Dark Skin Phototype Diagnosis Failures

Generated: 2026-04-09

This document records all cases where the device's diagnostic output changed significantly when clinical images were converted from their original Fitzpatrick phototype to FP-V or FP-VI. These cases need visual QC to determine whether the diagnosis change reflects a genuine device limitation on dark skin or a conversion artefact (lesion erased, unrealistic rendering, etc.).

Summary

• Total cases analyzed: 143 (cases with both original and FP-V API results)
• Unaffected (same Top-1 on both phototypes): 87 (61%)
• Affected (Top-1 changed on at least one phototype): 56 (39%)
  • Category A --- Accuracy lost: 39 cases (original was correct, at least one phototype is wrong)
  • Category B --- Diagnosis shifted, no accuracy change: 10 cases (original was already wrong, or converted is still correct)
  • Category C --- Text ground truth: 7 cases (GT is a text label the device never matches)

Accuracy overview

Metric	All cases	Excl. text GT
Original Top-1 accuracy	113/143 (79.0%)	113/135 (83.7%)
FP-V Top-1 accuracy	96/143 (67.1%)	96/135 (71.1%)
FP-VI Top-1 accuracy	88/142 (62.0%)	88/134 (65.7%)

---

Condition-level analysis: erythema-dependent conditions suffer most

There is a strong, clinically coherent relationship between the type of condition and the degree of performance degradation. Conditions where erythema (redness) is the primary diagnostic feature are disproportionately affected, while conditions diagnosed by pigmentation patterns or structural morphology are largely robust.

By broad clinical category

Category	N	Original	FP-V	FP-VI	Drop V	Drop VI
Inflammatory / erythema-dependent	88	83%	73%	64%	-10pp	-19pp
Infectious	19	95%	63%	78%	-32pp	-17pp
Pigmented lesions	23	78%	78%	65%	0pp	-13pp
Structural / other	5	80%	40%	60%	-40pp	-20pp

Most affected conditions (N >= 5)

Condition (ICD-11)	N	Orig	FP-V	FP-VI	Key diagnostic feature
Psoriasis, all subtypes (EA90.*)	27	70%	48%	44%	Erythematous scaly plaques
Dermatomycosis (1F28)	8	88%	50%	100%	Erythematous annular border
Herpes / impetigo (1B72)	7	100%	71%	33%	Erythematous base with vesicles
Seb. keratosis (2F21.0)	6	100%	83%	50%	Waxy surface on erythematous base
Seb. dermatitis (EA81)	6	100%	83%	67%	Erythema with greasy scale
Pemphigus (EB05)	5	100%	100%	60%	Flaccid blisters on erythematous skin

Least affected conditions (N >= 3)

Condition (ICD-11)	N	Orig	FP-V	FP-VI	Key diagnostic feature
Melanocytic naevi (2F20.*)	8	75%	75%	75%	Pigmentation pattern
BCC (2C32)	3	100%	100%	100%	Pearly / translucent structure
Palmoplantar pustulosis (EA90.42)	5	100%	100%	100%	Structural (pustules on palms/soles)
Acne (ED80.*)	20	80%	80%	70%	Structural (papules, comedones)
Eczema (EA89)	13	92%	85%	85%	Mixed (erythema + texture)

Clinical explanation

On dark skin, erythema presents as violaceous (purple/dark) rather than the bright red seen on lighter phototypes. This is a well-documented phenomenon in clinical dermatology and one of the main reasons dermatological atlases have historically underrepresented dark skin.

Two effects are likely interacting and cannot be separated without real (non-converted) dark-skin images:

1. Conversion artefact: Imagen 3 changes skin colour but may not properly transform how erythema presents on dark skin. The real clinical difference is red to violaceous, not just a hue overlay. The conversion may erase erythema cues that the device relies on.
2. Genuine device limitation: The device may rely on red-channel features trained predominantly on light-skin erythema patterns, causing reduced performance on any dark-skin image regardless of how it was obtained.

Conditions diagnosed by structural features (acne papules, comedones) or pigmentation patterns (naevi, melanoma, BCC) are robust because those features survive skin tone changes.

Implications

• For visual QC: Inflammatory/erythematous cases warrant the closest inspection --- conversion artefacts on erythema are harder to spot visually than a fully erased lesion. Pigmented lesions can be included with higher confidence.
• For the CIR/CER: This condition-dependent pattern is clinically expected. It can be framed as consistent with known evidence that erythema-dependent conditions present differently across Fitzpatrick phototypes, a gap the PMCF plan addresses through real-world dark-phototype data collection.
• For the Phase 2 reader study: Subgroup analysis by condition type (inflammatory vs pigmented) should be pre-specified in the study protocol, as the incremental diagnostic benefit of the device may differ by condition category.

