Demystifying PCCPs: How Predetermined Change Control Plans (PCCPs) are Shaping the Future of AI-Enabled Medical Devices

Artificial intelligence (AI) is revolutionizing the medical device industry by enabling continuous improvement while ensuring safety and effectiveness. A pivotal regulatory tool driving this innovation is the Predetermined Change Control Plan (PCCP). This mechanism allows AI-powered devices to evolve without needing repeated FDA submissions for each modification. The FDA may authorize PCCPs for “planned changes that may be made to the device” as long as “the device remains safe and effective without any such change” and, for 510(k) devices, “the device would remain substantially equivalent to the predicate." In this article, we highlight several FDA-cleared medical devices with PCCPs, focusing on the noteworthy, pre-authorized changes that can be applied to new AI/ML-enabled medical devices. The below image showcases all the curated devices as of 20.10.2024.

Typical Changes Permitted Under PCCP for AI/ML enabled Medical Devices

Algorithm retraining

Retraining with new datasets to improve performance, address data drift, or incorporate new patient populations. Some examples below (non-exhaustive):

• Natural Cycles (FDA Reference: K241006): Allows the fertility algorithm to be retrained when integrating new wearable devices that provide temperature data. This expands the device’s functionality while ensuring safety and effectiveness.
• Overjet Charting Assist (FDA Reference: K241684): Includes retraining of the machine learning model with new data to improve the accuracy of dental structure detection.
• LINQ II ICM (FDA Reference: K210484): Allows retraining of the Zelda AI ECG Classification System to enhance arrhythmia detection through additional labeled data.

Threshold adjustments

Modifying classification thresholds (e.g., sensitivity/specificity balance) to enhance clinical performance. Some examples below (non-exhaustive):

• Irregular Rhythm Notification Feature (IRNF) (FDA Reference: K231173): Enables changes to the Tachogram Classification Algorithm's thresholds for detecting atrial fibrillation, enhancing accuracy.
• Caption Interpretation Automated Ejection Fraction Software (FDA Reference: DEN220063): Adjusts operating thresholds for improving sensitivity and specificity in estimating left ventricular ejection fraction (LVEF).

Model architecture changes

Adjustments to the model structure, including hyperparameter tuning or the introduction of new layers, to improve prediction accuracy. Some examples below (non-exhaustive):

• Clarius OB AI (FDA Reference: K233955): Permits changes to the algorithm's structure to improve fetal biometric measurements during ultrasound exams.
• BoneMRI (FDA Reference: K233030): Allows changes to the underlying model to improve the visualization of bone structures in MRI images by increasing bone-soft tissue contrast.

Addition of new data sources:

Expanding input types such as new wearable sensors, additional signal types, or new imaging modalities (e.g., ultrasound scanners). Some examples below (non-exhaustive):

• Overjet Charting Assist (FDA Reference: K241684): Integrates additional radiographic data to enhance dental charting accuracy.
• Tyto Insights for Crackles Detection (FDA Reference: K240555): Expands the dataset to include new FDA-cleared stethoscopes, allowing the software to detect lung crackles with increased accuracy.

Post-processing refinements:

Adjustments to how the model's outputs are processed to improve the accuracy or relevance of final clinical outputs. Some examples below (non-exhaustive):

• AiMIFY (FDA Reference: K240290): Allows post-processing modifications to improve the contrast-to-noise ratio in brain MRI images.
• Caption Guidance (FDA Reference: DEN190040): Refinements to the AI’s prescriptive guidance for acquiring echocardiographic images, allowing non-specialists to capture clearer images.

Expansion of device use cases:

Extending compatibility with additional patient demographics or medical conditions without altering the intended use. Some examples below (non-exhaustive):

• BoneMRI (FDA Reference: K233030): Expands the model’s use case to additional body regions, including the spine, sacrum, and hip bones.

Recommended Implementation Strategies:

• Verification and Validation (V&V) Testing: Rigorous verification to ensure that modified software meets specific performance criteria. Validation involves comparing results against ground truth datasets, ideally across multiple, diverse geographic sites.
• Software Locking: Algorithms are “locked” to prevent continuous learning in the field, ensuring changes are controlled and tested in development environments.
• Retrospective Studies: Use of real-world datasets to validate performance under the new modifications, comparing sensitivity, specificity, PPV, and NPV to previous models.
• Substantial Equivalence Testing: New versions are tested to ensure they meet performance thresholds equivalent to the original device’s cleared version.
• Risk Analysis: Hazards introduced by changes are identified and mitigated through detailed risk management processes.
• Internal Testing: The modified models undergo extensive internal testing to ensure that performance metrics are met before release.
• Non-inferiority Analysis: Ensures that new modifications do not degrade the device's performance, with margins aligned to FDA standards.

Evaluation Strategies to Ensure Safety and Effectiveness:

• Performance Testing on Diverse Datasets: Validation datasets include demographic diversity (age, sex, race, etc.) to ensure that modifications work effectively across different populations.
• Benefit-Risk Analysis: Each modification is subject to a benefit-risk analysis to evaluate the overall impact of the change and identify appropriate risk mitigations.
• Software Verification Tests: Ensures that changes are in line with design requirements and do not negatively impact the device's safety or effectiveness.
• Real-world Monitoring (Post-market surveillance): Devices are monitored for performance issues or adverse events after deployment, ensuring long-term safety.
• Specific Testing for Device Compatibility: Ensures that new sensors or devices added to the system meet existing safety and accuracy standards, without compromising performance.

