Artificial Intelligence Applications in Dermatology: Where Do We Stand?
Arieh Gomolin
Elena Netchiporouk
Robert Gniadecki2* 
Ivan V. Litvinov- 1Division of Dermatology, McGill University Health Centre, Montreal, QC, Canada
- 2Division of Dermatology, University of Alberta, Edmonton, AB, Canada
Artificial intelligence (AI) has become a progressively prevalent Research Topic in medicine and is increasingly being applied to dermatology. There is a need to understand this technology's progress to help guide and shape the future for medical care providers and recipients. We reviewed the literature to evaluate the types of publications on the subject, the specific dermatological topics addressed by AI, and the most challenging barriers to its implementation. A substantial number of original articles and commentaries have been published to date and only few detailed reviews exist. Most AI applications focus on differentiating between benign and malignant skin lesions, however; others exist pertaining to ulcers, inflammatory skin diseases, allergen exposure, dermatopathology, and gene expression profiling. Applications commonly analyze and classify images, however, other tools such as risk assessment calculators are becoming increasingly available. Although many applications are technologically feasible, important implementation barriers have been identified including systematic biases, difficulty of standardization, interpretability, and acceptance by physicians and patients alike. This review provides insight into future research needs and possibilities. There is a strong need for clinical investigation in dermatology providing evidence of success overcoming the identified barriers. With these research goals in mind, an appropriate role for AI in dermatology may be achieved in not so distant future.
Introduction
Dermatology is a field with the growing interplay of digitalization, telehealth, and informatics (1). The increasing presence of artificial intelligence (AI) worldwide, has led to numerous attempts to leverage this technology for dermatological applications (2). In a recent international survey of 1,271 dermatologists, 85.1% of responders were aware of AI as an emerging topic in their field yet only 23.8% had good or excellent knowledge on the subject (3). Moreover, 77.3% agreed that AI will improve dermatologic care and 79.8% thought that AI should be a part of medical training. Informing stakeholders on the current stance of AI is thus necessary to promote what dermatologists themselves believe to be a growing, beneficial and potentially obligatory aspect of the field. To date, many publications exist on specific AI topics in dermatology but few provide a basic overview and address the wide landscape. In this article, we summarize the status of the literature on AI in dermatology using three sections. First, we highlight the types of articles published on this subject. We then focus on dermatologic diseases targeted by AI, and finally, we spotlight the identified barriers impeding AI implementation.
Types of Articles Published
Original Research
The overwhelming majority of articles published to date are original research articles. These describe the design of AI applications that can perform dermatology-related tasks. For example, articles have studied tools that can segment a psoriasis lesion or differentiate between benign and malignant skin lesions (4, 5). Although these AI tools have not yet been implemented clinically, these papers describe their technological feasibility and identify the potential clinical relevance should they be further validated. Most of these studies are authored by engineering researchers with few dermatologists as co-authors. Although the number is increasing, relatively few papers involve significant dermatologist collaboration in conceiving, designing and interpreting the studies. To overcome known barriers to clinical implementation, partnership with dermatologists is key. Zakhem et al. highlight in their recent review of melanoma screening applications, that when dermatologists were involved in the study design, the AI applications leveraged significantly larger patient datasets that were more representative of true clinical scenarios (6). Another important form of collaboration is conducting prospective clinical trials and very few of these exist (7). One landmark study was conducted by Dreiseitl et al. (8). In their protocol, patients with undiagnosed pigmented lesions presented to a dermatology clinic and were assessed separately by both non-expert physicians using an AI device and by expert dermatologists. The study design therefore resembled a realistic clinical practice scenario. The results indicated inferiority of the automated system, and unfortunately, no similar studies have been published to date. Whether the lack of similar studies is a publication bias or a literature gap remains unclear.
Reviews
A small number of systematic reviews exist at this time. Most of these cover the potential use of AI in differentiating between benign and malignant skin lesions. For example, studies have reviewed the specificities and sensitivities of AI tools for melanoma screening (9). To the best of our knowledge, only one systematic review has been published on dermatological applications of AI in general, not limited to neoplastic lesions (10).
Commentaries
An extensive amount of commentaries exist on the topic of AI in dermatology. These papers either highlight the potential impact of AI or stress the challenges of its implementation (11, 12).
Dermatological Applications of AI
Keratinocyte Carcinomas and Melanoma
There is an abundant and growing body of research demonstrating the preliminary success of AI applications at distinguishing between benign nevi vs. melanoma (5, 13–34). The main principle behind these applications is that dermatoscopic or non-dermatoscopic images of lesions can be broken down into individual pixels for analysis. A representative example by Jafari et al. describes an application that examines images pixel by pixel and extracts 60 features from each to predict disease classification (24). These applications are typically validated by comparing their ability to correctly diagnose lesions to the ability of certified dermatologists (32). One review of photo recognition applications by Safran et al. included 48 melanoma-screening tools and demonstrated a mean sensitivity of 87.60% and a mean specificity of 83.54% (9). Interest toward this topic has grown to the extent that an international skin imaging competition was founded in 2016 and has been occurring annually since (32, 35). Although these applications have become more robust, prospective clinical trials are rare and known implementation barriers are continuously debated.