---

Category A --- Accuracy lost (original correct, converted wrong)

39 cases where the device correctly identified the condition on the original image but failed after skin tone conversion. Ranked by ground-truth probability drop (largest drop first). These are the highest-priority cases for visual QC --- a large probability drop with a completely absent GT may indicate the conversion erased the lesion.

A1. fn-202401/case-016 --- HIGH --- GT absent from one phototype

• Images: converted-images/fitzpatrick-5/fn-202401/case-016.jpg, converted-images/fitzpatrick-6/fn-202401/case-016.jpg
• Source study: SAN_2024
• Ground truth: EE12.1

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	EE12.1 (88.5%)	EC22.0 (72.5%)	EE12.1 (83.7%)
Top-1 name	Onychomycosis	Inherited deformities of nails	Onychomycosis
Top-1 = GT?	Correct	Wrong	Correct
GT probability	88.5%	0.0%	83.7%
GT rank	#1	Absent	#1

A2. bh-202406/case-021 --- HIGH --- GT absent from one phototype

• Images: converted-images/fitzpatrick-5/bh-202406/case-021.jpg, converted-images/fitzpatrick-6/bh-202406/case-021.jpg
• Source study: BI_2024
• Ground truth: EA81

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	EA81 (86.6%)	None (43.5%)	EA81 (82.2%)
Top-1 name	Seborrhoeic dermatitis	None	Seborrhoeic dermatitis
Top-1 = GT?	Correct	Wrong	Correct
GT probability	86.6%	0.0%	82.2%
GT rank	#1	Absent	#1

A3. bh-202406/case-008 --- HIGH --- GT absent from one phototype

• Images: converted-images/fitzpatrick-5/bh-202406/case-008.jpg, converted-images/fitzpatrick-6/bh-202406/case-008.jpg
• Source study: BI_2024
• Ground truth: ED80.4

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	ED80.4 (85.4%)	ED80.4 (86.0%)	EB61 (9.9%)
Top-1 name	Severe inflammatory acne	Severe inflammatory acne	Morphoea
Top-1 = GT?	Correct	Correct	Wrong
GT probability	85.4%	86.0%	0.0%
GT rank	#1	#1	Absent

A4. nv-202406/case-025 --- CRITICAL --- GT absent from both

• Images: converted-images/fitzpatrick-5/nv-202406/case-025.jpg, converted-images/fitzpatrick-6/nv-202406/case-025.jpg
• Source study: PH_2024
• Ground truth: EA90.4

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	EA90.4 (85.4%)	EK91.0 (6.7%)	2F20.Z (12.1%)
Top-1 name	Pustular psoriasis	Large plaque parapsoriasis	Melanocytic naevus, unspecified
Top-1 = GT?	Correct	Wrong	Wrong
GT probability	85.4%	0.0%	0.0%
GT rank	#1	Absent	Absent

A5. bh-202406/case-025 --- HIGH --- GT absent from one phototype

• Images: converted-images/fitzpatrick-5/bh-202406/case-025.jpg, converted-images/fitzpatrick-6/bh-202406/case-025.jpg
• Source study: BI_2024
• Ground truth: 1F28

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	1F28 (84.0%)	EA89 (44.7%)	1F28 (84.5%)
Top-1 name	Dermatophytosis	Generalised eczematous dermatitis of unspecified type	Dermatophytosis
Top-1 = GT?	Correct	Wrong	Correct
GT probability	84.0%	0.0%	84.5%
GT rank	#1	Absent	#1

A6. bh-202406/case-097 --- HIGH --- GT absent from one phototype

• Images: converted-images/fitzpatrick-5/bh-202406/case-097.jpg, converted-images/fitzpatrick-6/bh-202406/case-097.jpg
• Source study: BI_2024
• Ground truth: ED80.3

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	ED80.3 (83.9%)	ED80.3 (80.1%)	EA93 (13.5%)
Top-1 name	Nodular acne	Nodular acne	Pityriasis lichenoides
Top-1 = GT?	Correct	Correct	Wrong
GT probability	83.9%	80.1%	0.0%
GT rank	#1	#1	Absent

A7. bh-202406/case-050 --- HIGH --- GT absent from one phototype

• Images: converted-images/fitzpatrick-5/bh-202406/case-050.jpg, converted-images/fitzpatrick-6/bh-202406/case-050.jpg
• Source study: BI_2024
• Ground truth: 1B72

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	1B72 (82.3%)	1B72 (86.3%)	None (78.0%)
Top-1 name	Impetigo	Impetigo	None
Top-1 = GT?	Correct	Correct	Wrong
GT probability	82.3%	86.3%	0.0%
GT rank	#1	#1	Absent

