Abstract
This cross-sectional survey finds that most dermatologists are receptive to assistance from AI tools and envision a role for augmented intelligence in clinical practice, but they also value the physician-patient relationship and human accuracy.
Artificial intelligence (AI) refers to the ability of a machine or computer program to solve problems that would be typically handled by humans. Augmented intelligence (AuI) focuses on the assistive role of AI and is designed to enhance, but not replace, human intelligence and the physician-patient relationship.1 Studies have demonstrated2,3 superior skin cancer classification using a combination of dermatologists and AI and improved prognostication of malignant neoplasms by dermatologists using AI. A recent systematic review4 called for dermatologists’ leadership in defining how these technologies fit into clinical practice.
Methods
To evaluate dermatologists’ perspectives on AI and AuI, the American Academy of Dermatology (AAD) Task Force on Augmented Intelligence conducted a cross-sectional survey of 3080 fellows from July 10 to July 30, 2020. Continuous variables were summarized with means and standard deviations. Categorical variables were reported as proportions and percentages. Descriptive statistical analyses were performed from August 3 to September 23, 2020, using Excel 14.7.1 (Microsoft Corp).
The AAD reviewed and approved the study, waiving informed consent because the study used only deidentified data. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline was followed.
Results
Characteristics of the 121 dermatologists (mean [SD] age, 51 [12] years; 64 [53%] men) who completed the survey (response rate, 3.9%) are detailed in Table 1. Most respondents self-reported their race/ethnicity as White (76; 84%) and non-Hispanic/Latino (93; 95%). Respondents represented all career stages and diverse physician practice settings.
Table 1. Characteristics of Dermatologist-Respondents.
| Characteristic | No. of Total (%) |
|---|---|
| Total No. of respondents | 121 |
| Age, mean (SD), y | 51 (12) |
| Sex | |
| Male | 64 of 121 (53) |
| Female | 57 of 121 (47) |
| Racea | |
| White | 76 of 90 (84) |
| Asian | 7 of 90 (8) |
| American Indian/Alaska Native | 2 of 90 (2) |
| Black | 1 of 90 (1) |
| Native Hawaiian/other Pacific Islander | 1 of 90 (1) |
| Other | 6 of 90 (7) |
| Ethnicity | |
| Hispanic/Latino | 5 of 98 (5) |
| Non-Hispanic/Latino | 93 of 98 (95) |
| Career stageb | |
| Early | 33 of 120 (28) |
| Mid | 43 of 120 (36) |
| Late | 44 of 120 (37) |
| Practice setting | |
| Dermatology group | 39 of 100 (39) |
| Hospital (academic) | 20 of 100 (20) |
| Solo practice | 19 of 100 (19) |
| Multispecialty group | 11 of 100 (11) |
| Veterans Affairs/military | 5 of 100 (5) |
| Hospital (nonacademic) | 3 of 100 (3) |
| Other | 3 of 100 (3) |
| Community served | |
| Suburban | 53 of 101 (52) |
| Urban | 40 of 101 (40) |
| Rural | 8 of 101 (8) |
| Practice in vulnerable population area | |
| Yes | 41 of 97 (42) |
| No | 56 of 97 (58) |
| Practice in HSPA | |
| Yes | 17 of 91 (19) |
| No | 74 of 91 (81) |
Abbreviation: HSPA, health professional shortage area.
Total percentages exceed 100% because categories are not mutually exclusive.
Early career was defined as fellows <40 years old or 45 years old with ≤8 years postresidency; mid, as fellows 40-55 years old with 8 years postresidency; and late, fellows ≥55 years old.
Dermatologists’ perspectives on AI and AuI are summarized in Table 2. While only 56 (46%) thought AI would positively influence their practice, that opinion rose to 78 (64%) for AuI. Sixty-nine (57%) stated they would use an AI tool with similar accuracy to a human dermatologist to help diagnose skin lesions in clinic. Ninety-one (76%) respondents stated that they would be more likely to perform a biopsy of a lesion that was not clinically suggestive of cancer if an AI tool indicated a possible malignant diagnosis. Also, 9 (8%) respondents stated that if an AI tool indicated a benign diagnosis, they would be less likely to perform a biopsy on a clinically suggestive lesion. Eighty-nine (74%) respondents said they would use an AI tool with similar accuracy to a human dermatologist to help monitor skin lesions.
