What happened
MELA Sciences of Irvington, New York, makes MelaFind, a handheld imager that takes a ten-band image of a pigmented skin lesion. A classifier algorithm, developed and tested on 9439 images of 9078 lesions, labels the lesion positive or negative. In a blinded study at seven sites in 2007 and 2008, dermatologists did not see results for lesions that they had chosen to biopsy. FDA reviewers report that MelaFind detected 112 of 114 melanomas, a sensitivity of 98.3%. Its specificity was 10.6%, against 5.5% for the study dermatologists. In a web-based reader study, MelaFind's sensitivity was 97%, while dermatologists who reviewed the images missed 28% of melanomas. Their specificity was 51%, against 9% for MelaFind. An advisory panel split 8 to 7 on whether benefits outweigh risks. FDA approved MelaFind on 1 November 2011 for trained dermatologists and required a post-approval study. About 9% of lesions could not be evaluated, and 18.7% of those were melanomas or high-grade lesions.
Sources
-
Regulatory document, by authors with no tie to the system.
U.S. Food and Drug Administration, Center for Devices and Radiological Health. Summary of Safety and Effectiveness Data (SSED), P090012: MelaFind. Approved 2011-11-01.
MelaFind sensitivity was 97%, which was statistically significantly superior to that of 110 dermatologists who, on the average, missed (i.e., elected not to biopsy) 28% of melanomas in this electronic study (p-value < 0.0001).
Details
AI system
- Workflow
-
Operators cleaned the lesion with alcohol and held the imager steady for 2-3 seconds. Five earlier clinical studies developed the classifier before the blinded study tested it.
Follow-up
- Conditions for replication
-
The lesion is clinically atypical and has no clinical diagnosis of melanoma. The user is a dermatologist who has completed training in the device. Lesions that the device cannot evaluate are considered for biopsy.
Entry details
- Sector
-
Healthcare.
- Purpose
-
MELA Sciences built MelaFind to give dermatologists more information about atypical pigmented lesions when they decide whether to biopsy to rule out melanoma.
Not yet checked: model, release, provider, role, configuration, inputs and outputs, autonomy, human oversight, AI contribution, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.
Cite this record
Peter H. Vartanian. (2026). MelaFind caught nearly all melanomas at very low specificity (AIBD-0795). Artificial Intelligence Benefits Database, edition of 31 October 2026. https://petervartanian.xyz/aibd/records/AIBD-0795.html (archived at https://doi.org/10.7910/DVN/AADSII, version 4.0)