How Changes Are Communicated to End Users:

• User Notifications and Software Updates: End users are notified of changes via in-app updates, software patch notes, or through device labeling updates. Updated manuals, user guides, and change logs are made available on company websites or through the device’s integrated software platform.
• Updated Instructions for Use (IFU): Revised user manuals and Instructions for Use are updated to reflect new performance metrics, features, and any changes to safety guidelines. Users receive summaries of the device’s updated performance, including sensitivity, specificity, and any changes to clinical interpretation.

Refer to FDA's latest draft guidance on PCCPs for more best guidance on writing PCCPs.

Methodology

To curate this list, I refined Brendan O'Leary's comprehensive collection of PCCP-authorized, FDA-cleared devices. (Reference: Brendan O'Leary Blog). My focus was specifically on AI/ML-enabled devices where PCCP modifications are integral to their functioning.

I manually downloaded the device summaries and used Google's NotebookLM to summarize key information, structuring my analysis around the following key questions:

1. Device Name and Regulatory Details: What are the essential regulatory details? (FDA reference number, trade/device name, regulation number, regulation name, regulatory class, and product codes).
2. Purpose and Clinical Need: What is the device’s primary purpose, and which clinical need does it address?
3. PCCP Inclusion and Permitted Modifications: Why was the PCCP included, and what types of modifications are permitted?
4. Implementation and Evaluation Strategies: How will these modifications be implemented, and what evaluation strategies are outlined to ensure continued safety and effectiveness?

The answers provide a structured breakdown, preserving crucial details and wording from the official public summaries while avoiding generic regulatory jargon.

Summary Process

After compiling and manually verifying the information for completeness and accuracy, I utilized ChatGPT 4o to generate a summary of noteworthy authorized changes for quick reference by readers. This overview serves as a resource to illustrate how PCCPs enable medical devices to stay ahead in an ever-evolving technological landscape while maintaining compliance with regulatory standards.

The following summaries were generated using a combination of NotebookLM and ChatGPT to provide key highlights for each of the medical devices listed above. While I have made every effort to verify the accuracy of the NotebookLM outputs, I encourage readers to refer to the original 510(k) documentation linked below for the most reliable and comprehensive information.

Overjet Charting Assist (K241684)

Device Name and Regulatory Details:

• FDA Reference Number: K241684
• Trade/Device Name: Overjet Charting Assist
• Regulation Number: 21 CFR 892.2050
• Regulation Name: Medical Image Management and Processing System
• Regulatory Class: Class II
• Product Code: QIH

Purpose and Clinical Need: Overjet Charting Assist is a Medical Image Management and Processing System (MIMPS) designed to assist dental professionals in identifying dental structures and generating dental charting data from 2D dental radiographs. The device detects natural tooth anatomy (such as enamel and pulp), tooth numbering, and restorative structures (implants, crowns, endodontic treatment, fillings). It aims to provide accurate, efficient dental charting by automating the detection of key dental features in bitewing, periapical, and panoramic radiographs.

It’s important to note that Overjet Charting Assist supports clinicians but is not a substitute for comprehensive clinical judgment.

PCCP Inclusion and Permitted Modifications: The Predetermined Change Control Plan (PCCP) was included because it is the only difference between the subject device and its predicate (K233590). The PCCP allows for modifications aimed at reducing false positives and negatives, primarily by retraining the machine learning model using real-world data to enhance the device's accuracy and clinical utility.

Implementation and Evaluation Strategies: Modifications under the PCCP will be implemented manually across all devices to ensure consistency. The PCCP outlines a modification protocol, which includes:

• Impact assessment considerations
• Data management requirements: covering data sources, collection methods, documentation, and reuse practices.
• Performance criteria: identical to those used for the pre-modified device (K233590), with additional analyses for identifying significant performance drops.

If modifications fail performance evaluation, they will not be implemented, and failures will be documented in the Software Development Life Cycle (SDLC). Users will be informed of changes through updated Instructions for Use, detailing algorithm performance and modifications.

FETOLY-HEART (K241380)

Device Name and Regulatory Details:

• FDA Reference Number: K241380
• Trade/Device Name: FETOLY-HEART
• Regulation Number: 21 CFR 892.1550
• Regulation Name: Ultrasonic Pulsed Doppler Imaging System
• Regulatory Class: Class II
• Product Codes: IYN, IYO, QIH

Purpose and Clinical Need: FETOLY-HEART is a machine-learning-based software designed to assist healthcare professionals during fetal ultrasound examinations in the second and third trimesters of pregnancy. It analyzes ultrasound images to automatically detect heart views and quality criteria, ensuring a complete fetal heart examination according to established guidelines.

The software addresses a critical clinical need for reliable, consistent assessments of fetal heart examinations, potentially improving the detection of cardiac abnormalities.

