Zepp Health Corp v. Worcester Polytechnic Institute

1. Case Identification

• Case #: IPR2025-00521
• Patent #: 10,653,362
• Filed: February 3, 2025
• Petitioner(s): Zepp Health Corp
• Patent Owner(s): Worcester Polytechnic Institute
• Challenged Claims: 1-20

2. Patent Overview

• Title: System and Method for Physiological Signal Monitoring and Motion Artifact Detection
• Brief Description: The ’362 patent discloses systems and methods for detecting motion and noise artifacts in photoplethysmogram (PPG) signals. The technology involves obtaining a time-frequency spectrum from a PPG signal segment, deriving a noise quality index (NQI) from various spectral features, and applying a statistical learning method, such as a Support Vector Machine (SVM), to classify whether the segment is corrupted, thereby improving the reliability of physiological measurements like heart rate.

3. Grounds for Unpatentability

Ground 1: Obviousness over Chon and Mollerus - Claims 1, 3, 5-8, 10, 12, 14-17, and 19 are obvious over Chon in view of Mollerus.

• Prior Art Relied Upon: Chon (Application # 2012/0190947) and Mollerus (Application # 2010/0049267).
• Core Argument for this Ground:
  • Prior Art Mapping: Petitioner argued that Chon taught the foundational method for detecting motion artifacts in PPG signals, including obtaining a time-frequency spectrum, deriving a noise quality index from it, and using that index to determine if a signal segment is corrupted. Chon’s method uses statistical indicators and compares measures from the time-frequency spectrum to a threshold to assess signal quality. However, Petitioner contended that Chon did not explicitly disclose determining first, second, and third harmonic traces or calculating residual noise power by subtracting these traces. Mollerus was argued to supply these missing elements. Mollerus disclosed analyzing the first three harmonics of a cardiac signal's frequency spectrum to distinguish between true arrhythmias and noise, including calculating energy ratios that inherently involve assessing the power remaining after accounting for primary harmonics, which Petitioner equated to the claimed "residual noise power."
  • Motivation to Combine: A POSITA would combine these references to enhance the artifact detection system taught by Chon. Petitioner asserted that a POSITA would look to Mollerus’s sophisticated harmonic analysis techniques to improve upon Chon’s more general statistical approach, thereby creating a more robust and accurate method for identifying and filtering noise in physiological signals.
  • Expectation of Success: Petitioner argued there would be a reasonable expectation of success because both references addressed the same technical problem—differentiating valid physiological signals from noise artifacts—using spectral analysis of similar signal types. Integrating Mollerus's specific harmonic-based feature extraction into Chon's framework was presented as a predictable combination of known signal processing techniques.

Ground 2: Obviousness over Chon, Mollerus, and Simon - Claims 2, 4, 9, 11, 13, 18, and 20 are obvious over Chon in view of Mollerus and further in view of Simon.

• Prior Art Relied Upon: Chon (Application # 2012/0190947), Mollerus (Application # 2010/0049267), and Simon (Patent 8,918,178).
• Core Argument for this Ground:
  • Prior Art Mapping: This ground built upon the combination of Chon and Mollerus from Ground 1 to address the dependent claims that specifically require the "statistical learning method" to be a Support Vector Machine (SVM). Petitioner asserted that the base combination taught all elements except for the specific use of an SVM. Simon was introduced to remedy this, as it explicitly disclosed using an SVM as an "algorithmic approach to the problem of classification within the larger context of supervised learning" for physiological data. Simon taught training an SVM on physiological data to classify a patient's state and forecast medical events. The petition also highlighted that the ’362 patent specification itself admitted that SVMs were widely known and used for classification and regression analysis.
  • Motivation to Combine: A POSITA, seeking to implement the "statistical learning method" taught by the Chon/Mollerus combination, would be motivated to select an SVM as taught by Simon. Petitioner argued that SVMs were a well-known, powerful, and obvious choice for a classification task like distinguishing clean PPG data from corrupted data, making it a routine design choice for a skilled artisan.
  • Expectation of Success: Success would be expected because implementing an SVM is a straightforward application of a known machine learning tool to a well-defined classification problem. Petitioner contended that combining the feature extraction methods of Chon and Mollerus with the classification algorithm of Simon was a predictable integration of existing technologies to achieve an expected result.

4. Arguments Regarding Discretionary Denial

• Petitioner argued that discretionary denial under §314(a) based on Fintiv factors would be inappropriate. The petition asserted that the related district court litigation is in its early stages with minimal investment, and the scheduled trial date is likely to be delayed. Crucially, Petitioner offered a Sotera stipulation, agreeing not to pursue the same invalidity grounds in the district court if an IPR is instituted.
• Petitioner also argued against denial under §325(d), stating that the prior art references and the specific invalidity arguments presented in this petition were not the same or substantially the same as those considered by the examiner during the original prosecution of the ’362 patent.

5. Relief Requested

• Petitioner requests institution of IPR and cancellation of claims 1-20 of Patent 10,653,362 as unpatentable.