating a neural network.
    • "means for determining... a quality assessment value": a processor and memory operating a neural network.
    • "means for determining... an image property": a processor and memory operating a neural network.
    • "means for producing signals...": a processor and memory.

5. Relief Requested

  • Petitioner requests institution of an inter partes review and cancellation of claims 1-30 of the ’029 patent as unpatentable.