Research and planning

Internal working document

This page is for internal planning only. It will not be included in the final response to BSI.

What BSI is asking

The test evidence for model 'Biofilm/Slough Surface' (C515_T355) appears to be for wound bed, not biofilm/slough surface. The response.json in the evidence folder was empty ([]).

Root cause analysis — CONFIRMED DATA ENTRY ERROR

Comparing test cases T354 and T355 in ai-models-integration-tests.csv:

Field	T354 (Wound Bed Surface)	T355 (Biofilm/Slough Surface)
Title	"Verify integrated model 'Wound Bed Surface'"	"Verify integrated model 'Biofilm/Slough Surface'"
Case ID	C514	C515
Evidence URI	s3://.../wound-bed-segmenter/case-001/evidence/	s3://.../wound-bed-segmenter/case-001/evidence/
Precondition image	wound-bed-segmenter/case-001/image.jpg	wound-biofilm-segmenter/case-001/image.jpg

The Evidence URI for T355 incorrectly points to wound-bed-segmenter instead of wound-biofilm-segmenter. The precondition correctly references the biofilm segmenter model, but the evidence storage path is wrong — it's identical to T354's path.

This means:

1. The test was likely executed correctly (the precondition image was from the biofilm segmenter)
2. But the evidence was stored at the wrong S3 path (wound-bed-segmenter instead of wound-biofilm-segmenter)
3. When BSI navigated to C515_T355/case-001/response.json, they found evidence from the wound-bed model (T354), not the biofilm/slough model

The correct evidence path should be: s3://legit-health-plus/integration-verification/wound-biofilm-segmenter/case-001/evidence/

Additional confirmations that the correct model exists:

• models.json (line 696-699): Defines "Biofilm and Slough" as a tissue type in the wound-surface-segmentation model
• R-TF-030-002-Software-Bills-Of-Materials.mdx (line 448): Confirms wound-biofilm-segmenter exists as a separate Docker container
• R-TF-029-001-Deployment-and-Configuration-Commissioning-Record.mdx (line 105): Confirms wound-biofilm-segmenter v1.0.0 is deployed in production (image ID: 3f64adebeadf)

Expected output format for all wound surface segmenters (T354-T363) is consistent: {image, mask, probability_map, percentage}. The difference between models is in the actual values (reflecting different tissue types from the same wound image).

Relevant QMS documents

Document	Path	Relevance
Integration tests CSV	ai-models-integration-tests.csv line 194 (T355)	Contains the Evidence URI error
models.json	models.json lines 696-699	Defines "Biofilm and Slough" as tissue type
SBOM	R-TF-030-002-Software-Bills-Of-Materials.mdx line 448	Confirms wound-biofilm-segmenter exists
Commissioning record	R-TF-029-001 line 105	Confirms wound-biofilm-segmenter v1.0.0 deployed

Gap analysis

• Already had: The test was executed and passed. The correct model (wound-biofilm-segmenter) exists and is deployed.
• BSI couldn't find: Correct evidence — the Evidence URI in the CSV points to the wrong S3 path.
• Needs updating: (a) Fix the Evidence URI in the CSV for T355 from wound-bed-segmenter to wound-biofilm-segmenter; (b) re-export the correct evidence from S3; (c) provide the corrected evidence to BSI.

Response strategy

Regulatory mapping for this response:

Requirement	How our response addresses it
Annex II 6.2(f)	V&V evidence for the biofilm/slough surface model was incorrectly linked (evidence packaging error), not missing. Corrected evidence is provided.
EN 62304 §5.5	Integration testing was executed correctly; the error was in evidence traceability, not in the verification itself

Action required:

1. Acknowledge the data entry error in the CSV: the Evidence URI for T355 incorrectly points to wound-bed-segmenter instead of wound-biofilm-segmenter. Frame this as an evidence packaging error per Annex II 6.2(f), not a V&V execution failure.

2. Explain that the test was executed correctly — the precondition references the correct model (wound-biofilm-segmenter/case-001/image.jpg), and the expected results show valid biofilm/slough segmentation output (percentage: 13.47%).

3. Verify the correct evidence exists at the right S3 path before committing to this response. If the evidence was stored at the wrong path, the correct path (s3://legit-health-plus/integration-verification/wound-biofilm-segmenter/case-001/evidence/) might be empty. This must be confirmed before writing the final response.

4. Provide corrected evidence: export the actual evidence from the correct S3 path and include it in the supplementary evidence PDF.

5. Fix the CSV and provide the red-lined version showing the corrected Evidence URI.

6. Describe corrective action to prevent recurrence. For a Major NC, BSI will expect systemic action beyond a point fix. Proposed corrective actions:

   • Add a validation step to the integration test evidence pipeline that verifies each evidence URI path contains the model name from the test case title (e.g., T355 title contains "Biofilm/Slough Surface" → evidence URI must contain wound-biofilm-segmenter)
   • Perform an audit of all 76 integration test evidence URIs against their test case definitions (see cross-cutting CAPA section in the M1 overview)

Response tone: "We identified a data entry error in the integration test specification: the evidence URI for test case T355 (Biofilm/Slough Surface, C515) incorrectly referenced the wound-bed-segmenter evidence path instead of wound-biofilm-segmenter. Per Annex II 6.2(f), the verification was executed correctly (as confirmed by the precondition image path and expected results), but the evidence traceability was broken by the incorrect URI. We have corrected the evidence URI and provide the actual biofilm/slough surface test evidence in the supplementary evidence PDF. Corrective action: (1) the integration test CSV has been corrected [red-lined version provided]; (2) a validation check has been added to the evidence collection pipeline to prevent similar URI mismatches; (3) a full audit of all 76 integration test evidence URIs has been performed (see supplementary evidence)."

Action items:

#	Action	Owner	Document affected	Priority
6	Verify correct biofilm evidence exists at correct S3 path	Gerardo	—	High
7	Fix Evidence URI for T355 in CSV	Gerardo	ai-models-integration-tests.csv	High
8	Export correct biofilm/slough evidence from S3	Gerardo	Supplementary evidence PDF	High
9	Add evidence URI validation to integration test pipeline	Gerardo	Integration test tooling	High