A8. fn-202401/case-015 --- HIGH --- GT absent from one phototype

• Images: converted-images/fitzpatrick-5/fn-202401/case-015.jpg, converted-images/fitzpatrick-6/fn-202401/case-015.jpg
• Source study: SAN_2024
• Ground truth: ED70

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	ED70 (80.6%)	2F20.Z (43.8%)	ED70 (81.5%)
Top-1 name	Alopecia or hair loss	Melanocytic naevus, unspecified	Alopecia or hair loss
Top-1 = GT?	Correct	Wrong	Correct
GT probability	80.6%	0.0%	81.5%
GT rank	#1	Absent	#1

A9. fn-202401/case-006 --- HIGH --- GT absent from one phototype

• Images: converted-images/fitzpatrick-5/fn-202401/case-006.jpg, converted-images/fitzpatrick-6/fn-202401/case-006.jpg
• Source study: SAN_2024
• Ground truth: EB05

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	EB05 (80.5%)	EB05 (48.9%)	1F2D.0 (20.5%)
Top-1 name	Urticaria of unspecified type	Urticaria of unspecified type	Pityriasis versicolor
Top-1 = GT?	Correct	Correct	Wrong
GT probability	80.5%	48.9%	0.0%
GT rank	#1	#1	Absent

A10. bh-202406/case-091 --- HIGH --- GT absent from one phototype

• Images: converted-images/fitzpatrick-5/bh-202406/case-091.jpg, converted-images/fitzpatrick-6/bh-202406/case-091.jpg
• Source study: BI_2024
• Ground truth: 1B72

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	1B72 (78.2%)	1B72 (77.7%)	EA89 (34.3%)
Top-1 name	Impetigo	Impetigo	Generalised eczematous dermatitis of unspecified type
Top-1 = GT?	Correct	Correct	Wrong
GT probability	78.2%	77.7%	0.0%
GT rank	#1	#1	Absent

A11. bh-202406/case-045 --- HIGH --- GT absent from one phototype

• Images: converted-images/fitzpatrick-5/bh-202406/case-045.jpg, converted-images/fitzpatrick-6/bh-202406/case-045.jpg
• Source study: BI_2024
• Ground truth: 1F28

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	1F28 (77.6%)	2F20.Z (75.3%)	1F28 (68.4%)
Top-1 name	Dermatophytosis	Melanocytic naevus, unspecified	Dermatophytosis
Top-1 = GT?	Correct	Wrong	Correct
GT probability	77.6%	0.0%	68.4%
GT rank	#1	Absent	#1

A12. bh-202406/case-068 --- CRITICAL --- GT absent from both

• Images: converted-images/fitzpatrick-5/bh-202406/case-068.jpg, converted-images/fitzpatrick-6/bh-202406/case-068.jpg
• Source study: BI_2024
• Ground truth: EA90.0

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	EA90.0 (77.6%)	None (53.6%)	None (75.6%)
Top-1 name	Plaque psoriasis	None	None
Top-1 = GT?	Correct	Wrong	Wrong
GT probability	77.6%	0.0%	0.0%
GT rank	#1	Absent	Absent

A13. bh-202406/case-018 --- CRITICAL --- GT absent from both

• Images: converted-images/fitzpatrick-5/bh-202406/case-018.jpg, converted-images/fitzpatrick-6/bh-202406/case-018.jpg
• Source study: BI_2024
• Ground truth: ED92.0

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	ED92.0 (76.4%)	None (7.2%)	EK50.0 (2.1%)
Top-1 name	Hidradenitis suppurativa	None	Cutaneous insect bite reactions
Top-1 = GT?	Correct	Wrong	Wrong
GT probability	76.4%	0.0%	0.0%
GT rank	#1	Absent	Absent

A14. nv-202406/case-026 --- HIGH --- GT absent from one phototype

• Images: converted-images/fitzpatrick-5/nv-202406/case-026.jpg, converted-images/fitzpatrick-6/nv-202406/case-026.jpg
• Source study: PH_2024
• Ground truth: ED92.0

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	ED92.0 (76.4%)	2F20.Z (27.5%)	ED92.0 (76.3%)
Top-1 name	Hidradenitis suppurativa	Melanocytic naevus, unspecified	Hidradenitis suppurativa
Top-1 = GT?	Correct	Wrong	Correct
GT probability	76.4%	0.0%	76.3%
GT rank	#1	Absent	#1

A15. bh-202406/case-028 --- HIGH --- GT absent from one phototype

• Images: converted-images/fitzpatrick-5/bh-202406/case-028.jpg, converted-images/fitzpatrick-6/bh-202406/case-028.jpg
• Source study: BI_2024
• Ground truth: EA90.0