Table 2. Dermatologists’ Perspectives on Artificial Intelligence (AI) and Augmented Intelligence (AuI).
| Survey item | No. of Total (%) |
|---|---|
| Artificial intelligencea | |
| What type of impact do you think AI will have on your dermatology practice? | |
| Positive | 56 of 121 (46) |
| None | 13 of 121 (11) |
| Negative | 22 of 121 (18) |
| Unsure | 30 of 121 (25) |
| In your opinion, what are the greatest potential benefits of AI tools for skin cancer screening? Select your top 3 answers.b | |
| More efficient triage | 78 of 119 (66) |
| Improved health care access | 56 of 119 (47) |
| Quicker diagnosis | 37 of 119 (31) |
| Less patient anxiety owing to avoiding/delaying diagnosis | 33 of 119 (28) |
| Reduced health care cost | 31 of 119 (26) |
| None of these | 25 of 119 (21) |
| Imagine an AI tool that analyzes an image of a skin lesion and suggests a likely diagnosis. For the purpose of this survey, assume that this tool has similar accuracy to a human dermatologist. Would you use this tool to help you diagnose skin lesions in your clinic? | |
| Yes | 69 of 121 (57) |
| No | 19 of 121 (16) |
| Unsure | 33 of 121 (27) |
| If you suspected that a skin lesion was benign, but the tool suggested a malignant diagnosis, would that increase your likelihood of performing a biopsy? | |
| Yes | 91 of 120 (76) |
| No | 4 of 120 (3) |
| Unsure | 25 of 120 (21) |
| What do you consider to be the greatest potential strengths of AI tools for skin cancer screening as compared to a human dermatologist? Select your top 3 answers.b | |
| Patient motivation to seek out skin cancer diagnosis and/or treatment | 45 of 115 (39) |
| More objective diagnosis | 40 of 115 (35) |
| More convenient diagnosis | 34 of 115 (30) |
| More accurate diagnosis | 33 of 115 (29) |
| More consistent diagnosis | 29 of 115 (25) |
| Patient education | 25 of 115 (22) |
| None of these | 25 of 115 (22) |
| Are you concerned that the development of AI may exacerbate health care disparities in dermatology? | |
| Yes | 33 of 117 (28) |
| No | 47 of 117 (40) |
| Unsure | 37 of 117 (32) |
| Augmented intelligencea | |
| What type of impact do you think AuI will have on your dermatology practice? | |
| Positive | 78 of 121 (64) |
| None | 12 of 121 (10) |
| Negative | 10 of 121 (8) |
| Unsure | 21 of 121 (17) |
| In your opinion, what are the greatest potential risks of AI tools for skin cancer screening? Select your top 3 answers.b | |
| Lack of patient follow-up because using AI without clinician support | 63 of 117 (54) |
| Clinician loss of control to AI | 55 of 117 (47) |
| Human deskilling | 49 of 117 (42) |
| Loss of doctor-patient personal interaction | 47 of 117 (40) |
| Greater patient anxiety from receiving diagnosis without clinician support | 42 of 117 (36) |
| Unethical use of AI | 31 of 117 (26) |
| Job losses | 14 of 117 (12) |
| Patient loss of privacy | 5 of 117 (4) |
| Reduced health care access | 4 of 117 (3) |
| None of these | 15 of 117 (13) |
| Imagine an AI tool that analyzes serial images of a skin lesion to detect change over time. For the purpose of this survey, assume that this tool has similar accuracy to a human dermatologist. Would you use this tool to help you monitor skin lesions in your clinic? | |
| Yes | 89 of 121 (74) |
| No | 15 of 121 (12) |
| Unsure | 17 of 121 (14) |
| If you suspected that a skin lesion was malignant, but the tool suggested a benign diagnosis, would that decrease your likelihood of performing a biopsy? | |
| Yes | 9 of 120 (8) |
| No | 87 of 120 (73) |
| Unsure | 24 of 120 (20) |
| What do you consider to be the greatest potential weaknesses of AI tools for skin cancer screening as compared with a human dermatologist? Select your top 3 answers.b | |
| Inability to perform TBSE | 69 of 108 (64) |
| Lack of creativity owing to algorithmic limitations | 43 of 108 (40) |
| Lack of social contract between AI and patient | 43 of 108 (40) |
| Lack of verbal communication | 34 of 108 (31) |
| Less accurate diagnosis | 32 of 108 (30) |
| Lack of emotion | 22 of 108 (20) |
| Lack of nonverbal communication | 21 of 108 (19) |
| None of these | 6 of 108 (6) |
| What would be the greatest potential challenge of implementing AI and/or AuI in dermatology? Select your top 3 answers.b | |
| Disruption of human physician-patient relationship | 57 of 105 (54) |
| Threats to accuracy | 56 of 105 (53) |
| Medical malpractice lawsuits | 49 of 105 (47) |
| Lack of credibility | 24 of 105 (23) |
| Threats to patient privacy and confidentiality | 16 of 105 (15) |
| Other | 6 of 105 (6) |
Abbreviation: TBSE, total body skin examination.