An increasing number of original studies have also begun classifying non-melanoma skin cancers (also known as keratinocyte carcinomas) vs. benign and pre-malignant lesions (36–44). For example, Spyridonos et al. developed an AI model that could differentiate between actinic keratosis and normal skin with a specificity of 89.8% and a sensitivity of 91.7% (37). Altogether, most of the research on the topic of skin cancer demonstrates technological feasibility combined with the growing evidence supporting clinical utility. What remains to be demonstrated is whether such tools can be implemented and relied upon in daily clinical practice.
AI has also been used beyond photo recognition. Rather than processing image pixels, applications can also process numerical values in various sequences and extract trends. For example, Tan et al. described an application that predicted the complexity of micrographic Mohs surgery based on variables assessed at the initial evaluation visit such as tumor size and patient age (43). They were able to create a preliminary model that could theoretically be used to triage patients and prioritize Mohs referrals. Although most of the research on AI and skin cancer is based on photo recognition algorithms, other opportunities exist.
Ulcer Assessment
There is a growing body of research on diabetic and pressure ulcer applications (45–48). Thus, far most studies demonstrate methods for improving wound assessments using image recognition (45).
Articles have described applications capable of measuring precise wound boundaries, and differentiating between the types of tissue involved (45–47, 49). For example, Dhane et al. demonstrated an AI application's ability to segment the area of ill-defined ulcers with a sensitivity of 87.3% and specificity of 95.7% (47). Mukerjee et al. demonstrated an AI application's ability to classify granulation, slough and necrotic tissue with 87.61% accuracy (46). Risk prediction tools also exist. Alderden et al. described a tool that leverages data in the electronic health records of admitted patients, to predict their tendency to develop pressure ulcers (50). Altogether, these applications have preliminarily been shown to be technologically feasible, they have not yet been validated extensively in clinically trials.
Psoriasis and Other Inflammatory Skin Diseases
Several original research articles exist on AI applications for inflammatory dermatoses. Most of these studies thus far have focused on improving psoriasis classification methods using image recognition (4, 51, 52). A representative example is a study conducted by Shrivasta et al., which compared the ability of several applications at classifying the severity of psoriasis lesions. The systems described achieved average sensitivities between 93.81 and 99.76% and average specificities between 97 and 99.99% (4).
Emam et al. described a psoriasis application beyond image recognition. They demonstrated a system's ability to predict psoriasis patient responses to biologic therapy using parameters gauged at an initial visit such as patient's weight and age of onset of psoriasis (53). They were able to create a preliminary model that could theoretically be used to optimize therapy for patients.
Beyond psoriasis, applications have been described classifying acne, lichen planus, pityriasis lichenoides and dermatomyositis (10, 54–58). Seite et al. developed a smartphone AI tool that grades and classifies types of acne lesions (e.g., comedonal, inflammatory, post inflammatory hyperpigmentation, etc.) (59). Huang et al. developed a multi disease classifier that could analyse 34 attributes (e.g., erythema, scaling, definite borders, etc.) and differentiate several papulosquamous diseases such as psoriasis, seborheic dermatitis, lichen planus, pityriasis, and chronic dermatitis (58).
Although most of these experiments relied on images of skin, one application assessed muscle ultrasound images and differentiated between normal muscle, dermatomyosisits, polymyositis, and inclusion body myositis with accuracies between 76.2 and 86.6% (60). Altogether, the theoretical utility of these applications for inflammatory diseases is significant, both further technological validation and clinical experimentation are needed.
Predicting Skin Sensitization Substances
Research is also accumulating on using AI to minimize exposure to skin-sensitizing substances (61–65). A representative example by Zang et al. described an application capable of analyzing physiochemical properties of substances (e.g., melting point) and determining whether the substance could be a sensitizer or not (65). This application yielded an accuracy of 81% when the substances were studied in a human cohort. Wilm et al. reviewed current advances in skin sensitization testing and highlighted several other examples, where AI has provided a method to reduce animal testing (66). While this use of AI can have an impact on a population wide level, significant technological and clinical validation studies are necessary.
Novel Applications in Pathology and Gene Expression Profiling
Applications have been described that can automate histology image processing and classification (67–71). For example, Arevalo et al. described a system that analyzes histopathological images and can classify basal cell carcinoma with 98.1% accuracy (67). Olsen et al. described a system that diagnosed dermal nevi and seborrheic keratosis with high accuracies and may serve as a future method to increase the efficiency of analyzing these prevalent benign tumors (72). Algorithms have also been described that can identify predictive genes and biomarkers for diseases (73–83). A representative example by Reimann et al., described an AI model capable of diagnosing psoriasis vulgaris based on the expression level of 4 genes with 96.4% accuracy (78). In another study assessing genetic differences in psoriasis genotypes, Patrick et al. used a combination of statistical learning and machine learning to identify new loci and predict the tendency of cutaneous psoriasis patients to develop psoriatic arthritis symptoms (84). While these investigations are in an early phase, the potential for their impact can be significant.
