Standardized Consent, De-Identification Preferred for AI Image Use in Dermatology

The implementation of a standardized and transparent process of consent and de-identification may be necessary if image-based artificial intelligence (AI) is to be used in dermatology, according to recent findings, as most patients would prefer to consent to the use of their images regardless of their identifiability.¹

These results were the conclusion of new research into patient views of informed consent, given the wide variability in different areas of regulations for the gathering and processing of personal data and the sharing of de-identified information.

This new research into views on the necessity of informed consent was led by Trina Salvador, BS, from the dermatology service at Memorial Sloan Kettering Cancer Center in New York City.

“Despite widespread use of imaging in dermatology and portability of digital image formats that enable both rapid intentional and inadvertent image sharing, information about attitudes and preferences on consent and identifiability are limited,” Salvador and colleagues wrote. “Consequently, processes for obtaining informed consent are not standardized across clinical practices and research journals.” ²

Background and Methods

The investigators recruited their subjects in December 2022, surveying 1316 adult study participants located in the US who also spoke English. They implemented an online survey recruitment platform which was known as Prolific.

With the goal of ensuring adequate representation, the research team stratified their sample size based upon participants’ racial identity, gender identity, and ethnicity. They decided their research would mirror demographic proportions which are seen within the US population and had the study subjects fill out a 29-item cross-sectional survey.

The team’s survey inquired about several different topics, including participants' attitudes towards the sharing of medical photographs by their health providers, for clinical purposes as well as research/education. Some examples of questions asked by investigators included the following:

• If your medical provider shares medical photographs of your skin with your other care providers, would this impact your trust in the quality of care you receive?
• How comfortable would you be with medical photographs of your skin being used in the following clinical scenarios?
• How comfortable would you be with the use of medical photographs of your skin for the following research or education purposes?
• Would you want to be asked before the use of identifiable photographs of your skin for research or education?
• Would you want to be asked before the use of non-identifiable photographs of your skin for research or education?

Subjects were compensated financially. The demographic details of the respondents such as race, ethnicity, and several other socio-demographic elements, were self-reported. The investigators’ analysis of the survey results was carried out through the use of R 4.2.2 for generation of descriptive statistics.

Findings

The investigators reported on the demographic data and general attitudes of the 1316 survey respondents, with 47.6% having been identified as men, 51.0% as women and 1.4% as individuals identifying as other or nonbinary genders. They added that there was a median age of 43 years.

A substantial 79.6% of survey respondents were found to have expressed the belief that images are useful for the purposes of skin-related medical care. In their assessment of trust in care quality when images were implemented in research or education, 64.1% of the subjects reported no change in trust and 56.3% reported an increase in trust if there was a process of permission-seeking for image use.

The research team noted wide variation in comfort levels among the subjects, given that 45.3% reported being comfortable with image publication in different journals, 45.3% for presentations at conferences, 54.6% with use of images in AI model training, and 62.0% for the purposes of educational use. They also found that 45.0% stated that they had concerns about identifiability of the images and 51.6% had little or no concern in this matter.

A significant 84.9% of the participants were found to report being comfortable with non-identifiable image utilization. This contrasted with the 21.1% who expressed being comfortable with use of images which were identifiable. Additionally, their preferences were shown to lean toward being asked prior to using either image type (88.6% for non-identifiable and and 97.7% for identifiable).

“Most respondents reported that their trust in the quality of care they receive would increase if they were asked for consent to use their medical images, regardless of identifiability,” they wrote. “This response aligns with studies that conclude health information transparency plays a role in engendering trust.”

References

1. ^{Salvador T, Gu L, Hay JL, et al. Consent and Identifiability for Patient Images in Research, Education, and Image-Based Artificial Intelligence. JAMA Dermatol. Published online March 13, 2024. doi:10.1001/jamadermatol.2024.0084.}
2. ^{Nanda JK, Marchetti MA. Consent and deidentification of patient images in dermatology journals: observational study. JMIR Dermatol. 2022;5(3):e37398. doi:10.2196/37398.}