Defined for participants as follows: AI refers to a machine or computer program that can solve problems that are typically handled by humans (eg, a machine that can classify skin lesions at the level of a clinician) and AuI focuses on AI’s assistive role and is designed to enhance, rather than replace, human intelligence and the physician-patient relationship.
Total percentage exceeds 100% because categories are not mutually exclusive.
The survey instrument asked dermatologists to prioritize 3 benefits, risks, strengths, and weaknesses of AI tools for skin cancer screening identified by patients in a qualitative study.5 As benefits, dermatologists collectively prioritized more efficient triage, improved health care access, and quicker diagnosis. As risks, dermatologists collectively prioritized lack of patient follow-up owing to using AI without clinician support, clinician loss of control to AI, and human deskilling. Thirty-three (28%) dermatologists thought AI would exacerbate health care disparities. As strengths, dermatologists collectively prioritized patient motivation to seek skin cancer diagnosis or treatment, more objective diagnosis, and more convenient diagnosis. As weaknesses, dermatologists collectively prioritized the inability to perform total body skin examinations, lack of creativity owing to algorithmic limitations, and a lack of social contact between AI and the patient. Finally, most of the dermatologist-respondents identified disruption of the human physician-patient relationship and threats to accuracy as implementation challenges.
Discussion
This cross-sectional study indicates that dermatologists are receptive to assistance from AI tools in diagnosing and monitoring skin lesions but value the human physician-patient relationship and accuracy. These findings confirm and complement those of a recently published survey of Chinese dermatologists,6 which found that 99.5% are attentive to information on AI and 95.4% envision a role for AuI in dermatology. Dermatologists were more likely to biopsy a nonsuggestive lesion if the tool indicated a malignant diagnosis than to forego a biopsy of a suggestive lesion if the tool indicated a benign diagnosis. Future research would be beneficial to determine whether AuI increases biopsy rates in clinical practice.
The very low survey response rate of this study strongly limits the conclusions and generalizability of its findings. It is unknown whether respondents were users of AI or AuI, which may have introduced bias into the spectrum of responses. Although respondents’ demographic information was aligned with that of the general population of AAD fellows (eg, mean age, 50 years; 47% men), further research is needed to confirm the key finding, that is, that most dermatologists are open to integrating AI and AuI with clinical practice.
References
- 1.Kovarik C, Lee I, Ko J; Ad Hoc Task Force on Augmented Intelligence . Commentary: position statement on augmented intelligence (AuI). J Am Acad Dermatol. 2019;81(4):998-1000. doi: 10.1016/j.jaad.2019.06.032 [DOI] [PubMed] [Google Scholar]
- 2.Hekler A, Utikal JS, Enk AH, et al. Superior skin cancer classification by the combination of human and artificial intelligence. Eur J Cancer. 2019;120:114-121. doi: 10.1016/j.ejca.2019.07.019 [DOI] [PubMed] [Google Scholar]
- 3.Han SS, Park I, Eun Chang S, et al. Augmented intelligence dermatology: deep neural networks empower medical professionals in diagnosing skin cancer and predicting treatment options for 134 skin disorders. J Invest Dermatol. 2020;140(9):1753-1761. doi: 10.1016/j.jid.2020.01.019 [DOI] [PubMed] [Google Scholar]
- 4.Zakhem GA, Fakhoury JW, Motosko CC, Ho RS. Characterizing the role of dermatologists in developing artificial intelligence for assessment of skin cancer: A systematic review. J Am Acad Dermatol. 2020;S0190-9622(20)30079-7. doi: 10.1016/j.jaad.2020.01.028 [DOI] [PubMed] [Google Scholar]
- 5.Nelson CA, Pérez-Chada LM, Creadore A, et al. Patient perspectives on the use of artificial intelligence for skin cancer screening: a qualitative study. JAMA Dermatol. 2020;156(5):501-512. doi: 10.1001/jamadermatol.2019.5014 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Shen C, Li C, Xu F, et al. Web-based study on Chinese dermatologists’ attitudes towards artificial intelligence. Ann Transl Med. 2020;8(11):698. doi: 10.21037/atm.2019.12.102 [DOI] [PMC free article] [PubMed] [Google Scholar]