Identified Barriers
Choice of Predictive Model
AI algorithms are continually being developed and each has advantages and challenges. Beam et al. discussed the relationship of AI compared to more classic statistical models (85). They detail how predictive technologies can be viewed on a machine-learning spectrum. Statistical models are lower on the spectrum because humans impose assumptions and guide many aspects of the algorithm. True machine learning is highest on the spectrum because the algorithms evolve without human involvement. A systematic review by Christodoulou et al. found no objective advantage of machine learning compared to longitudinal regression for binary clinical prediction (86). However, the review does summarize various theoretical reasons why machine learning may be superior to longitudinal regression in certain instances such as processing data with a strong signal to noise ratio (e.g., handwriting) or with a significant number of predictor subcategories (e.g., images) (86). There are therefore many unanswered questions regarding whether advanced forms of AI are actually needed or if more primitive technologies can accomplish the same tasks.
Generalizability
One of the main limitations to AI is that the decisions made by these technologies are ultimately a reflection of the input data used to train the system (87). This theoretically implies that applications can only be used reliably in populations they were trained to assess. If applications are trained in one population and tested in another, the results are technically not generalizable and are subject to systematic biases such as overfitting. For example, Han et al. experimented with a skin cancer detection algorithm and concluded that overall performance could be improved if trained with a wide variety of data from multiple ethnic populations (36). However, simply using more data does not necessarily solve this problem. For example, Navarrete-Dechent et al. took Han et al.'s established an AI algorithm that was trained with a relatively diverse set of data and tested it in a unique database of Caucasian Americans from the southern United States. They found that the performance was suboptimal compared to how it was reported originally (88). The issue of generalizability is thus not simple to solve and may require either unique or extended data depending on the composition of the population being tested. This tendency for systematic bias has numerous implications for dermatology given the various demographic factors that affect making a diagnosis such as age, gender, race, and ethnicity to name a few.
Standardization
Even if the application is trained using data from the correct population, images of new lesions need to be comparable. What angle should the image be taken at? What lighting should the room have? What should the background be? Are there pen markings? These are factors that can affect decision-making by AI. For example, a study by Winkler et al. showed that surgical markings significantly interfered with the ability of a system to correctly diagnose dermatoscopic images of melanoma and increased the false positive rate (89). Artificial intelligence relies on standardization and there are numerous non-standardized aspects to dermatology unlike in other specialties (e.g., radiology) (90). Although databases are intentionally large to account for variability, factors such as these create an infinite possibility for divergence.
Data Requirements
One large barrier is the prerequisite for copious quantities of data of appropriate quality to power AI algorithms (91). A growing effort in the United States has been to solve this barrier using DataDerm, the American Academy of Dermatology's electronic health record system. Worldwide collaboration is likely required to achieve the ideal scenario where all of the necessary data categories are represented.
Interpretability
Artificial intelligence algorithms are formed, re-evaluated and constantly changing without human input. This is why the technology has often been termed a “black box” technology (92). Although AI is therefore flexible and can theoretically accomplish more than humans and human-guided statistical algorithms, many aspects and certainly the logic behind the decision-making is often not interpretable. When a certified dermatologist conducts a personalized assessment and arrives at a conclusion, that conclusion can be rationalized and explained based on existing clinical evidence. At this time, decisions made by AI cannot be interpreted in this way. This is a strong limitation which influences whether society and regulatory bodies will accept it in the daily practice of medicine.
Acceptance
A proper history followed by a physical examination in a well-lit examining room, while assessing for texture and eliciting specific signs for a given lesion (e.g., Darier, dimple, buttonhole signs, etc.) complemented by additional investigations/imaging or a biopsy is a standard way to establish a diagnosis in dermatology. Furthermore, it is accepted that while some diagnoses are clinical, others rest solely on histologic findings or a combination of clinical and histologic results correlation. This holistic approach cannot be fully replaced by computer programs and this is felt to be one of the most important barriers to implementing AI (93). Many patients also want to see and partner with a physician who is vested in helping them and may not be satisfied with isolated computerized tools (12).
Liability
There is also the issue of liability (94). If AI is relied upon and an adverse outcome ensues, is the dermatologist responsible? With this in mind, a common belief is that AI will only become a guidance tool and not an absolute diagnostic tool.
Next Steps
To address these barriers, several broad recommendations have been made to date. One clear need is for prospective clinical trials. If the question is whether AI can improve a dermatologic clinical encounter then studies that revolve around the clinical encounter are crucial (7). Dermatologist collaboration has also been highlighted as essential (6). Systems need to be trained with the full spectrum of human populations and clinical presentations that challenge dermatologists in clinical practice (88). Systems can also benefit from receiving inputs on other metrics available to physicians such as anatomic location, duration of the lesion and images of unaffected skin (88). Standardization practices also need to be implemented for photographing new lesions. Finally, given the lack of interpretability of many AI applications, we hypothesize that improving the lay descriptions of the algorithms and study designs can lead to improved acceptance by physicians and society at large. This would also aid regulatory decision makers who will need to adopt stances on liability.
