Skip to main content
NCBI home page
BMC Primary Care logoLink to BMC Primary Care
. 2025 Nov 28;26:384. doi: 10.1186/s12875-025-03073-9

Diagnostic accuracy of artificial intelligence compared to family physicians and dermatologists for skin conditions: a systematic review and meta-analysis

Norhane Nadour 1,✉, Théo Duguet 1,2, Sophie Zahedi 1, Hugo Figoni 1, Roxane Liard 1,✉
PMCID: PMC12661747  PMID: 41315984

Abstract

Context

Artificial intelligence (AI) technologies are increasingly used for image recognition, especially for skin lesions. Due to what may be long wait times for dermatology appointments, general practitioners (GPs) are the gatekeepers when it comes to skin diseases requiring rapid treatments.

Objective

This study aims to examine the diagnostic accuracy of AI in diagnosing skin lesions encountered in primary care and to perform a meta-analysis of AI’s in diagnostic accuracy for melanoma detection.

Methodology

This systematic review and meta-analysis, conducted according to the 2020 PRISMA guidelines, included diagnostic accuracy studies using any type of AI applied to photographs or dermoscopy images to diagnose skin lesions encountered in primary care settings. The reference standard was dermatologist consensus or histopathological examination. Searches were conducted in PubMed, Web of Science and Cochrane in December 2023. Risk of bias and concerns of applicability were assessed using the QUADAS-2 tool. Data extraction was conducted by two investigators and meta-analysis was performed using a bivariate random effects model.

Results

Between 2013 and 2023, 382 studies were found and 38 met the inclusion criteria.AI's accuracy was reported as non-inferior or superior to that of dermatologists in 30 studies, while 4 studies reported that AI was less accurate than dermatologists. Similarly, AI's accuracy was reported as non-inferior or superior to that of GPs in 8 studies, and one study indicated that AI was less accurate than GPs. The meta-analysis showed that AI for the diagnosis of melanoma had a pooled sensitivity of 0.86 (95% CI: 0.80–0.90) and a specificity of 0.94 (95% CI: 0.89–0.97). The diagnostic odds ratio was 44.36 (95% CI: 29.28; 67.1), with an AUC of 0.922 for the SROC curve. Of the 38 included studies, 25 were at high risk of bias, primarily due to patient selection. Datasets were frequently not representative of the outpatient population, as malignant conditions were often overestimated.

Conclusion

AI appears to perform at a similar level to dermatologists, and the same is true when comparing AI to GPs. This is especially true for serious conditions like melanoma, suggesting that AI could be a valuable tool for GPs in improving patient care.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12875-025-03073-9.

Keywords: Artificial intelligence, Dermatology, Diagnosis

Background

Artificial intelligence (AI) was first coined by Professor McCarthy at Stanford in 1955 [1]. It is a broad informatics field involving computer systems that imitate human intelligence to perform complex tasks like conversing, reacting to environments, or recognizing objects with precision [2]. AI systems use sensors to detect data relevant to their tasks, which can include text, sounds, images, or physical measures, with the type of AI depending on the nature of the data [3].

Recent progress in machine learning, particularly convolutional neural networks (CNNs) in deep learning, has brought AI back into focus in healthcare research [4, 5]. Neural networks process data through layers of artificial neurons, each extracting increasingly complex characteristics. Initial layers detect simple features like outlines, while deeper layers identify entire objects. During training, with the large amounts of data processed, neural networks adjust the weights that connect neurons in different layers and reduce prediction error [3, 6].

Deep learning is well-suited for visual recognition tasks, such as in radiology, ophthalmology or dermatology [7, 8]. General practitioners (GPs) are the first step in healthcare, and 30.5% of the French population suffers from dermatological diseases [9]. In France, in 2020, there were an average of 4.4 private practice dermatologists per 100,000 inhabitants [10]. In 2024, approximately the same proportion of 3.66 dermatologists for 100,000 was documented in the USA, with a non-uniform distribution between regions. The same study estimated the population's need at 4 dermatologists per 100,000 inhabitants. Consequently, dermatology is a crucial part of a GP’s medical practice, amidst a wide range of diseases. The average waiting time for a dermatologist appointment in France can reach 95 days [11], making the GP's role in referring the right patients to specialists essential.

For melanoma, which represents 17,900 new cases per year in France [12], delayed diagnosis and delayed treatment can worsen prognosis [13]. Since 2017, AI devices developed by private industries to analyze skin lesions have shown progress, notably after a study by Esteva et al. demonstrated AI’s diagnostic accuracy is comparable to dermatologists in identifying malignant pigmented skin lesions [14]. The types of AI, data studied, targeted populations, and results vary, making AI’s potential in diagnosing skin conditions still uncertain. This study aims to examine the diagnostic accuracy of AI in diagnosing skin lesions encountered in primary care and to perform a meta-analysis of diagnostic accuracy of AI for melanoma detection.

Methods

This systematic review was conducted following the 2020 PRISMA (Preferred Reporting Items for Systematic reviews and Meta-Analyses) guidelines. A protocol was established, submitted online on PROSPERO (international prospective register of systematic reviews) on December 21 st 2023. It was registered on January 1 st 2024 (available online, identification: CRD42024495355).