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	EA90.0 (75.9%)	2C30 (37.7%)	EA90.0 (76.7%)
Top-1 name	Plaque psoriasis	Melanoma of skin	Plaque psoriasis
Top-1 = GT?	Correct	Wrong	Correct
GT probability	75.9%	0.0%	76.7%
GT rank	#1	Absent	#1

A16. fn-202401/case-002 --- HIGH --- GT absent from one phototype

• Images: converted-images/fitzpatrick-5/fn-202401/case-002.jpg, converted-images/fitzpatrick-6/fn-202401/case-002.jpg
• Source study: SAN_2024
• Ground truth: 2F20.Z

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	2F20.Z (75.8%)	2F20.Z (75.4%)	ED63.0 (14.0%)
Top-1 name	Melanocytic naevus, unspecified	Melanocytic naevus, unspecified	Vitiligo
Top-1 = GT?	Correct	Correct	Wrong
GT probability	75.8%	75.4%	0.0%
GT rank	#1	#1	Absent

A17. bh-202406/case-077 --- HIGH --- GT absent from one phototype

• Images: converted-images/fitzpatrick-5/bh-202406/case-077.jpg, converted-images/fitzpatrick-6/bh-202406/case-077.jpg
• Source study: BI_2024
• Ground truth: ED92.0

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	ED92.0 (75.4%)	ED92.0 (49.1%)	2F20.Z (11.4%)
Top-1 name	Hidradenitis suppurativa	Hidradenitis suppurativa	Melanocytic naevus, unspecified
Top-1 = GT?	Correct	Correct	Wrong
GT probability	75.4%	49.1%	0.0%
GT rank	#1	#1	Absent

A18. bh-202406/case-079 --- HIGH --- GT absent from one phototype

• Images: converted-images/fitzpatrick-5/bh-202406/case-079.jpg, converted-images/fitzpatrick-6/bh-202406/case-079.jpg
• Source study: BI_2024
• Ground truth: 1B72

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	1B72 (75.3%)	EB42 (6.1%)	1B72 (79.7%)
Top-1 name	Impetigo	Linear IgA bullous dermatosis	Impetigo
Top-1 = GT?	Correct	Wrong	Correct
GT probability	75.3%	0.0%	79.7%
GT rank	#1	Absent	#1

A19. bh-202406/case-003 --- HIGH --- GT absent from one phototype

• Images: converted-images/fitzpatrick-5/bh-202406/case-003.jpg, converted-images/fitzpatrick-6/bh-202406/case-003.jpg
• Source study: BI_2024
• Ground truth: EA90.0

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	EA90.0 (74.1%)	EA90.0 (18.8%)	EB5Z (23.0%)
Top-1 name	Plaque psoriasis	Plaque psoriasis	Cutaneous lupus erythematosus of unspecified type
Top-1 = GT?	Correct	Correct	Wrong
GT probability	74.1%	18.8%	0.0%
GT rank	#1	#1	Absent

A20. bh-202406/case-062 --- HIGH --- large probability drop

• Images: converted-images/fitzpatrick-5/bh-202406/case-062.jpg, converted-images/fitzpatrick-6/bh-202406/case-062.jpg
• Source study: BI_2024
• Ground truth: EA90.40

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	EA90.40 (79.1%)	EA90 (23.9%)	EA90.40 (78.5%)
Top-1 name	Generalised pustular psoriasis	Psoriasis	Generalised pustular psoriasis
Top-1 = GT?	Correct	Wrong	Correct
GT probability	79.1%	5.3%	78.5%
GT rank	#1	#2	#1

A21. bh-202406/case-014 --- HIGH --- GT absent from one phototype

• Images: converted-images/fitzpatrick-5/bh-202406/case-014.jpg, converted-images/fitzpatrick-6/bh-202406/case-014.jpg
• Source study: BI_2024
• Ground truth: 1B72

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	1B72 (73.1%)	1B72 (74.8%)	EE7Y (4.0%)
Top-1 name	Impetigo	Impetigo	Atrophoderma of Pasini and Pierini
Top-1 = GT?	Correct	Correct	Wrong
GT probability	73.1%	74.8%	0.0%
GT rank	#1	#1	Absent

A22. bh-202406/case-039 --- HIGH --- large probability drop

• Images: converted-images/fitzpatrick-5/bh-202406/case-039.jpg, converted-images/fitzpatrick-6/bh-202406/case-039.jpg
• Source study: BI_2024
• Ground truth: ED92.0