Conclusion
AI is being increasingly studied in dermatology. Although most applications involve analyzing and classifying images, there are other tools such as risk assessment calculators. The most progress thus far has taken place in the field of melanoma diagnosis, followed by ulcer and psoriasis assessment tools, then followed by numerous less frequently studied applications. However, critical barriers and literature gaps exist that significantly limit AI's applicability to clinical practice at this time. For the less common applications, technological papers and commentaries are needed to improve capabilities and provoke interest. For the more saturated topics, there is a larger need for clinical trials providing evidence of clinical efficacy, while successfully overcoming the identified barriers. With these research goals in mind, an appropriate role for AI in dermatology may be achieved.
Author Contributions
AG, EN, RG, and IL performed the literature review and wrote the paper. RG and IL supervised the project.
Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
References
1. El-Azhary RA. The inevitability of change. Clin Dermatol. (2019) 37:4–11. doi: 10.1016/j.clindermatol.2018.09.003
2. Jiang F, Jiang Y, Zhi H, Dong Y, Li H, Ma S, et al. Artificial intelligence in healthcare: past, present and future. Stroke Vasc Neurol. (2017) 2:230–43. doi: 10.1136/svn-2017-000101
3. Polesie S, Gillstedt M, Kittler H, Lallas A, Tschandl P, Zalaudek I, et al. Attitudes towards artificial intelligence within dermatology: an international online survey. Br J Dermatol. (2020). doi: 10.1111/bjd.18875. [Epub ahead of print].
4. Shrivastava VK, Londhe ND, Sonawane RS, Suri JS. A novel and robust Bayesian approach for segmentation of psoriasis lesions and its risk stratification. Comp Methods Programs Biomed. (2017) 150:9–22. doi: 10.1016/j.cmpb.2017.07.011
5. Xie F, Fan H, Li Y, Jiang Z, Meng R, Bovik A. Melanoma classification on dermoscopy images using a neural network ensemble model. IEEE Trans Med Imaging. (2017) 36:849–58. doi: 10.1109/TMI.2016.2633551
6. Zakhem GA, Fakhoury JW, Motosko CC, Ho RS. Characterizing the role of dermatologists in developing AI for assessment of skin cancer: a systematic review. J Am Acad Dermatol. (2020). doi: 10.1016/j.jaad.2020.01.028. [Epub ahead of print].
7. Dick V, Sinz C, Mittlbock M, Kittler H, Tschandl P. Accuracy of computer-aided diagnosis of melanoma: a meta-analysis. JAMA Dermatol. (2019) 155:1291–9. doi: 10.1001/jamadermatol.2019.1375
8. Dreiseitl S, Binder M, Hable K, Kittler H. Computer versus human diagnosis of melanoma: evaluation of the feasibility of an automated diagnostic system in a prospective clinical trial. Melan Res. (2009) 19:180–4. doi: 10.1097/CMR.0b013e32832a1e41
9. Safran T, Viezel-Mathieu A, Corban J, Kanevsky A, Thibaudeau S, Kanevsky J. Machine learning and melanoma: the future of screening. J Am Acad Dermatol. (2018) 78:620–1. doi: 10.1016/j.jaad.2017.09.055
10. Thomsen K, Iversen L, Titlestad TL, Winther O. Systematic review of machine learning for diagnosis and prognosis in dermatology. J Dermatol Treat. (2019) 1–15. doi: 10.1080/09546634.2019.1682500
11. Krittanawong C. The rise of artificial intelligence and the uncertain future for physicians. Eur J Internal Med. (2018) 48:e13–e4. doi: 10.1016/j.ejim.2017.06.017
12. Lallas A, Argenziano G. Artificial intelligence and melanoma diagnosis: ignoring human nature may lead to false predictions. Dermatol Pract Concept. (2018) 8:249–51. doi: 10.5826/dpc.0804a01
13. Yu C, Yang S, Kim W, Jung J, Chung KY, Lee SW, et al. Acral melanoma detection using a convolutional neural network for dermoscopy images. PLoS ONE. (2018) 13:e0193321. doi: 10.1371/journal.pone.0193321
14. Pour MP, Seker H, Ling S. Automated lesion segmentation and dermoscopic feature segmentation for skin cancer analysis. Conf Proc. (2017) 2017:640–3. doi: 10.1109/EMBC.2017.8036906
15. Jafari MH, Samavi S, Karimi N, Soroushmehr SM, Ward K, Najarian K. Automatic detection of melanoma using broad extraction of features from digital images. Conf Proc. (2016) 2016:1357–60. doi: 10.1109/EMBC.2016.7590959
16. Sabouri P, GholamHosseini H, Larsson T, Collins J. A cascade classifier for diagnosis of melanoma in clinical images. Conf Proc. (2014) 2014:6748–51. doi: 10.1109/EMBC.2014.6945177