Inclusion criteria

The studies reporting on the accuracy of AI model diagnoses focused on skin conditions evaluated by GPs or dermatologists in outpatient settings. Datasets using photographs or dermoscopy images only were sought. The primary outcome was the sensitivity and specificity, accuracy, or Youden index of AI tools in detecting a specific skin condition. The index test was any AI model applied to clinical photographs or dermoscopy images of outpatients or from datasets. The reference standard was diagnostic histopathological interpretation or dermatologists consensus or unremarkable follow-up data for benign lesions.

Exclusion criteria

Were excluded: case-reports, studies about hospitalized patients or follow-up of lesions previously diagnosed. Studies with images obtained from specialist devices such as confocal microscopy or optical coherence tomography were also excluded.

Assuming that AI technology performance improves over time, the most recent studies were considered representative of its optimal current performance. Therefore, studies published more than 10 years ago were excluded.

Studies in languages other than English or French were excluded. Unpublished studies, preprints, or conference proceedings were excluded.

Literature research

The literature search was conducted in December 2023. Studies were collected from 3 databases: PubMed, WebOfScience, Cochrane. The research equations (Appendix A) were developed with the help of a librarian from Paris Sorbonne University. The MeSH terms were ‘dermatology’ OR ‘skin diseases’, OR ‘dermoscopy’ AND ‘artificial intelligence’, AND ‘diagnosis’.

Two reviewers screened titles separately (N.N. and S.Z.). The same two reviewers screened abstracts separately. When conflicts occurred, they were discussed at each step to reach agreement. One reviewer (N.N.) read full texts to include the studies.

Citation screening of included studies was performed. The reference lists of the selected review studies were screened. Studies that met the inclusion criteria and were not already screened were added.

Risk of bias assessment

The QUADAS—2 tool was used to evaluate the risk of bias and applicability of the studies of this review [15]. For each study, an assessment was made regarding whether potential biases were present. Elements related to these potential biases were identified by reading the full texts. If such elements were noticeable, it presumed a high risk of bias. If not, it presumed a low risk of bias [16].

Qualitative synthesis

For the qualitative synthesis, the outcome for diagnosis performance in each study was collected. These results were considered alongside study characteristics and potential biases to evaluate their possible influence.

Meta-analysis

Diagnostic accuracy of AI in detecting melanoma was evaluated using a hierarchical summary receiver operating characteristic (HSROC) approach implemented with a bivariate random-effects model (Reitsma method) through the R package mada [17]. Study-specific sensitivity and specificity estimates were pooled, and corresponding forest plots with global summary estimates (random-effects models) were generated. Additionally, an HSROC curve was plotted, including a summary estimate with a 95% confidence ellipse, and the area under the curve (AUC) was computed.

Results

Article selection

In December 2023, the research equations found 382 references. After deleting duplicates, 370 references remained.

The title and abstract screenings excluded 199 and 93 articles; 78 articles were left. After reading full texts and adding articles from reference lists, 38 articles were included in the final analysis.

The reasons for exclusion are detailed in the flowchart (Fig. 1).

Fig. 1.

Fig. 1

Open in a new tab

Flowchart

Characteristics of the studies included

The characteristics of the studies included are detailed in Table 1

Table 1.