Predetermined Change Control Plan (PCCP): The PCCP was included to outline future modifications that can be implemented without requiring a new premarket notification. This allows flexibility in updating the software to improve performance and maintain alignment with clinical guidelines.

Permitted modifications under the PCCP include:

• Modifying model training hyperparameters to optimize performance.
• Retraining the model with new datasets to address data drift and enhance accuracy.
• Adding or removing heart quality criteria to stay consistent with updated international guidelines.

Implementation and Evaluation Strategies: Modifications to the FETOLY-HEART algorithm will be implemented through software updates. Each modification will undergo rigorous testing to ensure safety and effectiveness, including:

• Internal testing to compare the modified model's performance with the original.
• Evaluation of performance metrics using unseen test data.
• Ensuring any new quality criteria meet the same acceptance standards as the existing criteria.

Users will be notified of updates through software update notifications and updated labeling.

Natural Cycles (K241006)

Device Name and Regulatory Details:

• FDA Reference Number: K241006
• Trade/Device Name: Natural Cycles
• Regulation Number: 21 CFR 884.5370
• Regulation Name: Software Application for Contraception
• Regulatory Class: II
• Product Code: PYT (Device, fertility diagnostic, contraceptive, software application)

Purpose and Clinical Need: Natural Cycles is an over-the-counter, web and mobile-based software application that monitors a woman's menstrual cycle. Designed for women aged 18 and older, it helps monitor fertility for contraception or conception. The software uses a proprietary algorithm to evaluate user-entered data, including daily temperature measurements, menstruation cycle details, and optional ovulation or pregnancy test results.

By analyzing this data, Natural Cycles provides predictions of "not fertile" (green days) and "use protection" (red days), enabling users to make informed decisions about their fertility status and plan accordingly.

PCCP Inclusion and Permitted Modifications: The Predetermined Change Control Plan (PCCP) included in this submission (K241006) allows Natural Cycles to integrate with additional wearable devices for temperature measurement without requiring a new 510(k) submission. This plan supports future integration with a broader range of temperature-monitoring devices while maintaining the same algorithm.

The current submission does not modify the existing application, previously cleared under K231274, but focuses solely on expanding temperature input capabilities via the PCCP.

Implementation and Evaluation Strategies: Modifications under the PCCP will involve validating future wearables for use with Natural Cycles. Key Performance Indicators (KPIs) used to evaluate new wearables include:

• Positive Percent Agreement (PPA): Measures contraceptive effectiveness by evaluating how often a non-fertile day is correctly flagged, with a CI lower bound of ≥ 96.5%.
• Fraction of Green Days: Ensures no more than 2 additional red days per cycle compared to using an oral thermometer, maintaining usability.
• Day-to-Day Variability: Evaluates consistency of daily temperature readings, with a CI upper bound of ≤ 0.234.
• Ratio: Assesses the temperature phase separation, ensuring accurate detection of the shift between pre- and post-ovulatory phases, with a CI lower bound of ≥ 2.04.
• Detected Ovulations: Ensures the wearable accurately detects ovulation with an LH test, requiring a CI lower bound of ≥ 85.5%.
• Ovulation Resolution: Measures ovulation detection accuracy within two days of the LH-only ovulation day, with a CI lower bound of ≥ 76.1%.

Caption Guidance (DEN190040)

Device Name and Regulatory Details:

• FDA Reference Number: DEN190040
• Trade/Device Name: Caption Guidance
• Regulation Number: 21 CFR 892.2100
• Regulation Name: Radiological Acquisition and/or Optimization Guidance System
• Regulatory Class: Class II
• Product Code: QJU

Purpose and Clinical Need: Caption Guidance is a software designed to assist medical professionals in acquiring cardiac ultrasound images. It serves as an accessory to compatible diagnostic ultrasound systems, addressing the clinical need for improved access to echocardiography. Given the shortage of skilled cardiac sonographers and the extensive training required for echocardiography, this tool empowers non-cardiac professionals (such as nurses) to capture standard echocardiography images. These images can then be reviewed by a qualified cardiac healthcare professional.

PCCP Inclusion and Permitted Modifications: The Predetermined Change Control Plan (PCCP) was included due to the deep learning algorithms powering Caption Guidance. The PCCP allows for future algorithm improvements, focusing on refining the device’s ability to provide Prescriptive Guidance for maneuvering the ultrasound probe to the optimal position.

This plan mitigates the risk of negative impacts on the device’s clinical performance following algorithm changes.

Implementation and Evaluation Strategies: Modifications to Caption Guidance’s algorithm under the PCCP will undergo non-clinical and feasibility-level clinical testing. These tests will evaluate core functionalities, particularly the algorithm’s ability to accurately predict the optimal probe position.

To ensure safety and effectiveness, the PCCP defines specific assessment metrics, acceptance criteria, and statistical methods for performance evaluation. However, detailed evaluation strategies were not fully outlined in the available information.