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	ED92.0 (80.6%)	ED92.0 (69.3%)	1G04 (20.8%)
Top-1 name	Hidradenitis suppurativa	Hidradenitis suppurativa	Scabies
Top-1 = GT?	Correct	Correct	Wrong
GT probability	80.6%	69.3%	8.8%
GT rank	#1	#1	#2

A23. bh-202406/case-006 --- HIGH --- large probability drop

• Images: converted-images/fitzpatrick-5/bh-202406/case-006.jpg, converted-images/fitzpatrick-6/bh-202406/case-006.jpg
• Source study: BI_2024
• Ground truth: 2F21.0

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	2F21.0 (71.6%)	2F21.0 (71.6%)	2F20.Z (56.3%)
Top-1 name	Seborrhoeic keratosis	Seborrhoeic keratosis	Melanocytic naevus, unspecified
Top-1 = GT?	Correct	Correct	Wrong
GT probability	71.6%	71.6%	0.3%
GT rank	#1	#1	#10

A24. bh-202406/case-073 --- HIGH --- large probability drop

• Images: converted-images/fitzpatrick-5/bh-202406/case-073.jpg, converted-images/fitzpatrick-6/bh-202406/case-073.jpg
• Source study: BI_2024
• Ground truth: 1F28

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	1F28 (81.2%)	EA90 (21.0%)	1F28 (83.1%)
Top-1 name	Dermatophytosis	Psoriasis	Dermatophytosis
Top-1 = GT?	Correct	Wrong	Correct
GT probability	81.2%	10.9%	83.1%
GT rank	#1	#2	#1

A25. nv-202406/case-013 --- HIGH --- GT absent from one phototype

• Images: converted-images/fitzpatrick-5/nv-202406/case-013.jpg, converted-images/fitzpatrick-6/nv-202406/case-013.jpg
• Source study: PH_2024
• Ground truth: EB05

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	EB05 (70.0%)	EB05 (67.3%)	None (82.4%)
Top-1 name	Urticaria of unspecified type	Urticaria of unspecified type	None
Top-1 = GT?	Correct	Correct	Wrong
GT probability	70.0%	67.3%	0.0%
GT rank	#1	#1	Absent

A26. fn-202401/case-019 --- HIGH --- large probability drop

• Images: converted-images/fitzpatrick-5/fn-202401/case-019.jpg, converted-images/fitzpatrick-6/fn-202401/case-019.jpg
• Source study: SAN_2024
• Ground truth: 2F20.Z

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	2F20.Z (73.4%)	EK70 (4.2%)	2F20.Z (74.2%)
Top-1 name	Melanocytic naevus, unspecified	Cutaneous cysts	Melanocytic naevus, unspecified
Top-1 = GT?	Correct	Wrong	Correct
GT probability	73.4%	3.6%	74.2%
GT rank	#1	#4	#1

A27. fn-202401/case-005 --- HIGH --- large probability drop

• Images: converted-images/fitzpatrick-5/fn-202401/case-005.jpg, converted-images/fitzpatrick-6/fn-202401/case-005.jpg
• Source study: SAN_2024
• Ground truth: ED70

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	ED70 (70.8%)	EA81 (54.7%)	EA81 (53.7%)
Top-1 name	Alopecia or hair loss	Seborrhoeic dermatitis	Seborrhoeic dermatitis
Top-1 = GT?	Correct	Wrong	Wrong
GT probability	70.8%	9.8%	1.7%
GT rank	#1	#2	#2

A28. bh-202406/case-026 --- HIGH --- GT absent from one phototype

• Images: converted-images/fitzpatrick-5/bh-202406/case-026.jpg, converted-images/fitzpatrick-6/bh-202406/case-026.jpg
• Source study: BI_2024
• Ground truth: EA81

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	EA81 (67.4%)	EA81 (28.7%)	EA89 (8.9%)
Top-1 name	Seborrhoeic dermatitis	Seborrhoeic dermatitis	Generalised eczematous dermatitis of unspecified type
Top-1 = GT?	Correct	Correct	Wrong
GT probability	67.4%	28.7%	0.0%
GT rank	#1	#1	Absent

A29. bh-202406/case-083 --- HIGH --- large probability drop

• Images: converted-images/fitzpatrick-5/bh-202406/case-083.jpg, converted-images/fitzpatrick-6/bh-202406/case-083.jpg
• Source study: BI_2024
• Ground truth: EA89

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	EA89 (73.3%)	EA89 (73.9%)	1B70 (43.8%)
Top-1 name	Generalised eczematous dermatitis of unspecified type	Generalised eczematous dermatitis of unspecified type	Bacterial cellulitis, erysipelas or lymphangitis
Top-1 = GT?	Correct	Correct	Wrong
GT probability	73.3%	73.9%	6.3%
GT rank	#1	#1	#2