Characteristics of the studies included

Country Type of skin lesion studied Type of image studied Data source Comparison Study design Accuracy test
Esteva, 2017, [14] USA Malignant, pigmented or not Dermoscopic and clinical Outpatient department Dermatologists Retrospective AUC,
Yu C, 2018, [18] South Korea Acral, malignant and pigmented Dermoscopic Outpatient departments Dermatologists and Family physicians Prospective Se, Spe, AUC Youden, accuracy
Marchetti, 2018, [19] USA Malignant and pigmented Dermoscopic ISIC web database Dermatologists Retrospective Se, Sp, AUC
Han, 2018, [20] South Korea Malignant, pigmented or not Clinical Outpatient department Dermatologists Retrospective Se, Sp, AUC
Haenssle, 2018, [21] Germany Malignant and pigmented Dermoscopic Outpatient department Dermatologists Retrospective Se, Sp, AUC
Tschandl, 2018, [22] Austria Malignant, not pigmented Dermoscopic and Clinical Outpatient department Dermatologists and Family physicians Retrospective Se, Sp, AUC
Fujisawa, 2019, [23] Japan Malignant, pigmented or not Clinical Outpatient department Dermatologists Retrospective Se, Accuracy
Brinker, 04/2019, [24] Germany Malignant and pigmented Dermoscopic ISIC web database Dermatologists Retrospective Se, Sp, Youden
Tschandl, 2019, [25] Austria Malignant, pigmented or not Dermoscopic Outpatient departments Dermatologists Retrospective difference of correct diagnosis
Brinker, 08/2019, [26] Germany Malignant and pigmented Dermoscopic ISIC web database Dermatologists Retrospective Se, Sp
Maron, 2019, [27] Germany Malignant, pigmented or not Dermoscopic HAM10000 database Dermatologists Retrospective Se, Sp
Cho, 2019, [28] South Korea Lips, malignant, pigmented or not Clinical Outpatient departments Dermatologists and Non dermatologists Retrospective Se, Sp, AUC, Youden
Kim, 2020, [29] South Korea Onychomycosis Dermoscopic and Clinical Outpatient department Dermatologists Prospective Se, Sp, AUC Youden
Haenssle, 2020, [30] Germany Malignant, pigmented or not Dermoscopic and Clinical Unknown Dermatologists Prospective Se, Sp
Huang, 2020, [31] China Malignant, not pigmented Clinical Outpatient department Dermatologists Prospective Se, Spe, AUC, ROCc
Han, 01/2020, [32] South Korea Head and neck, malignant, pigmented or not Clinical Outpatient departments Dermatologists and Non dermatologists Prospective Se, Sp
Marchetti, 2020, [33] Italy Malignant and pigmented Dermoscopic ISIC web database Dermatologists Retrospective Se, Sp, AUC
Liu, 2020, [34] USA All types Clinical Primary care structures and Teledermatology Dermatologists and Family physicians Retrospective Accuray, Se
Fink, 2020, [35] Germany Malignant and pigmented Dermoscopic Outpatient departments Dermatologists Retrospective Se, Sp, DOR
Jinnai, 2020, [36] Japan Malignant and pigmented Dermoscopic Outpatient department Dermatologists Retrospective Se, Sp, accuracy
Lee, 2020, [37] South Korea Acral, malignant and pigmented Dermoscopic Outpatient department Dermatologists and Family physicians Retrospective Accuracy, Se, Sp
Han, 09/2020, [38] South Korea All types Clinical Outpatient departments Dermatologists Retrospective AUC, accuracy, Se
Wang, 2020, [39] China Malignant, pigmented or not. Psoriasis Dermoscopic and Clinical Outpatient department Dermatologists Prospective Se, Sp, accuracy
Lucius, 2020, [40] Spain Malignant and pigmented Dermoscopic HAM10000 database Family physicians Retrospective accuracy, TPR, TNR
Han, 11/2020, [41] South Korea Malignant, pigmented or not Clinical Outpatient department Dermatologists Retrospective Se, Sp, AUC
Muñoz-Lopez, 2021, [42] Chile All types Clinical Teledermatology Dermatologists and Family physicians Prospective accuracy
Tognetti, 2021, [43] Italy Malignant and pigmented Dermoscopic Outpatient departments Dermatologists Retrospective Se, Sp
Minagawa, 2021, [44] Japan Malignant, pigmented or not Dermoscopic Outpatient department Dermatologists Retrospective Se, Sp, accuracy
Zhao, 2021, [45] China Rosacea Clinical Unknown Dermatologists Retrospective ROC curve, accuracy
Jain, 2021, [46] USA All types Clinical Primary care structures Family physicians Retrospective agreement rate of the primary differential diagnosis
Zhu C-Y, 2021, [47] China All types Dermoscopic and Clinical Outpatient department Dermatologists Retrospective Se, Spe, Youden
Yang, 2021, [48] China Psoriasis Dermoscopic and Clinical Outpatient department Dermatologists Retrospective Se, Sp
Zhu X, 2022 [49] China Onychomycosis Dermoscopic Outpatient department Dermatologists Prospective OR, Se, Sp
Combalia, 2022, [50] Spain Malignant, pigmented or not Dermoscopic Outpatient departments Dermatologists Retrospective accuracy
Yu Z, 2022, [51] Chine Psoriasis or Seborrheic dermatitis Dermoscopic Outpatient department Dermatologists and Family physicians Prospective AUC, accuracy, Se, Sp
Escalé-Besa, 2023, [52] Spain All types Dermoscopic and Clinical Primary care structures and Teledermatology Dermatologists and Family physicians Prospective accuracy, Se, Sp
Anderson, 2023, [53] USA Malignant and pigmented Dermoscopic ISIC web database Dermatologists and Family physicians Retrospective Se, Sp, Accuracy
Li, 2023, [54] China All types Clinical Outpatient departments Dermatologists and Family physicians Prospective Se, Sp, accuracy
Open in a new tab

Among the 38 included studies, the focus varied: 12 addressed malignant pigmented skin lesions, 12 examined skin cancers generally (including pigmented lesions), 2 targeted skin cancers other than melanoma, 7 covered a broad range of skin diseases, and 6 investigated other specific conditions like onychomycosis or psoriasis. One study focused on both psoriasis and skin cancers.

The types of images studied were either dermoscopic images in 17 articles, or clinical images in 13 articles, and both dermoscopic and clinical images in 8 articles.

Cases from dermatology outpatient hospital departments were reported by 25 studies, 4 were collected in a primary care setting and 7 were collected from databases without information on the original setting of consultation. 2 articles did not provide any information on the patient setting.

The diagnosis performance of artificial intelligence was compared to dermatologists alone in 25 studies, to dermatologists and other physicians in 11 articles (9 of these mentioned “GPs” and 2 mentioned “non-dermatologists”) and to GPs alone in 2 articles.

Twenty-seven studies used retrospective data or pre-existing datasets, while 11 used prospectively collected data.

Conflicts of interest were declared in 17 studies out of the 38 included.

Risk of bias

Figure 2, based on the QUADAS-2 tool [16], summarizes the risk of bias assessment for each included study. Considering patient selection, few studies included a random or consecutive sample of images but instead selected specific proportions of each skin disease. Datasets were often not representative of outpatients, as malignant conditions were usually overestimated.

Fig. 2.