17. Abedini M, von Cavallar S, Chakravorty R, Davis M, Garnavi R. A cloud-based infrastructure for feedback-driven training and image recognition. Stud Health Technol Inform. (2015) 216:691–5. doi: 10.3233/978-1-61499-564-7-691
18. Sabbaghi S, Aldeen M, Garnavi R. A deep bag-of-features model for the classification of melanomas in dermoscopy images. Conf Proc. (2016) 2016:1369–72. doi: 10.1109/EMBC.2016.7590962
19. 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) 542:115–8. doi: 10.1038/nature21056
20. Bi L, Kim J, Ahn E, Kumar A, Fulham M, Feng D. Dermoscopic image segmentation via multistage fully convolutional networks. IEEE Trans Bio-Med Eng. (2017) 64:2065–74. doi: 10.1109/TBME.2017.2712771
21. Riaz F, Hassan A, Javed MY, Tavares Coimbra M. Detecting melanoma in dermoscopy images using scale adaptive local binary patterns. Conf Proc. (2014) 2014:6758–61. doi: 10.1109/EMBC.2014.6945179
22. Garcia Arroyo JL, Garcia Zapirain B. Detection of pigment network in dermoscopy images using supervised machine learning and structural analysis. Comp Biol Med. (2014) 44:144–57. doi: 10.1016/j.compbiomed.2013.11.002
23. Liu Z, Sun J, Smith L, Smith M, Warr R. Distribution quantification on dermoscopy images for computer-assisted diagnosis of cutaneous melanomas. Med Biol Eng Comp. (2012) 50:503–13. doi: 10.1007/s11517-012-0895-7
24. Jafari MH, Nasr-Esfahani E, Karimi N, Soroushmehr SMR, Samavi S, Najarian K. Extraction of skin lesions from non-dermoscopic images for surgical excision of melanoma. Int J Comp Assist Radiol Surg. (2017) 12:1021–30. doi: 10.1007/s11548-017-1567-8
25. Shimizu K, Iyatomi H, Celebi ME, Norton KA, Tanaka M. Four-class classification of skin lesions with task decomposition strategy. IEEE Trans Bio-Med Eng. (2015) 62:274–83. doi: 10.1109/TBME.2014.2348323
26. Lingala M, Stanley RJ, Rader RK, Hagerty J, Rabinovitz HS, Oliviero M, et al. Fuzzy logic color detection: blue areas in melanoma dermoscopy images. Comput Med Imaging Graph. (2014) 38:403–10. doi: 10.1016/j.compmedimag.2014.03.007
27. Nasr-Esfahani E, Samavi S, Karimi N, Soroushmehr SM, Jafari MH, Ward K, et al. Melanoma detection by analysis of clinical images using convolutional neural network. Conf Proc. (2016) 2016:1373–6. doi: 10.1109/EMBC.2016.7590963
28. Souza S, Abe JM. Nevus and melanoma paraconsistent classification. Studies Health Technol Inform. (2014) 207:244–50. doi: 10.1007/978-3-319-19722-7_9
29. Premaladha J, Ravichandran KS. Novel approaches for diagnosing melanoma skin lesions through supervised and deep learning algorithms. J Med Syst. (2016) 40:96. doi: 10.1007/s10916-016-0460-2
30. Tschandl P, Kittler H, Argenziano G. A pretrained neural network shows similar diagnostic accuracy to medical students in categorizing dermatoscopic images after comparable training conditions. Br J Dermatol. (2017) 177:867–9. doi: 10.1111/bjd.15695
31. Kaur R, Albano PP, Cole JG, Hagerty J, LeAnder RW, Moss RH, et al. Real-time supervised detection of pink areas in dermoscopic images of melanoma: importance of color shades, texture and location. Skin Res Technol. (2015) 21:466–73. doi: 10.1111/srt.12216
32. 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) 78:270–7.e1. doi: 10.1016/j.jaad.2017.08.016
33. Li Y, Shen L. Skin lesion analysis towards melanoma detection using deep learning network. Sensors. (2018) 18:E556. doi: 10.3390/s18020556
34. Afifi S, GholamHosseini H, Sinha R. SVM classifier on chip for melanoma detection. Conf Proc. (2017) 2017:270–4. doi: 10.1109/EMBC.2017.8036814
35. 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) 82:622–7. doi: 10.1016/j.jaad.2019.07.016
36. 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) 138:1529–38. doi: 10.1016/j.jid.2018.01.028
37. Spyridonos P, Gaitanis G, Likas A, Bassukas ID. Automatic discrimination of actinic keratoses from clinical photographs. Comput Biol Med. (2017) 88:50–9. doi: 10.1016/j.compbiomed.2017.07.001
38. Kefel S, Pelin Kefel S, LeAnder RW, Kaur R, Kasmi R, Mishra NK, et al. Adaptable texture-based segmentation by variance and intensity for automatic detection of semitranslucent and pink blush areas in basal cell carcinoma. Skin Res Technol. (2016) 22:412–22. doi: 10.1111/srt.12281
39. Cheng B, Joe Stanley R, Stoecker WV, Stricklin SM, Hinton KA, Nguyen TK, et al. Analysis of clinical and dermoscopic features for basal cell carcinoma neural network classification. Skin Res Technol. (2013) 19:e217–22. doi: 10.1111/j.1600-0846.2012.00630.x