A30. bh-202406/case-071 --- HIGH --- GT absent from one phototype

• Images: converted-images/fitzpatrick-5/bh-202406/case-071.jpg, converted-images/fitzpatrick-6/bh-202406/case-071.jpg
• Source study: BI_2024
• Ground truth: 2F21.0

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	2F21.0 (62.7%)	2F21.0 (62.8%)	2F20.Z (14.4%)
Top-1 name	Seborrhoeic keratosis	Seborrhoeic keratosis	Melanocytic naevus, unspecified
Top-1 = GT?	Correct	Correct	Wrong
GT probability	62.7%	62.8%	0.0%
GT rank	#1	#1	Absent

A31. bh-202406/case-019 --- HIGH --- large probability drop

• Images: converted-images/fitzpatrick-5/bh-202406/case-019.jpg, converted-images/fitzpatrick-6/bh-202406/case-019.jpg
• Source study: BI_2024
• Ground truth: EA90.40

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	EA90.40 (73.0%)	EA90.40 (70.0%)	EA90 (23.7%)
Top-1 name	Generalised pustular psoriasis	Generalised pustular psoriasis	Psoriasis
Top-1 = GT?	Correct	Correct	Wrong
GT probability	73.0%	70.0%	12.7%
GT rank	#1	#1	#2

A32. nv-202406/case-005 --- HIGH --- large probability drop

• Images: converted-images/fitzpatrick-5/nv-202406/case-005.jpg, converted-images/fitzpatrick-6/nv-202406/case-005.jpg
• Source study: PH_2024
• Ground truth: EA90.0

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	EA90.0 (75.1%)	EA90.0 (61.8%)	EA90 (33.2%)
Top-1 name	Plaque psoriasis	Plaque psoriasis	Psoriasis
Top-1 = GT?	Correct	Correct	Wrong
GT probability	75.1%	61.8%	21.7%
GT rank	#1	#1	#2

A33. bh-202406/case-032 --- HIGH --- GT absent from one phototype

• Images: converted-images/fitzpatrick-5/bh-202406/case-032.jpg, converted-images/fitzpatrick-6/bh-202406/case-032.jpg
• Source study: BI_2024
• Ground truth: 1B72

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	1B72 (52.9%)	1F28 (65.5%)	None (76.6%)
Top-1 name	Impetigo	Dermatophytosis	None
Top-1 = GT?	Correct	Wrong	Wrong
GT probability	52.9%	1.3%	0.0%
GT rank	#1	#5	Absent

A34. fn-202401/case-020 --- MODERATE

• Images: converted-images/fitzpatrick-5/fn-202401/case-020.jpg, converted-images/fitzpatrick-6/fn-202401/case-020.jpg
• Source study: SAN_2024
• Ground truth: EA90

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	EA90 (60.0%)	EA90 (80.1%)	EA89 (14.3%)
Top-1 name	Psoriasis	Psoriasis	Generalised eczematous dermatitis of unspecified type
Top-1 = GT?	Correct	Correct	Wrong
GT probability	60.0%	80.1%	10.5%
GT rank	#1	#1	#2

A35. bh-202406/case-001 --- HIGH --- GT absent from one phototype

• Images: converted-images/fitzpatrick-5/bh-202406/case-001.jpg, converted-images/fitzpatrick-6/bh-202406/case-001.jpg
• Source study: BI_2024
• Ground truth: EA90.40

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	EA90.40 (39.4%)	1B74 (12.7%)	EA90.4 (43.1%)
Top-1 name	Generalised pustular psoriasis	Superficial bacterial folliculitis	Pustular psoriasis
Top-1 = GT?	Correct	Wrong	Wrong
GT probability	39.4%	0.0%	11.6%
GT rank	#1	Absent	#2

A36. fn-202401/case-013 --- MODERATE

• Images: converted-images/fitzpatrick-5/fn-202401/case-013.jpg, converted-images/fitzpatrick-6/fn-202401/case-013.jpg
• Source study: SAN_2024
• Ground truth: EA90

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	EA90 (50.4%)	EA90.50 (34.2%)	EA90 (29.3%)
Top-1 name	Psoriasis	Scalp psoriasis	Psoriasis
Top-1 = GT?	Correct	Wrong	Correct
GT probability	50.4%	12.9%	29.3%
GT rank	#1	#2	#1

A37. bh-202406/case-024 --- CRITICAL --- GT absent from both

• Images: converted-images/fitzpatrick-5/bh-202406/case-024.jpg, converted-images/fitzpatrick-6/bh-202406/case-024.jpg
• Source study: BI_2024
• Ground truth: 2F21.0

