{"doi":"10.1093/bjd/ljac047","title":"Evaluation of diagnosis diversity in artificial intelligence datasets: a scoping review","abstract":"https://doi.org/10.1093/bjd/ljac047 Dear Editor, Artificial intelligence (AI) algorithms are increasingly used for clinical tasks within dermatology, with some algorithms having regulatory approval in Europe and Australia.1 However, there are concerns about the transparency of these algorithms regarding patient diversity.1,2 Recent work showed that Fitzpatrick skin tone information was reported in only 10% of dermatological AI datasets.1 To our knowledge, differences in dermatology AI performance based on sex and gender have not been investigated. A randomized, prospective clinical trial found that an AI algorithm performed poorly on diagnoses it had never encountered during algorithmic training.3 To understand whether certain diagnoses are not represented among dermatology AI algorithms, we conducted a scoping review of published algorithms and assessed potential biases against women or patients with skin of colour. We queried PubMed and MEDLINE using the following terms: deep learning, machine learning or artificial intelligence; with dermatology, dermatologist or skin. We included clinically relevant, English-language, peer-reviewed articles from 1 January 2015 to 1 March 2021. All papers utilizing deep learning applied to imaging, histopathology or sociodemographic or clinical data were included. The title and abstract of each paper were reviewed to select studies meeting the inclusion criteria, and full-text reviews of the selected papers were conducted to confirm adherence to the selection criteria. For each study dataset, M.L.C. collected the disease list when the number of cases per diagnosis was provided in the article, using a previously described process.1 Two-hundred and three articles met our inclusion criteria. The data types included imaging only (168 papers), histopathology only (21 papers), sociodemographic or clinical data only (seven papers) and combined imaging and sociodemographic or clinical data (seven papers). Of the 203 articles, 183 included a partial or full breakdown of disease distribution. We grouped diagnoses using a previously reported taxonomy.4 We observed 242 dermatological diseases and identified 49 high-level diagnostic groupings. The most frequently identified diagnostic groupings were malignant (135 papers) and benign (106 papers) pigmented lesions. Figure 1(a) depicts frequently represented categories. (a) Overview of the most common diagnostic categories among the 203 publications. (b) Number of images in diagnoses more common or enriched in patients with skin of colour (blue), diagnoses more common or enriched in women (orange), and diagnoses more common or enriched in both patients with skin of colour and women (green). The number of papers containing at least one case of each diagnosis is also shown. We identified diagnoses either commonly seen in or with higher prevalence in women or patients with skin of colour based on previously reported studies and consultation with a board-certified dermatologist and skin of colour expert (J.C.L.) and two board-certified dermatologists (A.S.C. and R.D.).5 Our final list of diagnoses was confirmed for relevance using targeted literature search terms based on the selected diagnoses. Example findings included one study of a skin of colour clinic that identified acne and dyschromia as two common diagnoses.6 Uncommon conditions that have increased prevalence or incidence in patients with skin of colour include cutaneous T-cell lymphoma, sarcoidosis and hidradenitis suppurativa. Diagnoses with increased prevalence in women included contact dermatitis, atopic dermatitis and urticaria.7 Seventeen diagnoses were identified as more common or enriched in patients with skin of colour and seven were identified for women (Figure 1b). We determined the number of papers containing at least one case of each diagnosis (Figure 1b). Diagnoses seen more often by skin of colour specialists were underrepresented: acne (13 papers), atopic dermatitis (five papers), seb","journal":"British Journal of Dermatology","year":2022,"id":283030,"datarank":0.0,"base_score":0.0,"endowment":0.0,"self_citation_contribution":0.0,"citation_network_contribution":0.0,"self_endowment_contribution":0.0,"citer_contribution":0.0,"corpus_percentile":null,"corpus_rank":null,"citation_count":7,"citer_count":0,"citers_with_citation_signal":0,"citers_with_endowment":0,"datacite_reuse_total":0,"is_dataset":false,"is_dataset_confidence":0.9576,"is_data_producer":false,"deposit_databanks":null,"is_oa":false,"file_count":0,"downloads":0,"has_version_chain":false,"published_date":"2022-01-01","fair_score":null,"fair_percentile":null,"algorithm_id":"datarank_citation_only_1hop_v6","ranking_scope":"data_only","authors":[{"id":620922,"name":"Veronica Rotemberg","orcid":"0000-0003-0639-2677","position":1,"is_corresponding":false},{"id":759528,"name":"Jenna Lester","orcid":"0000-0003-1849-1082","position":2,"is_corresponding":false},{"id":737581,"name":"Roberto A. Novoa","orcid":"0000-0002-7955-0536","position":3,"is_corresponding":false},{"id":759529,"name":"Albert S. Chiou","orcid":"0000-0002-8436-0008","position":4,"is_corresponding":false},{"id":177142,"name":"Roxana Daneshjou","orcid":"0000-0001-7988-9356","position":5,"is_corresponding":false},{"id":852791,"name":"Michael L. Chen","orcid":"0000-0002-0168-1122","position":0,"is_corresponding":true}],"reference_count":9,"raw_metadata":null,"created_at":"2026-07-19T00:29:23.777809Z","pmid":"36763858","pmcid":null,"fwci":null,"citation_percentile":null,"influential_citations":0,"oa_status":null,"license":null,"views":0,"total_file_size_bytes":0,"version_count":0,"fair_f":null,"fair_a":null,"fair_i":null,"fair_r":null,"fair_zscore":null,"fair_rationale":null,"fair_model":null,"fair_agent_version":null,"fair_fulltext_source":null,"fair_has_llm":null,"fair_computed_at":null,"clinical_trials":[],"software_tools":[],"db_accessions":[],"linked_datasets":[],"topics":[]}