40. Cheng B, Joe Stanley R, Stoecker WV, Osterwise CT, Stricklin SM, Hinton KA, et al. Automatic dirt trail analysis in dermoscopy images. Skin Res Technol. (2013) 19:e20–6. doi: 10.1111/j.1600-0846.2011.00602.x
41. Odeh SM, Baareh AK. A comparison of classification methods as diagnostic system: a case study on skin lesions. Comp Methods Programs Biomed. (2016) 137:311–9. doi: 10.1016/j.cmpb.2016.09.012
42. Chang WY, Huang A, Yang CY, Lee CH, Chen YC, Wu TY, et al. Computer-aided diagnosis of skin lesions using conventional digital photography: a reliability and feasibility study. PLoS ONE. (2013) 8:e76212. doi: 10.1371/journal.pone.0076212
43. Tan E, Lin F, Sheck L, Salmon P, Ng S. A practical decision-tree model to predict complexity of reconstructive surgery after periocular basal cell carcinoma excision. JEADV. (2017) 31:717–23. doi: 10.1111/jdv.14012
44. Guvenc P, LeAnder RW, Kefel S, Stoecker WV, Rader RK, Hinton KA, et al. Sector expansion and elliptical modeling of blue-gray ovoids for basal cell carcinoma discrimination in dermoscopy images. Skin Res Technol. (2013) 19:e532–6. doi: 10.1111/srt.12006
45. Wang L, Pedersen PC, Agu E, Strong DM, Tulu B. Area determination of diabetic foot ulcer images using a cascaded two-stage SVM-based classification. IEEE Trans Bio-Med Eng. (2017) 64:2098–109. doi: 10.1109/TBME.2016.2632522
46. Mukherjee R, Manohar DD, Das DK, Achar A, Mitra A, Chakraborty C. Automated tissue classification framework for reproducible chronic wound assessment. BioMed Res Int. (2014) 2014:851582. doi: 10.1155/2014/851582
47. Manohar Dhane D, Maity M, Mungle T, Bar C, Achar A, Kolekar M, et al. Fuzzy spectral clustering for automated delineation of chronic wound region using digital images. Comp Biol Med. (2017) 89:551–60. doi: 10.1016/j.compbiomed.2017.04.004
48. de Franciscis S, Fregola S, Gallo A, Argiro G, Barbetta A, Buffone G, et al. PredyCLU: a prediction system for chronic leg ulcers based on fuzzy logic; part I - exploring the venous side. Int Wound J. (2016) 13:1349–53. doi: 10.1111/iwj.12529
49. Garcia-Zapirain B, Elmogy M, El-Baz A, Elmaghraby AS. Classification of pressure ulcer tissues with 3D convolutional neural network. Med Biol Eng Comp. (2018) 56:2245–58. doi: 10.1007/s11517-018-1835-y
50. Alderden J, Pepper GA, Wilson A, Whitney JD, Richardson S, Butcher R, et al. Predicting pressure injury in critical care patients: a machine-learning model. Am J Crit Care. (2018) 27:461–8. doi: 10.4037/ajcc2018525
51. Lu J, Kazmierczak E, Manton JH, Sinclair R. Automatic segmentation of scaling in 2-D psoriasis skin images. IEEE Trans Med Imaging. (2013) 32:719–30. doi: 10.1109/TMI.2012.2236349
52. Shrivastava VK, Londhe ND, Sonawane RS, Suri JS. Computer-aided diagnosis of psoriasis skin images with HOS, texture and color features: a first comparative study of its kind. Comp Methods Programs Biomed. (2016) 126:98–109. doi: 10.1016/j.cmpb.2015.11.013
53. Emam SDA, Thomsen SF, Greiner R, Gniadecki R. Predicting the long-term outcomes of biologics in psoriasis patients using machine learning. Br J Dermatol. (2019) doi: 10.1111/bjd.18741. [Epub ahead of print].
54. Min S, Kong HJ, Yoon C, Kim HC, Suh DH. Development and evaluation of an automatic acne lesion detection program using digital image processing. Skin Res Technol. (2013) 19:e423–32. doi: 10.1111/j.1600-0846.2012.00660.x
55. Daliri MR. Feature selection using binary particle swarm optimization and support vector machines for medical diagnosis. Biomed Tech Biomed Eng. (2012) 57:395–402. doi: 10.1515/bmt-2012-0009
56. Begic Fazlic L, Avdagic K, Omanovic S. GA-ANFIS expert system prototype for prediction of dermatological diseases. Studies Health Technol Inform. (2015) 210:622–6. doi: 10.3233/978-1-61499-512-8-622
57. Khan J, Malik AS, Kamel N, Dass SC, Affandi AM. Segmentation of acne lesion using fuzzy C-means technique with intelligent selection of the desired cluster. Conf Proc. (2015) 2015:3077–80. doi: 10.1109/EMBC.2015.7319042
58. Huang ML, Hung YH, Lee WM, Li RK, Jiang BR. SVM-RFE based feature selection and Taguchi parameters optimization for multiclass SVM classifier. Scient World J. (2014) 2014:795624. doi: 10.1155/2014/795624