Caption Interpretation Automated Ejection Fraction Software (DEN220063)

Device Name and Regulatory Details:

• Trade/Device Name: Caption Interpretation Automated Ejection Fraction Software
• FDA Reference Number: DEN220063
• Regulation Number: 21 CFR 892.2055
• Regulation Name: Radiological Machine Learning-Based Quantitative Imaging Software with Predetermined Change Control Plan
• Regulatory Class: Class II
• Product Code: QVD

Purpose and Clinical Need: This software is designed to process transthoracic cardiac ultrasound images and provide an automated estimation of left ventricular ejection fraction (LVEF). LVEF is a critical parameter used to evaluate heart systolic function, aiding clinicians in managing cardiovascular diseases by assessing the severity of heart dysfunction.

PCCP Inclusion and Permitted Modifications: The Predetermined Change Control Plan (PCCP) was added to allow future modifications to the software while ensuring safety and effectiveness. The PCCP permits the following types of changes:

• Training on Additional Data to improve accuracy.
• Incorporation of additional 2D TTE views for more comprehensive evaluation.
• Optimization of core algorithm implementation.
• Improving algorithm operating speed for faster processing.

Implementation and Evaluation Strategies:

• Implementation: Machine learning algorithms are "locked" prior to release and do not continuously learn in the field. Caption Health will deploy a single version of the software within the United States. Upon a new release, Caption Health will notify all existing customers and request an upgrade to the latest version.
• Evaluation: Risk Analysis will identify new hazards from algorithm changes and track mitigation activities. Evaluation involves: Verification: Algorithm, subsystem, and end-to-end system levels. Validation: End-to-end system level with expert-labeled ground truth data.
• The PCCP specifies criteria for test datasets: minimum new data percentage, limitations on usage for the same purpose.
• Key acceptance criteria include: Accurate clip identification (mode, view, frames). Clip suitability for EF estimation. Auto EF accuracy compared to expert EF. Individual view performance. Confidence Metric equivalence to expert EF criteria.

User manuals and software interfaces will be updated to reflect any modifications in version numbering and new features.

LINQ II ICM with Zelda AI ECG Classification System (K210484)

Device Name and Regulatory Details:

• FDA Reference Number: K210484
• Trade/Device Name: LINQ II Insertable Cardiac Monitor, Zelda AI ECG Classification System
• Regulation Number: 21 CFR 870.1025
• Regulation Name: Arrhythmia Detector and Alarm (including ST-segment measurement and alarm)
• Regulatory Class: Class II
• Product Codes: MXD

Purpose and Clinical Need: The LINQ II ICM is an insertable cardiac monitor that continuously records subcutaneous electrocardiograms (ECG), designed to detect and automatically record arrhythmias. It can also be patient-activated. The device is intended for patients at an increased risk of cardiac arrhythmias or those experiencing transient symptoms such as dizziness, palpitations, syncope, or chest pain—symptoms potentially caused by arrhythmias.

PCCP Inclusion and Permitted Modifications: A Predetermined Change Control Plan (PCCP) was included to allow for ongoing improvements to the Zelda AI ECG Classification System, which uses deep-learning neural networks to detect atrial fibrillation and pauses.

The PCCP permits the following types of modifications:

• Threshold adjustments for arrhythmia detection.
• Algorithm retraining using either the original protocol or an alternate labeled data protocol.
• Algorithm pre-training to enhance model performance.

Implementation and Evaluation Strategies: Modifications under the PCCP will be implemented under controlled conditions, with rigorous testing before release to ensure superior performance. Key strategies include:

• Ensuring that sensitivity and specificity are maintained or improved without compromising other performance metrics.
• Locking the algorithms to prevent continuous learning in the field.

The PCCP defines specific assessment metrics, acceptance criteria, and statistical methods to evaluate the modified algorithms. Post-market surveillance will be conducted to monitor the device’s performance and safety after any modifications.

AliveCor Corvair (K231010)

Device Name and Regulatory Details:

• FDA Reference Number: K231010
• Trade/Device Name: Corvair
• Regulation Number: 21 CFR 870.1025
• Regulation Name: Arrhythmia Detector and Alarm (including ST-segment measurement and alarm)
• Regulatory Class: Class II
• Product Code: MHX

Purpose and Clinical Need: The Corvair ECG analysis system is a software tool designed to assist healthcare professionals in measuring and interpreting resting diagnostic electrocardiograms (ECGs). It provides an initial automated interpretation of ECGs for rhythm and morphological information, which can be confirmed, edited, or deleted by the healthcare professional. This device addresses the clinical need for accurate and efficient ECG analysis, helping professionals quickly identify potential cardiac abnormalities for prompt diagnosis and treatment.

PCCP Inclusion and Permitted Modifications: The Predetermined Change Control Plan (PCCP) was included to allow for algorithm performance improvements through retraining with additional data, without requiring a new 510(k) submission. The PCCP permits the retraining of algorithms with additional high-quality, diverse data from major clinical institutions, as long as the data is similar to the data used in the original model’s training.

Implementation and Evaluation Strategies:

• Implementation: Modifications to the algorithms will be implemented through retraining with new data, while keeping the model architecture unchanged.
• Evaluation Strategies: The performance of the retrained models will be evaluated using the same datasets used in the original 510(k) submission. Additional large validation datasets will be created from sites independent of the training data to ensure model generalization. The overall performance of the retrained models must be noninferior to the performance of the original models. Minor variations in individual determination performance are acceptable.