Fig. 2

Open in a new tab

Risk of bias assessment. Un = Unclear. N° of question in Appendix 2

Regarding the index test, interpretations were conducted without knowledge of the reference standard. Practitioner reader tests were sometimes carried out on datasets whose characteristics varied greatly from the algorithm's validation test which can impact practitioners’ performance. Considering reference, some studies had different reference standards between different skin condition, that could lead to classification bias.

Artificial intelligence performance

The results extracted from the articles included (Fig. 3), showed that AI was more accurate than dermatologists in 18 articles, AI’s accuracy was non-inferior to dermatologists in 12 articles and AI was less accurate than dermatologists in 4 articles (Fig. 3A).

Fig. 3.

Fig. 3

Open in a new tab

Artificial intelligence’s diagnosis performance compared to dermatologists and GPs

Regarding family physicians or non-dermatologists, 8 studies reported that AI was more accurate, 1 study found that AI’s accuracy was non-inferior to GPs and 1 study reported AI was less accurate than GPs (Fig. 3B).

Figure 4 shows the results for diagnosis performance extracted from each study included.

Fig. 4.

Fig. 4

Open in a new tab

Results for diagnosis performance from the studies included

Six articles evaluated whether human reader performance could be improved when their diagnosis was made after evaluating AI’s interpretation

In 2 studies, about malignant pigmented skin lesions [37] and all types of skin lesions [38], AI was shown to enhance dermatologists’ performance. 1 study on malignant, pigmented or not, skin lesions showed that AI would not increase their initial performance [28].

In the 5 studies evaluating AI as a tool for family physicians (or non-dermatologists), they achieved higher diagnosis performance when using artificial intelligence. These studies were related to malignant, pigmented or not, skin lesions [28], malignant and pigmented lesions [37, 40], all types of lesions [46], and psoriasis or seborrheic dermatitis lesions [51].

To analyze AI’s performance in subgroups, we focused on the three categories of skin lesions with the most studies: malignant and pigmented skin lesions, malignant lesions that are pigmented or not, and all types of skin lesions. This result is shown in Fig. 5.

Fig. 5.

Fig. 5

Open in a new tab

Studies results of AI’s diagnostic accuracy compared to physicians, depending on the type of lesion

AI is more likely to show better performance than humans’ when diagnosing malignant pigmented lesions, in identifying naevi from melanoma. For multiple skin cancer type diagnosis including pigmented cancers, AI would show non-inferior or better performance. When AI is facing a global spectrum of dermatological diseases, it doesn’t show a tendency to be more accurate, non-inferior, or inferior to human readers.

When excluding studies with biases on the outcome, for this subgroup analysis on types of skin lesions, there was no modification in the diagnosis performance tendency. From these excluded studies, for malignant pigmented lesions, 3 showed AI was non-inferior to human readers. For multiple skin cancer types, 3 showed AI was non-inferior to human readers, and one showed AI was more accurate. For multiple skin diseases, one showed AI was non-inferior to human readers, and one showed it was more accurate.

The type of image used often differed depending on the disease type. It is particularly noticeable for malignant pigmented skin lesions. In the 12 studies regarding this type of lesion, images were dermoscopic images, which offer more details in visual characteristics that AI can analyze.

Another subgroup analysis was carried out with the 4 studies taking place in a true primary care setting. Two of them did not report any benefits of using AI [42, 52]. On the other side, Liu et al. showed AI was non-inferior to dermatologists, and more accurate than GPs [34]. Jain et al. showed it could improve family physicians’ performance [46].

In the studies where AI was used as a tool for physicians, and in studies conducted in true primary care settings, AI’s diagnosis is represented by a degree of probability for multiple diseases. The different predictions are ranked in order of probability. Given this information, a physician would classify the lesion based on his knowledge and/or AI’s ratings. AI can be presented as a web application with instructions on how the image should be captured and uploaded, leading to an output of several potential diagnoses [42].

Meta-analysis results

Of the 27 studies looking at malignant melanoma diagnosis, after a detailed screening process, 11 were included in the meta-analysis (Fig. 1). Among the 11 studies with accessible or retrievable raw datasets, only three conducted a comparative evaluation of AI diagnostic accuracy against both board-certified dermatologists and general practitioners. Given the limited number of studies evaluating the diagnostic performance of general practitioners, the meta-analysis was therefore focused on comparing AI with dermatologists. Forest plots and SROC curves were generated to calculate the combined sensitivity, specificity and diagnostic odds ratio, along with the AUC of the SROC curve. The combined sensitivity and specificity were 0.86 (95% CI: 0.80; 0.90) and 0,88 (95% CI: 0,81; 0.93), respectively. The diagnostic odds ratio was 44.36 (95% CI: 29.28; 67.1), with an AUC of 0.922 for the SROC curve. The forest plots and SROC curves for the AI diagnosis of malignant melanoma are shown in Fig. 6.

Fig. 6.