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	2F21.0 (34.8%)	ED52 (4.8%)	None (55.3%)
Top-1 name	Seborrhoeic keratosis	Porokeratoses	None
Top-1 = GT?	Correct	Wrong	Wrong
GT probability	34.8%	0.0%	0.0%
GT rank	#1	Absent	Absent

A38. bh-202406/case-063 --- MODERATE

• Images: converted-images/fitzpatrick-5/bh-202406/case-063.jpg, converted-images/fitzpatrick-6/bh-202406/case-063.jpg
• Source study: BI_2024
• Ground truth: EA81

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	EA81 (32.4%)	EA81 (46.7%)	AA12 (16.4%)
Top-1 name	Seborrhoeic dermatitis	Seborrhoeic dermatitis	Chondrodermatitis nodularis
Top-1 = GT?	Correct	Correct	Wrong
GT probability	32.4%	46.7%	5.5%
GT rank	#1	#1	#2

A39. bh-202406/case-065 --- MILD

• Images: converted-images/fitzpatrick-5/bh-202406/case-065.jpg, converted-images/fitzpatrick-6/bh-202406/case-065.jpg
• Source study: BI_2024
• Ground truth: EA89

	Original	Fitzpatrick V	Fitzpatrick VI
Top-1 code	EA89 (19.5%)	EA86.0 (28.9%)	EA89 (32.5%)
Top-1 name	Generalised eczematous dermatitis of unspecified type	Stasis dermatitis of lower legs	Generalised eczematous dermatitis of unspecified type
Top-1 = GT?	Correct	Wrong	Correct
GT probability	19.5%	9.5%	32.5%
GT rank	#1	#3	#1

---

Category B --- Diagnosis shifted, no accuracy change

10 cases where the Top-1 diagnosis changed but the original was already wrong (or the converted output is still correct). These are less concerning for accuracy analysis but should still be reviewed during QC for conversion quality.

#	Study	Case	GT	Orig Top-1	FP-V Top-1	FP-VI Top-1	FP-V = GT?	FP-VI = GT?
B1	BI_2024	bh-202406/case-031	1F28	EA10 (24%)	EA90 (17%)	1F28 (17%)	Wrong	Correct
B2	SAN_2024	fn-202401/case-024	2C30	2F20.Z (55%)	EJ20.1 (27%)	2F20.Z (41%)	Wrong	Wrong
B3	BI_2024	bh-202406/case-054	EA89	4A41.0 (13%)	EB5Z (20%)	4A41.0 (23%)	Wrong	Wrong
B4	BI_2024	bh-202406/case-081	EB40	EB44 (4%)	EB40 (9%)	None (81%)	Correct	Wrong
B5	BI_2024	bh-202406/case-085	EA90.0	ED50.Z (37%)	ED50.Z (34%)	EA90 (27%)	Wrong	Wrong
B6	BI_2024	bh-202406/case-030	EB40	ED80.4 (23%)	1E91 (16%)	1E91 (19%)	Wrong	Wrong
B7	PH_2024	nv-202406/case-029	EA90.0	EB5Z (17%)	EA90 (42%)	EA94 (12%)	Wrong	Wrong
B8	SAN_2024	fn-202401/case-027	EH90.Z	EB21 (9%)	1B75.3 (10%)	1B75.3 (10%)	Wrong	Wrong
B9	PH_2024	nv-202406/case-016	2C30	2C30.3 (42%)	2C30 (71%)	2C30.3 (40%)	Correct	Wrong
B10	PH_2024	nv-202406/case-027	2F20.Z	2F20.1 (78%)	2F20.Z (34%)	2F20.Z (73%)	Correct	Correct

---

Category C --- Text ground truth (device never matches)

7 cases where the ground truth is a text label (not an ICD-11 code). The device outputs ICD-11 codes, so it can never match these labels. These cases are excluded from accuracy calculations. The diagnosis changes observed here reflect the device producing different (but always non-matching) outputs on dark skin.

#	Study	Case	GT	Orig Top-1	FP-V Top-1	FP-VI Top-1
C1	BI_2024	bh-202406/case-002	Subcorneal pustular dermatosis	EB2Y (67%)	1B72 (9%)	EB41 (48%)
C2	BI_2024	bh-202406/case-004	Acute generalized exanthematous pustulosis	EA94 (26%)	EA94 (38%)	None (22%)
C3	BI_2024	bh-202406/case-017	Subcorneal pustular dermatosis	EA90.40 (41%)	EA90 (35%)	EB5Z (19%)
C4	BI_2024	bh-202406/case-040	Subcorneal pustular dermatosis	EB40 (15%)	EA90.40 (10%)	EA89 (27%)
C5	BI_2024	bh-202406/case-044	Acute generalized exanthematous pustulosis	EA89 (50%)	EA89 (61%)	EB05 (23%)
C6	BI_2024	bh-202406/case-057	Acute generalized exanthematous pustulosis	EA89 (6%)	EA89 (18%)	None (33%)
C7	BI_2024	bh-202406/case-080	Subcorneal pustular dermatosis	EB2Y (71%)	EA90.40 (7%)	2F20.Z (69%)

---

QC Priority List

Cases to inspect first during manual visual QC, ordered by likelihood of conversion artefact.