Communication of Changes: Once improvements are validated and accepted, Corvair's device labeling will be updated to reflect the changes. These updates will be communicated to software integrators via the Corvair API, enabling them to inform end users accordingly.

Irregular Rhythm Notification Feature (IRNF) (K231173)

Device Name and Regulatory Details:

• FDA Reference Number: K231173
• Trade/Device Name: Irregular Rhythm Notification Feature (IRNF)
• Regulation Number: 21 CFR 870.2790
• Regulation Name: Photoplethysmograph Analysis Software for Over-the-Counter Use
• Regulatory Class: Class II
• Product Code: QDB

Purpose and Clinical Need: The Irregular Rhythm Notification Feature (IRNF) is a software-only mobile medical application designed for the Apple Watch. It analyzes pulse rate data to identify irregular heart rhythms that may indicate atrial fibrillation (AFib) and notifies the user. As an over-the-counter (OTC) screening tool, IRNF is intended for early AFib detection, supplementing traditional screening decisions, especially when used in conjunction with a user’s specific risk factors. It is not intended to replace standard diagnostic or treatment methods.

PCCP Inclusion and Permitted Modifications: The Predetermined Change Control Plan (PCCP) was included to allow certain modifications to the IRNF 2.0 software without requiring a new premarket notification. This ensures the device remains as safe and effective as the predicate device while allowing controlled improvements.

The PCCP permits modifications to:

• Tachogram Classification Algorithm: Adjustments to the numerical threshold for classifying a tachogram as AFib.
• Confirmation Cycle Algorithm: Modifications to the number of sequential tachograms classified as irregular to trigger a notification and adjustments to the time period of the confirmation cycle.

Implementation and Evaluation Strategies:

• Implementation: All algorithm modifications will be trained, tuned, and locked before the software release. Adaptive algorithms that continuously learn in the field are not permitted under the PCCP.
• Evaluation Strategies: Evaluation strategies includes verification and validation, specific test methods for substantial equivalence ?relative to IRNF 2.0, including sample size determination, analysis methods, and acceptance criteria. To ensure representativeness of the intended use population, validation test datasets will adhere to minimum demographic requirements for age, sex, race, and skin tone based on United States demographics.

Instructions for Use Updates: Users will be notified about algorithm changes made under the PCCP, and updated Instructions for Use will be available on the Apple website and within the Health App, summarizing changes and performance updates.

REMI-AI Discrete Detection Module (K231779)

Device Name and Regulatory Details:

• FDA Reference Number: K231779
• Trade/Device Name: REMI AI Discrete Detection Module
• Regulation Number: 21 CFR 882.1400
• Regulation Name: Electroencephalograph
• Regulatory Class: Class II
• Product Code: OMB

Purpose and Clinical Need: The REMI-AI Discrete Detection Module (REMI-AI DDM) is a software as a medical device (SaMD) designed to assist physicians in analyzing electroencephalogram (EEG) recordings taken with the REMI Remote EEG Monitoring System. The software identifies and marks EEG sections that may represent seizures lasting 10 seconds or longer, making it easier for physicians to review recordings. This tool is intended for use in adult and pediatric patients (6+ years), specifically by physicians trained in EEG analysis. Importantly, REMI-AI DDM does not operate in real-time and does not provide diagnostic conclusions.

PCCP Inclusion and Permitted Modifications: The Predetermined Change Control Plan (PCCP) was included to allow future improvements to the algorithm, focusing on:

• Expanding the training data to improve the algorithm’s performance.
• Optimizing the algorithm's internal operations for enhanced efficiency and effectiveness.

Implementation and Evaluation Strategies: Modifications to the REMI-AI DDM algorithm will follow the guidelines in the PCCP and the standard software update procedures. Users will be informed of updates through:

• Manual updates to reflect the changes.
• Customer notifications about the update and any potential impacts.
• Release notes available on the Epitel website.

To ensure continued safety and effectiveness, the updated algorithm will undergo validation using:

• A previously established validation dataset.
• An updated validation dataset to assess the changes.

Low Ejection Fraction AI-ECG Algorithm (K232699)

Device Name and Regulatory Details:

• FDA Reference Number: K232699
• Trade/Device Name: Low Ejection Fraction AI-ECG Algorithm
• Regulation Number: 21 CFR 870.2380
• Regulation Name: Cardiovascular Machine Learning-Based Notification Software
• Regulatory Class: Class II
• Product Code: QYE

Purpose and Clinical Need: The Anumana Low Ejection Fraction AI-ECG Algorithm is designed to screen adults at risk for heart failure with a Left Ventricular Ejection Fraction (LVEF) of ≤ 40%. It is aimed at patients with conditions such as cardiomyopathies, past myocardial infarctions, aortic stenosis, chronic atrial fibrillation, cardiotoxic medication usage, and postpartum women. The tool provides a point-of-care screening option in settings like primary care, urgent care, and emergency departments where cardiac imaging might not be readily available.

PCCP Inclusion and Permitted Modifications: The Predetermined Change Control Plan (PCCP) is included in the submission but does not provide specific details on permitted modifications or the rationale behind its inclusion.