Fig. 6

Open in a new tab

Forest plot of sensitivity (a) and specificity (b) and HSROC curve (c) for malignant melanoma diagnosis by AI

Discussion

As shown in this systematic review, artificial intelligence can enhance diagnostic performance in dermatology, particularly when compared to dermatologists and family physicians in controlled settings. The results of this systematic review and meta-analysis demonstrate that AI exhibits high sensitivity and specificity in diagnosing malignant melanoma. With a pooled sensitivity of 0.86 and a specificity of 0.88, AI effectively distinguishes between melanoma and benign nevi. Notably, these results were obtained using dermoscopic images, implying that general practitioners would require training in dermoscopy to effectively use the AI tools studied. Moreover, when clinicians were provided with AI-generated probability scores, AI showed potential to augment human diagnostic performance.

In contrast, across the four studies conducted in real-world primary care settings—where general practitioners or nurse practitioners managed patients with a wide range of dermatological conditions—AI did not demonstrate any significant diagnostic advantage over routine clinical practice.

Following the PRISMA guidelines, the study was conducted with references from multiple databases. Two reviewers selected studies independently, according to the inclusion/exclusion criteria defined prior to screening. Risk of bias was assessed using the QUADAS-2 tool. Subgroup evaluations were performed to identify specific factors influencing AI diagnostic performance characteristics.

This study has some limitations. Two deviations from the pre-established research protocol occurred: Embase, initially included in the planned databases, was not accessible, and gray literature was not screened. While Google Scholar offers broad coverage, its systematic use and quality filtering present challenges and that is why search was not conducted on it. Francophone databases were deemed less likely to hold unique primary studies on this specific international technology topic not captured elsewhere. This focused approach prioritized feasibility and evidence quality within the project's scope. Among the 27 studies included in the literature review, only 11 could be retained for the meta-analysis due to limited availability of source data. Additionally, only 3 studies with accessible data directly compared the diagnostic performance of AI systems to that of general practitioners in the context of melanoma diagnosis. This limited inclusion may have introduced selection bias and reduced the generalizability and statistical power of the meta-analysis findings.

For this review only MeSH terms were used. We aimed to maximize specificity and retrieve highly relevant articles, reducing the noise often associated with broad free-text searching, especially in rapidly evolving fields with inconsistent terminology. Many of the studies included were affected by the differential reference bias. For skin lesions, depending on the type of disease, the diagnosis must be obtained after biopsy and anatomopathological examination, which seems to be the most unbiased test reference. However, we often reach sufficient confidence in diagnosis with visual examination or non-invasive exams, avoiding the morbidity of going through biopsy. This explains why the differential reference bias has potentially occurred in a majority of studies.

Conclusion

Artificial intelligence appears to be either as accurate as, or more accurate than, dermatologists or GPs. Therefore, our study shows that artificial intelligence could be an effective tool for GPs to efficiently screen which cases need a dermatologist’s referral. Such a tool can take the form of a web application, receiving real-life images to analyze.

For a better appreciation, more prospective studies should be conducted in authentic primary care settings using real-life photographs, with the same gold-standard reference for each case, comparing how diagnosis performance improves, before and after a GP uses an AI-based tool, on a patient seen in real-life conditions.

Supplementary Information

Supplementary Material 1. Appendix A. (44.8KB, pdf)
Supplementary Material 2. Appendix B. (56.2KB, pdf)
Supplementary Material 3. Appendix C.  (411.7KB, docx)

Acknowledgements

Not applicable

Abbreviations

AI

Artificial intelligence

CNN

Convolutional neural network

GP

General practitioner

PICO

Population, intervention, comparison, outcome

PRISMA

Preferred reporting items for systematic reviews and meta-analyses

USA

United States of America

Authors’ contributions

N. N. and S. Z. screened articles based on titles and abstracts. N. N. screened articles based on full texts, extracted data. R. L. provided methodology guidance and risk of bias assessment. T. D. conducted the meta-analysis. All authors contributed to the main manuscript text, prepared figures and tables. All authors reviewed the manuscript.

Funding

Not applicable.

Data availability

In the references section, identification of each scientific articles published in the literature, from which data were collected.

Declarations

Ethics approval and consent to participate

Data collected from published scientific articles.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Footnotes

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Contributor Information

Norhane Nadour, Email: norhane.nadour@gmail.com.

Roxane Liard, Email: roxane.liard@gmail.com.