59. Seite S, Khammari A, Benzaquen M, Moyal D, Dreno B. Development and accuracy of an artificial intelligence algorithm for acne grading from smartphone photographs. Exp Dermatol. (2019) 28:1252–7. doi: 10.1111/exd.14022
60. Burlina P, Billings S, Joshi N, Albayda J. Automated diagnosis of myositis from muscle ultrasound: exploring the use of machine learning and deep learning methods. PLoS ONE. (2017) 12:e0184059. doi: 10.1371/journal.pone.0184059
61. Forreryd A, Zeller KS, Lindberg T, Johansson H, Lindstedt M. From genome-wide arrays to tailor-made biomarker readout - Progress towards routine analysis of skin sensitizing chemicals with GARD. Toxicol vitro. (2016) 37:178–88. doi: 10.1016/j.tiv.2016.09.013
62. Johansson H, Rydnert F, Kuhnl J, Schepky A, Borrebaeck C, Lindstedt M. Genomic allergen rapid detection in-house validation–a proof of concept. Toxicol Sci. (2014) 139:362–70. doi: 10.1093/toxsci/kfu046
63. Strickland J, Zang Q, Paris M, Lehmann DM, Allen D, Choksi N, et al. Multivariate models for prediction of human skin sensitization hazard. J Appl Toxicol. (2017) 37:347–60. doi: 10.1002/jat.3366
64. Tsujita-Inoue K, Hirota M, Ashikaga T, Atobe T, Kouzuki H, Aiba S. Skin sensitization risk assessment model using artificial neural network analysis of data from multiple in vitro assays. Toxicol in vitro. (2014) 28:626–39. doi: 10.1016/j.tiv.2014.01.003
65. Zang Q, Paris M, Lehmann DM, Bell S, Kleinstreuer N, Allen D, et al. Prediction of skin sensitization potency using machine learning approaches. J Appl Toxicol. (2017) 37:792–805. doi: 10.1002/jat.3424
66. Wilm A, Kuhnl J, Kirchmair J. Computational approaches for skin sensitization prediction. Crit Rev Toxicol. (2018) 48:738–60. doi: 10.1080/10408444.2018.1528207
67. Arevalo J, Cruz-Roa A, Arias V, Romero E, Gonzalez FA. An unsupervised feature learning framework for basal cell carcinoma image analysis. Artif Intell Med. (2015) 64:131–45. doi: 10.1016/j.artmed.2015.04.004
68. Cruz-Roa AA, Arevalo Ovalle JE, Madabhushi A, Gonzalez Osorio FA. A deep learning architecture for image representation, visual interpretability and automated basal-cell carcinoma cancer detection. Med Image Comp Comp Assist Interv. (2013) 16(Pt 2):403–10. doi: 10.1007/978-3-642-40763-5_50
69. Masood A, Al-Jumaily A. Differential evolution based advised SVM for histopathalogical image analysis for skin cancer detection. Conf Proc. (2015) 2015:781–4. doi: 10.1109/EMBC.2015.7318478
70. Zhang G, Yin J, Li Z, Su X, Li G, Zhang H. Automated skin biopsy histopathological image annotation using multi-instance representation and learning. BMC Med Genomics. (2013) 6 (Suppl. 3):S10. doi: 10.1186/1755-8794-6-S3-S10
71. Zhang G, Yin J, Su X, Huang Y, Lao Y, Liang Z, et al. Augmenting multi-instance multilabel learning with sparse bayesian models for skin biopsy image analysis. BioMed Res Int. (2014) 2014:305629. doi: 10.1155/2014/305629
72. Olsen TG, Jackson BH, Feeser TA, Kent MN, Moad JC, Krishnamurthy S, et al. Diagnostic performance of deep learning algorithms applied to three common diagnoses in dermatopathology. J Pathol Inform. (2018) 9:32. doi: 10.4103/jpi.jpi_31_18
73. Ghosh D, Ding L, Sivaprasad U, Geh E, Biagini Myers J, Bernstein JA, et al. Multiple transcriptome data analysis reveals biologically relevant atopic dermatitis signature genes and pathways. PLoS ONE. (2015) 10:e0144316. doi: 10.1371/journal.pone.0144316
74. Mays AC, Chou J, Craddock AL, Miller L, Browne JD. Gene variability between perineural-positive and perineural-negative squamous cell skin cancers. Antican Res. (2016) 36:4007–11. doi: 10.4172/2472-128X.1000133
75. Son YA, Yucebas SC. A novel SVM-ID3 hybrid feature selection method to build a disease model for melanoma using integrated genotyping and phenotype data from dbGaP. Studies Health Technol Inform. (2014) 205:501–5. doi: 10.3233/978-1-61499-432-9-501
76. Wei D. A multigene support vector machine predictor for metastasis of cutaneous melanoma. Mol Med Rep. (2018) 17:2907–14. doi: 10.3892/mmr.2017.8219
77. Yang S, Xu J, Zeng X. A six-long non-coding RNA signature predicts prognosis in melanoma patients. Int J Oncol. (2018) 52:1178–88. doi: 10.3892/ijo.2018.4268
78. Reimann E, Lattekivi F, Keermann M, Abram K, Koks S, Kingo K, et al. Multicomponent biomarker approach improves the accuracy of diagnostic biomarkers for psoriasis vulgaris. Acta Derm Venereol. (2019) 99:1258–65. doi: 10.2340/00015555-3337