Implementation and Evaluation Strategies: The summary do not elaborate on how modifications under the PCCP will be implemented or how safety and effectiveness will be ensured following the modifications.

BoneMRI (K233030)

Device Name and Regulatory Details:

• FDA Reference Number: K233030
• Trade/Device Name: BoneMRI
• Regulation Number: 21 CFR 892.2050
• Regulation Name: Medical Image Management and Processing System
• Regulatory Class: Class II
• Product Code: QIH

Purpose and Clinical Need: BoneMRI is an image processing software designed to enhance MRI images, improving the visualization of bone structures by increasing contrast between bone and surrounding soft tissue. This software is used in imaging of the pelvic region (sacrum, hip bones, femoral heads) and the spine (cervical, thoracic, lumbar, S1 vertebrae). Its enhanced visualization helps radiologists and orthopedic surgeons assess bone morphology, tissue radiodensity, and radiodensity contrast in patients aged 12 and older.

PCCP Inclusion and Permitted Modifications: The Predetermined Change Control Plan (PCCP) was included to provide a structured framework for making algorithm improvements without requiring a new premarket notification for each change. The PCCP supports iterative development for the machine learning models used in BoneMRI, allowing for:

1. Re-training the ML model with additional data: Enhancing accuracy, especially in challenging cases such as rare pathologies or imaging artifacts, while increasing the model’s robustness and generalization.
2. Validating additional scanner support: Ensuring compatibility and performance of the ML model with additional MRI vendors or varying MRI field strengths, with or without model re-training.

Implementation and Evaluation Strategies: Modifications under the PCCP will be executed by training, tuning, and locking the algorithm before releasing the updated application. To ensure safety and effectiveness, the following evaluation strategies will be applied:

• Software verification and validation testing: These will follow FDA guidelines.
• Performance validation: This will include quantitative analysis of BoneMRI images compared to CT scans from the same patients, focusing on bone morphology, radiodensity, and radiodensity contrast. Subgroup analyses will consider factors like MRI vendor, field strength, age, location, and BMI.
• 3D bone morphology accuracy: Evaluated by measuring the mean absolute cortical delineation error, aiming for an average of below 1.0 mm.
• Tissue radiodensity: Assessed by comparing the mean deviation in Hounsfield Units (HU) between BoneMRI and CT images, with a target of below 25 HU overall and below 55 HU for bone.
• Tissue radiodensity contrast: Measured using the mean HU correlation coefficient, which should be above 0.75, specifically for bone.

CLEWICU System (K233216)

Device Name and Regulatory Details:

• FDA Reference Number: K233216
• Trade/Device Name: CLEWICU System
• Regulation Number: 21 CFR 870.2210
• Regulation Name: Adjunctive Predictive Cardiovascular Indicator
• Regulatory Class: Class II
• Product Codes: QNL

Purpose and Clinical Need: The CLEWICU System is an analytical software designed for hospital critical care settings for patients aged 18 and older. Using machine learning models, the system calculates the likelihood of critical clinical events, such as hemodynamic instability requiring vasopressor/inotrope support. It provides clinicians with insights into a patient’s predicted risk of clinical deterioration or low risk status, helping to make informed decisions in intensive care settings. The early identification of patients at risk can greatly improve patient outcomes.

PCCP Inclusion and Permitted Modifications: The Predetermined Change Control Plan (PCCP) was included in the CLEWICU submission to allow modifications to the CLEW models without needing a new 510(k) submission. The PCCP permits:

• Training and validation of the CLEW models for new input datasets, addressing variations in hospital electronic medical record (EMR) systems and patient monitoring protocols. This includes situations where hospitals may have reduced or additional input data types compared to the original models.
• Re-training models with new datasets to improve the models' sensitivity, while maintaining or enhancing performance specifications.

Implementation and Evaluation Strategies: Modifications under the PCCP will be applied using the same protocols and procedures approved by the FDA for the original CLEWICU models. Key strategies include:

• Validation using patient data from at least three geographically diverse sites, ensuring no single site contributes more than 50% of the total validation dataset.
• Performance Specifications: New models must meet the same minimum performance standards as the original models:CLEWHI: Minimum sensitivity of 0.6 and PPV of 0.1.CLEWLR: Minimum sensitivity of 0.25 and SPC of 0.9.
• New models may show improved PPV/SPC while maintaining sensitivity within previous performance ranges.
• Adjustments to the PPV target will be based on the event prevalence.

Important Note: The modifications under the PCCP will not affect the device's indications for use or its operation once deployed. The user interface and criteria for raising notifications about hemodynamic events or identifying low-risk patients will remain unchanged.

SleepStageML (K233438)

Device Name and Regulatory Details:

• FDA Reference Number: K233438
• Trade/Device Name: SleepStageML
• Regulation Number: 21 CFR 882.1400
• Regulation Name: Electroencephalograph
• Regulatory Class: Class II
• Product Code: OLZ

Purpose and Clinical Need: SleepStageML is an AI/ML-powered software designed to analyze polysomnography (PSG) recordings and automatically score sleep stages. It assists sleep physicians and technicians in assessing sleep quality in patients aged 18 and older. Accurate sleep stage scoring is vital for diagnosing and managing sleep disorders, helping clinicians make informed treatment decisions.