References

  • 1.Manning C. Artificial intelligence definitions. Stanford (U.S.A.): Stanford University; 2020 [cited 2024 Jun 28]. Available from: https://hai.stanford.edu/sites/default/files/2020-09/AI-Definitions-HAI.pdf
  • 2.Oxford Reference. Artificial Intelligence. Oxford (U.K.): Oxford University. 2024 [cited 2024 Jun 28]. Available from: https://www.oxfordreference.com/display/10.1093/oi/authority.20110803095426960
  • 3.High-Level Expert Group on Artificial Intelligence Rossi F. A definition of AI: Main capabilities and scientific disciplines. European Comission ; 2018. 7. Available from : https://ec.europa.eu/futurium/en/system/files/ged/ai_hleg_definition_of_ai_18_december_1.pdf
  • 4.Haiech J. Parcourir l’histoire de l’intelligence artificielle, pour mieux la définir et la comprendre. Med Sci (Paris). 2020;36(10):919–23. [DOI] [PubMed] [Google Scholar]
  • 5.Topol EJ. High-performance medicine: the convergence of human and artificial intelligence. Nat Med. 2019;25(1):44–56. [DOI] [PubMed] [Google Scholar]
  • 6.LeCun Y, Bengio Y, Hinton G. Deep learning. Nature. 2015;521(7553):436–44. [DOI] [PubMed] [Google Scholar]
  • 7.Lindsey R, Daluiski A, Chopra S, Lachapelle A, Mozer M, Sicular S, et al. Deep neural network improves fracture detection by clinicians. Proc Natl Acad Sci U S A. 2018;115(45):11591–6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Gulshan V, Peng L, Coram M, Stumpe MC, Wu D, Narayanaswamy A, et al. Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs. JAMA. 2016;316(22):2402. [DOI] [PubMed] [Google Scholar]
  • 9.Société Française de Dermatologie. Dans la peau des Français : Dossier de presse. SFD ; 2017. 20. Available from: https://www.sfdermato.org/upload/presse/sfd-dossier-de-presse-dans-la-peau-des-francais-698b341f63725ea142271ed9b39e0980.pdf
  • 10.Observatoire des Territoires. Densité de dermatologues libéraux. France. 2018 [cited 2024 Jun 28]. Available from: https://www.observatoire-des-territoires.gouv.fr/densite-de-dermatogologues-liberaux
  • 11.Le Guide Santé. 95 jours de délais moyen en France pour une consultation en dermatologie [Internet]. France : Caducee.net. 2020 [cited 2024 Jun 28]. Available from: https://www.caducee.net/actualite-medicale/14682/95-jours-de-delais-moyen-en-france-pour-une-consultation-en-dermatologie.html
  • 12.Santé publique France. Cancers de la peau. France. 2023 [cited 2024 Jul 30]. Available from: https://www.santepubliquefrance.fr/maladies-et-traumatismes/cancers/cancers-de-la-peau/donnees/#tabs
  • 13.Institut National du Cancer (INCa). Épidémiologie des cancers cutanés - Détection précoce des cancers de la peau [Internet]. France. 2023 [cited 2024 Jun 28]. Available from: https://www.e-cancer.fr/Professionnels-de-sante/Depistage-et-detection-precoce/Detection-precoce-des-cancers-de-la-peau/Epidemiologie
  • 14.Esteva A, Kuprel B, Novoa RA, Ko J, Swetter SM, Blau HM, et al. Dermatologist-level classification of skin cancer with deep neural networks. Nature. 2017;542(7639):115–8. 10.1038/nature21056. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Cochrane Handbook for Systematic Reviews of Diagnostic Test Accuracy. [cited 2025 April 21]. Available from: https://training.cochrane.org/handbook-diagnostic-test-accuracy/current
  • 16.Whiting PF, Rutjes AW, Westwood ME, Mallett S, Deeks JJ, Reitsma JB. QUADAS-2: a revised tool for the quality assessment of diagnostic accuracy studies. Ann Intern Med. 2011;155:529–36. 10.7326/0003-4819-155-8-201110180-00009. [DOI] [PubMed] [Google Scholar]
  • 17.Lee J, Kim KW, Choi SH, Huh J, Park SH. Systematic review and meta-analysis of studies evaluating diagnostic test accuracy: a practical review for clinical researchers-Part II. statistical methods of meta-analysis. Korean J Radiol. 2015;16(6):1188–96. 10.3348/kjr.2015.16.6.1188. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.C Yu S Yang W Kim J Jung KY Chung SW Lee et al Haass NK. Acral melanoma detection using a convolutional neural network for dermoscopy images. PLoS One. 2018; 13(3): e0193321 10.1371/journal.pone.0193321 [DOI] [PMC free article] [PubMed]
  • 19.Marchetti MA, Codella NCF, Dusza SW, Gutman DA, Helba B, Kalloo A, et al. Results of the 2016 International Skin Imaging Collaboration International Symposium on Biomedical Imaging challenge: comparison of the accuracy of computer algorithms to dermatologists for the diagnosis of melanoma from dermoscopic images. J Am Acad Dermatol. 2018;78(2):270-277.e1. 10.1016/j.jaad.2017.08.016. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Han SS, Kim MS, Lim W, Park GH, Park I, Chang SE. Classification of the clinical images for benign and malignant cutaneous tumors using a deep learning algorithm. J Invest Dermatol. 2018;138(7):1529–38. 10.1016/j.jid.2018.01.028. [DOI] [PubMed] [Google Scholar]
  • 21.Haenssle HA, Fink C, Schneiderbauer R, Toberer F, Buhl T, Blum A, et al. Man against machine: diagnostic performance of a deep learning convolutional neural network for dermoscopic melanoma recognition in comparison to 58 dermatologists. Ann Oncol. 2018;29(8):1836–42. 10.1093/annonc/mdy166. [DOI] [PubMed] [Google Scholar]
