FDA-RFQ-1226065_QandA_02.doc

FDA-RFQ-1226065 Q&A 02

Artificial Intelligence Pharmaceutical Imaging Services

Small Size Standard $27.5 million

1. Question:

Is there any guidance on how many classes the images need to be classified into by the AI? Would it suffice if the AI classifies the images into 2 categories: “Particles Present” vs. “No Particles” or is there more nuanced classes that the AI should be trained for?

Answer:

It is not enough just to categorize particles “Particles Present” vs. “No Particles”. We need to categorize the particles into different classes (for e.g., silicone particles, glass particles, fiber, dust etc.). Approaches shall be able to differentiate particles from the product, container closure system, debris, and bubbles. (This is explained on Page 3 in the FDA-RFQ-1226065.)

2. Question:

For CLIN1, it is mentioned that “Samples cover three different contain closure systems exposed to at least three stress conditions.”. That would imply that there are 9 distinct Closure System x Stress Condition sample types. However, the Quantity mentioned is 30. Is it safe to assume we expect 10 stress conditions for each of the 3 contain closure systems?

Answer:

It may have referred to sampling time points as well (3x3x3 samples per time point= ~27 samples per time point, planning to have at least 3 time points).

3. Question:

For CLIN2, it is mentioned that “Samples include two manufacturers assessed at least three stability time points.”. That would imply that there are 6 distinct manufacturers x stability time points sample types. However, the Quantity mentioned is 30. Is it safe to assume we expect 15 stability time points for each of the 2 manufacturers?

Answer:

Same as Q2 see above

4. Question:

Out of the 60,000 images (30 classes x 1000 images/class x 2 image types) how many are going to be labeled already? Who is responsible for providing the ground-truth for what is considered a “Particles Present” vs. “No Particles” (assuming we are only training for 2 categories).

Answer:

FDA is responsible for labeling the particles. This is anticipated to include an iterative process and a back and forth communication between vendor and FDA. Refer to Q1.

5. Question:

Can the Government please define the Government-furnished equipment (GFE) provided on this contract?

Answer:

There is no government furnished equipment. Contractor should have access to the software tools needed to perform the analysis.

6. Question:

Can the Government please clarify the point at which contractor-provided equipment connects to GFE and/or the Government network?

Answer:

There is no anticipated need to connect to the government network

7. Question:

Will the successful vendor need to provide the imaging devices for the service, or will DPQR provide images for the service?

Answer:

We will provide images to the vendor. They need to help us with the analysis of the images.

RESPONSE DATE – 15 April - Extended RFQ Due Date: All quotes shall be submitted via email to Christine Russman at Christine.Russman@fda.hhs.gov before 2:00 PM Eastern Time on 15 April 2020 .

20 | Page