PCCP Inclusion and Permitted Modifications: The Predetermined Change Control Plan (PCCP) was included to enable future updates and improvements to SleepStageML without requiring a new FDA 510(k) submission for each change. The PCCP permits changes to four key components:

1. Machine Learning Model: Retraining with new datasets, adjusting hyperparameters, loss functions, and optimizers, or making architectural changes within limits.
2. Signal Preprocessing: Adjustments to the processing steps applied to EEG signals before they are input to the machine learning model.
3. Probability Postprocessing: Updates to methods for converting model outputs into sleep stages.
4. Signal Quality Check: Modifying criteria to ensure the analyzability of EEG signals before processing.

Changes to the machine learning model and signal preprocessing components would require retraining, while updates to probability postprocessing and signal quality checks would not.

Implementation and Evaluation Strategies: Modifications will follow rigorous software verification and validation processes. This mirrors the testing procedures from SleepStageML's original development and includes:

• Software Verification: All tests related to requirements and specifications must pass for a modification to be validated.
• Clinical Performance Validation: Modifications must meet specific performance criteria:Non-inferiority to the original SleepStageML device and predicate device in per-stage performance metrics across full overnight recordings.Minimal unanalyzable recordings.Non-inferiority in 2-hour recording segments and for recordings using the minimum number of AASM-recommended EEG channels.Non-inferiority in multi-stage agreement across all five sleep stages compared to the best-performing released version of SleepStageML.

User Notification: Following the release of any updated version under the PCCP, clinical users will be informed about the new version, its features, and any updated performance information.

Clarius OB AI (K233955)

Device Name and Regulatory Details:

• FDA Reference Number: K233955
• Trade/Device Name: Clarius OB AI
• Regulation Number: 21 CFR 892.1550
• Regulation Name: Ultrasonic Pulsed Doppler Imaging System
• Regulatory Class: Class II
• Product Codes: IYN, QIH

Purpose and Clinical Need: Clarius OB AI is a machine learning algorithm designed to help healthcare professionals measure fetal biometric parameters during obstetric ultrasounds. Accurate fetal measurements are essential for monitoring fetal growth and development, estimating gestational age, and identifying potential complications.

PCCP Inclusion and Permitted Modifications: The Predetermined Change Control Plan (PCCP) allows for modifications to the Clarius OB AI algorithm without the need for a new premarket notification. This ensures the software can continuously improve while maintaining safety and effectiveness. The PCCP permits modifications to:

• Model architecture to optimize performance.
• Model training methods and parameters to enhance performance and generalization.
• Post-processing algorithms to refine results and improve robustness.
• Data input sources to ensure compatibility with newer models of Clarius Ultrasound Scanners.

Implementation and Evaluation Strategies: Modifications will be implemented through retraining, testing, and locking the algorithm before release. The following steps ensure continued safety and effectiveness:

• Internal Testing: To compare the modified model's performance against the original.
• Clinical Performance Testing: Verification and validation studies to confirm accuracy and clinical utility.
• Benefit-Risk Analysis: Evaluating the potential risks and benefits of modifications.
• Risk Mitigation: Strategies like cross-validation, testing diverse datasets, and preventing overfitting.

Acorn 3D (K234009)

Device Name and Regulatory Details:

• FDA Reference Number: K234009
• Trade/Device Name: Acorn 3D Software (AC-SEG-4009); Acorn 3DP Model (AC-101-XX)
• Regulation Number: 21 CFR 892.2050
• Regulation Name: Medical Image Management And Processing System
• Regulatory Class: Class II
• Product Codes: QIH, LLZ

Purpose and Clinical Need: Acorn 3D is an image processing software enabling users to import, visualize, and segment medical images and create 3D models for diagnostic purposes in musculoskeletal and craniomaxillofacial applications. These models are vital for treatment planning and diagnostics.

PCCP Inclusion and Permitted Modifications: The PCCP allows modifications to the device without requiring a new 510(k) submission, provided they are consistent with the approved plan. Any major changes affecting safety or effectiveness (e.g., changes to the design, materials, or manufacturing process) would require a new submission.

Implementation and Evaluation Strategies: The source does not specify the exact modifications allowed under the PCCP.

AiMIFY (1.x) Overview (K240290)

Device Name and Regulatory Details:

• FDA Reference Number: K240290
• Trade/Device Name: AiMIFY (1.x)
• Regulation Number: 21 CFR 892.2050
• Regulation Name: Medical Image Management and Processing System
• Regulatory Class: Class II
• Product Code: LLZ

Purpose and Clinical Need: AiMIFY is an image processing software designed to enhance MRI images by improving contrast-to-noise ratio (CNR), contrast enhancement (CEP), and lesion-to-brain ratio (LBR). It is especially useful for visualizing enhancing tissue in brain MRI images acquired with gadolinium-based contrast agents.

PCCP Inclusion and Permitted Modifications: The PCCP outlines specific modifications allowed without the need for a new 510(k) submission. These modifications focus on:

• Improving the generalizability of the AI/ML model.
• Reducing processing time.
• Improving perceived image quality.