  • 22.Tschandl P, Rosendahl C, Akay BN, Argenziano G, Blum A, Braun RP, et al. Expert-level diagnosis of nonpigmented skin cancer by combined convolutional neural networks. JAMA Dermatol. 2019;155(1):58. 10.1001/jamadermatol.2018.4378. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Fujisawa Y, Otomo Y, Ogata Y, Nakamura Y, Fujita R, Ishitsuka Y, et al. Deep-learning-based, computer-aided classifier developed with a small dataset of clinical images surpasses board-certified dermatologists in skin tumour diagnosis. Br J Dermatol. 2019;180(2):373–81. 10.1111/bjd.16924. [DOI] [PubMed] [Google Scholar]
  • 24.Brinker TJ, Hekler A, Enk AH, Klode J, Hauschild A, Berking C, et al. Deep learning outperformed 136 of 157 dermatologists in a head-to-head dermoscopic melanoma image classification task. Eur J Cancer. 2019;113:47–54. 10.1016/j.ejca.2019.04.001. [DOI] [PubMed] [Google Scholar]
  • 25.Tschandl P, Codella N, Akay BN, Argenziano G, Braun RP, Cabo H, et al. Comparison of the accuracy of human readers versus machine-learning algorithms for pigmented skin lesion classification: an open, web-based, international, diagnostic study. Lancet Oncol. 2019;20(7):938–47. 10.1016/S1470-2045(19)30333-X. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Brinker TJ, Hekler A, Enk AH, Berking C, Haferkamp S, Hauschild A, et al. Deep neural networks are superior to dermatologists in melanoma image classification. Eur J Cancer. 2019;119:11–7. 10.1016/j.ejca.2019.05.023. [DOI] [PubMed] [Google Scholar]
  • 27.Maron RC, Weichenthal M, Utikal JS, Hekler A, Berking C, Hauschild A, et al. Systematic outperformance of 112 dermatologists in multiclass skin cancer image classification by convolutional neural networks. Eur J Cancer. 2019;119:57–65. 10.1016/j.ejca.2019.06.013. [DOI] [PubMed] [Google Scholar]
  • 28.Cho SI, Sun S, Mun J-H, Kim C, Kim SY, Cho S, et al. Dermatologist-level classification of malignant lip diseases using a deep convolutional neural network. Br J Dermatol. 2020;182(6):1388–94. 10.1111/bjd.18459. [DOI] [PubMed] [Google Scholar]
  • 29.Kim YJ, Han SS, Yang HJ, Chang SE. Prospective, comparative evaluation of a deep neural network and dermoscopy in the diagnosis of onychomycosis. PLoS ONE. 2020;15(6):e0234334. 10.1371/journal.pone.0234334. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Haenssle HA, Fink C, Toberer F, Winkler J, Stolz W, Deinlein T, et al. Man against machine reloaded: performance of a market-approved convolutional neural network in classifying a broad spectrum of skin lesions in comparison with 96 dermatologists working under less artificial conditions. Ann Oncol. 2020;31(1):137–43. 10.1016/j.annonc.2019.10.013. [DOI] [PubMed] [Google Scholar]
  • 31.Huang K, He X, Jin Z, Wu L, Zhao X, Wu Z, et al. Assistant diagnosis of basal cell carcinoma and seborrheic keratosis in Chinese population using convolutional neural network. J Healthc Eng. 2020;2020:1–8. 10.1155/2020/1713904. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Han SS, Moon IJ, Lim W, Suh IS, Lee SY, Na JI, et al. Keratinocytic skin cancer detection on the face using region-based convolutional neural network. JAMA Dermatol. 2020;156(1):29. 10.1001/jamadermatol.2019.3807. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Marchetti MA, Liopyris K, Dusza SW, Codella NCF, Gutman DA, Helba B, et al. Computer algorithms show potential for improving dermatologists’ accuracy to diagnose cutaneous melanoma: results of the International Skin Imaging Collaboration 2017. J Am Acad Dermatol. 2020;82(3):622–7. 10.1016/j.jaad.2019.07.016. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Liu Y, Jain A, Eng C, Way DH, Lee K, Bui P, et al. A deep learning system for differential diagnosis of skin diseases. Nat Med. 2020;26(6):900–8. 10.1038/s41591-020-0842-3. [DOI] [PubMed] [Google Scholar]
  • 35.Fink C, Blum A, Buhl T, Mitteldorf C, Hofmann-Wellenhof R, Deinlein T, et al. Diagnostic performance of a deep learning convolutional neural network in the differentiation of combined naevi and melanomas. J Eur Acad Dermatol Venereol. 2020;34(6):1355–61. 10.1111/jdv.16165. [DOI] [PubMed] [Google Scholar]
  • 36.Jinnai S, Yamazaki N, Hirano Y, Sugawara Y, Ohe Y, Hamamoto R. The development of a skin cancer classification system for pigmented skin lesions using deep learning. Biomolecules. 2020;10(29):1123. 10.3390/biom10081123. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Lee S, Chu YS, Yoo SK, Choi S, Choe SJ, Koh SB, et al. Augmented decision-making for acral lentiginous melanoma detection using deep convolutional neural networks. J Eur Acad Dermatol Venereol. 2020;34(8):1842–50. 10.1111/jdv.16185. [DOI] [PubMed] [Google Scholar]
  • 38.Han SS, Park I, Eun Chang S, Lim W, Kim MS, Park GH, 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):1753–61. 10.1016/j.jid.2020.01.019. [DOI] [PubMed] [Google Scholar]
