Exploring the Differences Between the AI/ML Framework and the Pre-Cert Program

The AI/ML Framework and Pre-Cert Program can be seen as the next generation of FDA regulatory requirements. This article uses a holographic diagram to compare the two systems, exploring their similarities and differences, and possible ways to reconcile them. Finally, the article lays the groundwork for building a suitable organizational structure for next-generation software development.

The purpose of this article is to provide information and discussion on the differences between the AI/ML Framework and Pre-Cert Program, including:

1. Displaying the differences between the two systems to help readers gain a clearer understanding of them.
2. Understanding the FDA’s thinking, such as TPLC.
3. As it is highly likely that many companies will have both types of products in the future, exploring how to design a process control mechanism that can balance the needs of both types of software products.

The basic core technique — TPLC

Compared to traditional medical software (SaMD), FDA adopts a more flexible but also more demanding approach for these two types of medical software.

In addition to the basic principle of risk-based approach in medical device management, introducing a complete product lifecycle management will be a milestone in entering the next generation of management.

The approach to implementing a complete product lifecycle management is TPLC (Total Product Life Cycle).

The characteristic of TPLC is that it can span across premarket and postmarket monitoring of products and evaluate both the product and organizational aspects, so the management level is already quite sophisticated.

However, depending on the characteristics of different products, the components of the TPLC process may also differ.

Differences inTPLC components and process

Fig 1 shows the official version of TPLC introduced in the Pre-Cert Program, while Fig 2 is a holographic diagram of the TPLC process and components created by the author.

Fig1. TPLC of Pre-Cert Program

Fig 2 Holographic diagram of TPLC process and components of the Pre-Cert Program

Fig. 3 shows the TPLC approach of the AI/ML Framework, while Fig. 4 is a holographic diagram of the TPLC process and components compiled by the author.

Fig. 3 TPLC approach of the AI/ML Framework

Fig 4 Holographic diagram of the TPLC process and components

Similarities and Differences

The most obvious difference is that Pre-Cert may go through 3 to 4 steps in one TPLC cycle, while AI/ML only goes through 3 steps regardless of the state. Additionally, the product and organizational reviews differ due to the differences in the products.

However, similarities can be found in the requirements for (1) organizational excellence required by CQOE and (2) real-world performance analysis, which are both basic requirements. Please refer to Table 1 for a comparison table.

Table 1 Comparison between Pre-cet and AI/ML Framework

Comparing with the existing review process (Figure 5, using the Pre-Cert Program as an example), in short, traditional review only relates to the product aspect prior to market launch, while the review of AI/ML and Pre-Cert Program covers both premarket and postmarket, product and organizational aspects due to the adoption of TPLC.

Figure 5 shows the differences between the current review method and the Pre-Cert method.

Discussions

In the future, the regulatory review process is expected to become more comprehensive and cover a wider range of requirements, with increasing consistency between FDA requirements and those of MDR or IVDR.

Personally, I believe this is also an excellent opportunity for development companies to upgrade their systems.

In addition, the use of next-generation SaMD and AI/ML is almost inevitable, so a company may have both types of SaMD coexisting. Therefore, possible reconciliatory approaches are discussed below.

Starting with the basic requirements of CQOE and RWPA, a common document can be established, and a corresponding secondary tier documents can be created for the review path according to the different products. The proposed harmonization framework for the development process is shown in Figure 6.

Fig 6 suggested development flow and harmonized structure

The grey portion of Figure 6 shows the common structure shared by both the Pre-Cert Program and the AI/ML Framework, while the other parts are specifically related to each program and can be separated into their own second-tier development processes such as procedures in ISO 13485 structure.

The author recommends having a separate development process for each product, as these products can undergo frequent changes, especially with AI/ML. Additionally, it is important to manage these products with a product management approach rather than a project management approach, as the TPLC concept is focused on the product lifecycle.

The UXA, PMS, and PMCF components of RWPA can be combined with customer service information to provide feedback to the design end, which also demonstrates how to establish a framework that meets the requirements of MDR and IVDR.

Further practical discussions on operational aspects will be presented in another article.