Permitted changes include:

• Expanding applicability to new gadolinium-based contrast agents and broader patient populations.
• Retraining the AI model to reduce processing time.
• Introducing an optional vessel suppression algorithm for better image quality.

Implementation and Evaluation Strategies: All changes will adhere to Subtle Medical's Design Change Control process, following ISO 13485:2016 standards. Evaluation strategies include:

• Documentation updates for each modification.
• Version-specific release notes and user manual updates.
• Ensuring adherence to the device’s Indications for Use, performance endpoints, and acceptance criteria.

Tyto Insights for Crackles Detection: Device Overview and PCCP (K240555)

Device Name and Regulatory Details:

• FDA Reference Number: K240555
• Trade/Device Name: Tyto Insights for Crackles Detection
• Regulation Number: 21 CFR 868.1900
• Regulation Name: Diagnostic Pulmonary-Function Interpretation Calculator
• Regulatory Class: Class II
• Product Code: PHZ

Purpose and Clinical Need: Tyto Insights for Crackles Detection is an over-the-counter AI-enabled decision support software designed to analyze lung sounds in adults and children aged 2 years and older. It works with the Tyto Stethoscope, detecting potential crackle sounds in lung recordings. The system supports healthcare providers by automatically flagging abnormal lung sounds suggestive of "crackles."

PCCP Inclusion and Permitted Modifications: The Predetermined Change Control Plan (PCCP) included in this device submission enables the system to make modifications in three key areas:

1. Performance Specifications: Retraining the machine learning model with new data to improve accuracy.
2. Technical Specifications: Enhancements to preprocessing, architecture, or hyper-parameters for computational efficiency.
3. Device Inputs: Adding compatibility with new FDA-cleared stethoscopes of the same type.

Implementation and Evaluation Strategies: For each type of modification, Tyto Insights will follow detailed software verification, validation, and clinical performance studies to ensure continued safety and effectiveness. Performance metrics such as sensitivity and specificity will be assessed using validation datasets to ensure the software meets regulatory expectations. Users will be notified of any significant changes through updated labeling and documentation.

Sleep Apnea Notification Feature (SANF): Device Overview and PCCP (K240929)

Device Name and Regulatory Details:

• FDA Reference Number: K240929
• Trade/Device Name: Sleep Apnea Notification Feature (SANF)
• Regulation Number: 21 CFR 868.2378
• Regulation Name: Over-the-counter device to assess risk of sleep apnea
• Regulatory Class: Class II
• Product Code: QZW

Purpose and Clinical Need: The SANF software analyzes data from Apple Watch to detect breathing disturbances suggestive of moderate-to-severe sleep apnea. Designed for over-the-counter use by adults, it flags potential risks for sleep apnea, guiding users to seek further professional evaluation.

PCCP Inclusion and Permitted Modifications: The PCCP allows the system to implement modifications related to:

1. Breathing Disturbances (BD) Computation: Adjusting or retraining algorithms for better performance.
2. Sleep Apnea Estimation: Modifying parameters to enhance sensitivity and specificity of notifications.

Implementation and Evaluation Strategies: All modifications will be rigorously tested using FDA-approved verification and validation methods, ensuring substantial equivalence to the initial model. Performance will be evaluated with demographic-representative datasets. Updates will be communicated through user notifications and revised Instructions for Use on Apple platforms.

Exo AI Platform 2.0: Device Overview and PCCP (K240953)

Device Name and Regulatory Details:

• FDA Reference Number: K240953
• Trade/Device Name: AI Platform 2.0 (AIP002)
• Regulation Number: 21 CFR 892.2050
• Regulation Name: Medical Image Management and Processing System
• Regulatory Class: Class II
• Product Code: QIH

Purpose and Clinical Need: Exo AI Platform 2.0 assists healthcare providers with the analysis and reporting of ultrasound images. It supports workflow optimization by offering real-time quality scores for cardiac and lung scans, helping users acquire high-quality images more efficiently.

PCCP Inclusion and Permitted Modifications: The PCCP allows for:

• Modifications to AI architecture and pre/post-processing to enhance performance and adaptability.
• Introduction of new training data to improve model robustness and reduce bias.

Implementation and Evaluation Strategies: Modifications will be implemented following a robust modification protocol, with stringent verification and validation procedures to ensure substantial equivalence. Each modification will undergo non-inferiority testing to ensure the model performs within acceptable safety and efficacy thresholds. User updates and release notes will ensure transparency and keep stakeholders informed.

Disclaimers

• Personal View: The views expressed in this post are my own and do not represent the views of my employer.
• Use of Gen-AI Tools: I leveraged generative AI tools like NotebookLM and ChatGPT for summarization purposes. While I manually verified the summaries, these tools are known to sometimes produce mistakes or hallucinations. Users should refer to the original device summaries for the most accurate information.
• Information Source: The information presented here is based on sources available in the public domain and is provided "as-is" without any warranties, express or implied.
• FDA's Original Source: For reliable, up-to-date, and comprehensive details, please refer directly to the FDA’s website or the original device summary.
• No Warranty: This information is presented without any warranty or guarantee regarding its accuracy or completeness. Users should rely on the FDA’s official publications for authoritative and final information.