Priority 1: GT absent from both phototype outputs

These cases had a correct diagnosis on the original but the ground-truth condition completely vanished from both FP-V and FP-VI outputs. Most likely the conversion erased or fundamentally altered the lesion.

• nv-202406/case-025 --- GT EA90.4 was at 85% on original
• bh-202406/case-068 --- GT EA90.0 was at 78% on original
• bh-202406/case-018 --- GT ED92.0 was at 76% on original
• bh-202406/case-024 --- GT 2F21.0 was at 35% on original

Priority 2: GT absent from one phototype output

The condition survived one conversion but not the other. Could be a borderline conversion quality issue or a genuine device sensitivity difference between FP-V and FP-VI.

• fn-202401/case-016 --- GT EE12.1 absent from FP-V (orig 89%)
• bh-202406/case-021 --- GT EA81 absent from FP-V (orig 87%)
• bh-202406/case-008 --- GT ED80.4 absent from FP-VI (orig 85%)
• bh-202406/case-025 --- GT 1F28 absent from FP-V (orig 84%)
• bh-202406/case-097 --- GT ED80.3 absent from FP-VI (orig 84%)
• bh-202406/case-050 --- GT 1B72 absent from FP-VI (orig 82%)
• fn-202401/case-015 --- GT ED70 absent from FP-V (orig 81%)
• fn-202401/case-006 --- GT EB05 absent from FP-VI (orig 81%)
• bh-202406/case-091 --- GT 1B72 absent from FP-VI (orig 78%)
• bh-202406/case-045 --- GT 1F28 absent from FP-V (orig 78%)
• nv-202406/case-026 --- GT ED92.0 absent from FP-V (orig 76%)
• bh-202406/case-028 --- GT EA90.0 absent from FP-V (orig 76%)
• fn-202401/case-002 --- GT 2F20.Z absent from FP-VI (orig 76%)
• bh-202406/case-077 --- GT ED92.0 absent from FP-VI (orig 75%)
• bh-202406/case-079 --- GT 1B72 absent from FP-V (orig 75%)
• bh-202406/case-003 --- GT EA90.0 absent from FP-VI (orig 74%)
• bh-202406/case-014 --- GT 1B72 absent from FP-VI (orig 73%)
• nv-202406/case-013 --- GT EB05 absent from FP-VI (orig 70%)
• bh-202406/case-026 --- GT EA81 absent from FP-VI (orig 67%)
• bh-202406/case-071 --- GT 2F21.0 absent from FP-VI (orig 63%)
• bh-202406/case-032 --- GT 1B72 absent from FP-VI (orig 53%)
• bh-202406/case-001 --- GT EA90.40 absent from FP-V (orig 39%)

Priority 3: Large probability drop but GT still present

The device still recognizes the condition but with much lower confidence. May indicate partial lesion degradation in the conversion.

• bh-202406/case-062 --- GT EA90.40 dropped from 79% to FP-V 5% / FP-VI 78%
• bh-202406/case-039 --- GT ED92.0 dropped from 81% to FP-V 69% / FP-VI 9%
• bh-202406/case-006 --- GT 2F21.0 dropped from 72% to FP-V 72% / FP-VI 0%
• bh-202406/case-073 --- GT 1F28 dropped from 81% to FP-V 11% / FP-VI 83%
• fn-202401/case-019 --- GT 2F20.Z dropped from 73% to FP-V 4% / FP-VI 74%
• fn-202401/case-005 --- GT ED70 dropped from 71% to FP-V 10% / FP-VI 2%
• bh-202406/case-083 --- GT EA89 dropped from 73% to FP-V 74% / FP-VI 6%
• bh-202406/case-019 --- GT EA90.40 dropped from 73% to FP-V 70% / FP-VI 13%
• nv-202406/case-005 --- GT EA90.0 dropped from 75% to FP-V 62% / FP-VI 22%
• fn-202401/case-020 --- GT EA90 dropped from 60% to FP-V 80% / FP-VI 11%
• fn-202401/case-013 --- GT EA90 dropped from 50% to FP-V 13% / FP-VI 29%
• bh-202406/case-063 --- GT EA81 dropped from 32% to FP-V 47% / FP-VI 6%