  • 39.Wang SQ, Zhang XY, Liu J, Tao C, Zhu CY, Shu C, et al. Deep learning-based, computer-aided classifier developed with dermoscopic images shows comparable performance to 164 dermatologists in cutaneous disease diagnosis in the Chinese population. Chin Med J (Engl). 2020;133(17):2027–36. 10.1097/CM9.0000000000001023. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Lucius M, De All J, De All JA, Belvisi M, Radizza L, Lanfranconi M, et al. Deep neural frameworks improve the accuracy of general practitioners in the classification of pigmented skin lesions. Diagnostics. 2020;10(11):969. 10.3390/diagnostics10110969. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Han SS, Moon IJ, Kim SH, Na JI, Kim MS, Park GH, et al. Assessment of deep neural networks for the diagnosis of benign and malignant skin neoplasms in comparison with dermatologists: a retrospective validation study. PLoS Med. 2020;17(11):e1003381. 10.1371/journal.pmed.1003381. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Muñoz-López C, Ramírez-Cornejo C, Marchetti MA, Han SS, Del Barrio-Díaz P, Jaque A, et al. Performance of a deep neural network in teledermatology: a single-centre prospective diagnostic study. J Eur Acad Dermatol Venereol. 2021;35(2):546–53. 10.1111/jdv.16979. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Tognetti L, Bonechi S, Andreini P, Bianchini M, Scarselli F, Cevenini G, et al. A new deep learning approach integrated with clinical data for the dermoscopic differentiation of early melanomas from atypical nevi. J Dermatol Sci. 2021;101(2):115–22. 10.1016/j.jdermsci.2020.11.009. [DOI] [PubMed] [Google Scholar]
  • 44.Minagawa A, Koga H, Sano T, Matsunaga K, Teshima Y, Hamada A, et al. Dermoscopic diagnostic performance of Japanese dermatologists for skin tumors differs by patient origin: a deep learning convolutional neural network closes the gap. J Dermatol. 2021;48(2):232–6. 10.1111/1346-8138.15640. [DOI] [PubMed] [Google Scholar]
  • 45.Zhao Z, Wu CM, Zhang S, He F, Liu F, Wang B, et al. A novel convolutional neural network for the diagnosis and classification of rosacea: usability study. JMIR Med Inform. 2021;9(3):e23415. 10.2196/23415. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46.Jain A, Way D, Gupta V, Gao Y, De Oliveira MG, Hartford J, et al. Development and Assessment of an Artificial Intelligence-Based Tool for Skin Condition Diagnosis by Primary Care Physicians and Nurse Practitioners in Teledermatology Practices. JAMA Netw Open. 2021;4(4):e217249. 10.1001/jamanetworkopen.2021.7249. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.Zhu CY, Wang YK, Chen HP, Gao KL, Shu C, Wang JC, et al. A Deep Learning Based Framework for Diagnosing Multiple Skin Diseases in a Clinical Environment. Front Med. 2021;16(8):626369. 10.3389/fmed.2021.626369. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48.Yang Y, Wang J, Xie F, Liu J, Shu C, Wang Y, et al. A convolutional neural network trained with dermoscopic images of psoriasis performed on par with 230 dermatologists. Comput Biol Med. 2021;139:104924. 10.1016/j.compbiomed.2021.104924. [DOI] [PubMed] [Google Scholar]
  • 49.Zhu X, Zheng B, Cai W, Zhang J, Lu S, Li X, et al. Deep learning-based diagnosis models for onychomycosis in dermoscopy. Mycoses. 2022;65(4):466–72. 10.1111/myc.13427. [DOI] [PubMed] [Google Scholar]
  • 50.Combalia M, Codella N, Rotemberg V, Carrera C, Dusza S, Gutman D, et al. Validation of artificial intelligence prediction models for skin cancer diagnosis using dermoscopy images: the 2019 International Skin Imaging Collaboration Grand Challenge. Lancet Digit Health. 2022;4(5):e330–9. 10.1016/S2589-7500(22)00021-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 51.Yu Z, Kaizhi S, Jianwen H, Guanyu Y, Yonggang W. A deep learning-based approach toward differentiating scalp psoriasis and seborrheic dermatitis from dermoscopic images. Front Med. 2022;9:965423. 10.3389/fmed.2022.965423. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 52.Escalé-Besa A, Yélamos O, Vidal-Alaball J, Fuster-Casanovas A, Miró Catalina Q, Börve A. Exploring the potential of artificial intelligence in improving skin lesion diagnosis in primary care. Sci Rep. 2023;13(1):4293. 10.1038/s41598-023-31340-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 53.Anderson JM. Artificial Intelligence vs Medical Providers in the Dermoscopic Diagnosis of Melanoma. Cutis. 2023 May [cited 2024 Jun 28];111(5). Available from: https://www.mdedge.com/dermatology/article/262697/melanoma/artificial-intelligence-vs-medical-providers-dermoscopic. 10.12788/cutis.0764 [DOI] [PubMed]
  • 54.Li H, Zhang P, Wei Z, Qian T, Tang Y, Hu K, et al. Deep skin diseases diagnostic system with dual-channel image and extracted text. Front Artif Intell. 2023;6:1213620. 10.3389/frai.2023.1213620. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Material 1. Appendix A. (44.8KB, pdf)
Supplementary Material 2. Appendix B. (56.2KB, pdf)
Supplementary Material 3. Appendix C.  (411.7KB, docx)

Data Availability Statement

In the references section, identification of each scientific articles published in the literature, from which data were collected.

Articles from BMC Primary Care are provided here courtesy of BMC

RESOURCES