79. Lefrancois P, Tetzlaff MT, Moreau L, Watters AK, Netchiporouk E, Provost N, et al. TruSeq-based gene expression analysis of formalin-fixed paraffin-embedded (FFPE) cutaneous T-cell lymphoma samples: subgroup analysis results and elucidation of biases from FFPE sample processing on the TruSeq Platform. Front Med. (2017) 4:153. doi: 10.3389/fmed.2017.00153
80. Lefrançois P, Xie P, Wang L, Tetzlaff MT, Moreau L, Watters AK, et al. Gene expression profiling and immune cell-type deconvolution highlight robust disease progression and survival markers in multiple cohorts of CTCL patients. OncoImmunology. (2018) 7:e1467856. doi: 10.1080/2162402X.2018.1467856
81. Litvinov IV, Cordeiro B, Huang Y, Zargham H, Pehr K, Dore MA, et al. Ectopic expression of cancer-testis antigens in cutaneous T-cell lymphoma patients. Clin Cancer Res. (2014) 20:3799–808. doi: 10.1158/1078-0432.CCR-14-0307
82. Litvinov IV, Netchiporouk E, Cordeiro B, Dore MA, Moreau L, Pehr K, et al. The use of transcriptional profiling to improve personalized diagnosis and management of cutaneous T-cell lymphoma (CTCL). Clin Cancer Res. (2015) 21:2820–9. doi: 10.1158/1078-0432.CCR-14-3322
83. Litvinov IV, Tetzlaff MT, Thibault P, Gangar P, Moreau L, Watters AK, et al. Gene expression analysis in Cutaneous T-Cell Lymphomas (CTCL) highlights disease heterogeneity and potential diagnostic and prognostic indicators. Oncoimmunology. (2017) 6:e1306618. doi: 10.1080/2162402X.2017.1306618
84. Patrick MT, Stuart PE, Raja K, Gudjonsson JE, Tejasvi T, Yang J, et al. Genetic signature to provide robust risk assessment of psoriatic arthritis development in psoriasis patients. Nat Commun. (2018) 9:4178. doi: 10.1038/s41467-018-06672-6
85. Beam AL, Kohane IS. Big data and machine learning in health care. JAMA. (2018) 319:1317–8. doi: 10.1001/jama.2017.18391
86. Christodoulou E, Ma J, Collins GS, Steyerberg EW, Verbakel JY, Van Calster B. A systematic review shows no performance benefit of machine learning over logistic regression for clinical prediction models. J Clin Epidemiol. (2019) 110:12–22. doi: 10.1016/j.jclinepi.2019.02.004
87. Du-Harpur X, Watt FM, Luscombe NM, Lynch MD. What is AI? Applications of artificial intelligence to dermatology. Br J Dermatol. (2020) doi: 10.1111/bjd.18880. [Epub ahead of print].
88. Navarrete-Dechent C, Dusza SW, Liopyris K, Marghoob AA, Halpern AC, Marchetti MA. Automated dermatological diagnosis: hype or reality? J Invest Dermatol. (2018) 138:2277–9. doi: 10.1016/j.jid.2018.04.040
89. Winkler JK, Fink C, Toberer F, Enk A, Deinlein T, Hofmann-Wellenhof R, et al. Association between surgical skin markings in dermoscopic images and diagnostic performance of a deep learning convolutional neural network for melanoma recognition. JAMA Dermatol. (2019) 155:1135–41. doi: 10.1001/jamadermatol.2019.1735
90. Narla A, Kuprel B, Sarin K, Novoa R, Ko J. Automated classification of skin lesions: from pixels to practice. J Invest Dermatol. (2018) 138:2108–10. doi: 10.1016/j.jid.2018.06.175
91. Park AJ, Ko JM, Swerlick RA. Crowdsourcing dermatology: dataderm, big data analytics, and machine learning technology. J Am Acad Dermatol. (2018) 78:643–4. doi: 10.1016/j.jaad.2017.08.053
92. Wang F, Kaushal R, Khullar D. Should health care demand interpretable artificial intelligence or accept “black box” medicine? Ann Internal Med. (2019). doi: 10.7326/M19-2548. [Epub ahead of print].
93. Lim BCW, Flaherty G. Artificial intelligence in dermatology: are we there yet? Br J Dermatol. (2019) 181:190–1. doi: 10.1111/bjd.17899
Keywords: artificial intelligence, barriers, contact allergens, dermatology, melanoma, nevi, psoriasis, machine learning
Citation: Gomolin A, Netchiporouk E, Gniadecki R and Litvinov IV (2020) Artificial Intelligence Applications in Dermatology: Where Do We Stand? Front. Med. 7:100. doi: 10.3389/fmed.2020.00100
Received: 10 December 2019; Accepted: 05 March 2020;
Published: 31 March 2020.
Edited by:
H. Peter Soyer, University of Queensland, AustraliaReviewed by:
Katie June Lee, University of Queensland, AustraliaRoberto Novoa, Stanford University, United States
Copyright © 2020 Gomolin, Netchiporouk, Gniadecki and Litvinov. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Robert Gniadecki, r.gniadecki@ualberta.ca; Ivan V. Litvinov, ivan.litvinov@mcgill.ca