Comment on Artificial Intelligence in Trichology

ABSTRACT

Artificial intelligence (AI) is intelligence exhibited by computation, as opposed to natural human intelligence. Founded as an academic discipline in 1956, AI went through cycles of optimism, followed by periods of disappointment. The growing use of AI in our century is influencing a shift toward increased automation, data-driven decision-making, and the integration of AI systems into various areas of life, including health care. On the occasion of the European Academy of Dermatology and Venereology Spring Symposium 2024, reference was made to AI in dermatology, specifically trichology. Particularly in the domain of trichoscopy, AI’s contribution to automated image analysis stands out. Due to the fairly standardized imaging and a limited amount of diagnoses, trichoscopic images have become a center of interest for automated medical image analysis and its commercial utilization. Yet, costly, multi-parameter computer-assisted trichoscopic image analysis is the least effective in medical practice. Using dermoscopy, signature patterns are seen in a range of scalp and hair conditions that have provided the foundation for diagnostic algorithms. Algorithms discourage physicians from thinking independently and with creativity. Doctors are trained to romance with technology. However, healing is replaced with treating, caring is supplanted by managing, and the art of listening is taken over by technology. The power of technology, particularly computer-based, may shake the confidence of a specialist in his initial diagnosis. Nevertheless, machines cannot replace the doctor’s mind, and his thinking about what he sees and what he does not see. Ultimately, common AI mistakes are often not too different from mistakes born out of natural human error. Taking all of this into account, it does not seem wise to precociously hail an evolving technology in public without the respective experimentation and experience to back up its factual utility in our practice.

I think of AI itself as a monster of capitalism.

-Trevor Paglen

Artificial intelligence (AI) is intelligence exhibited by computation, as opposed to natural human intelligence. It has become a field of research in computer science that develops and studies methods and software which enable machines to perceive their environment and use learning and intelligence to take actions that maximize their chances of achieving defined goals.

AI was founded as an academic discipline in 1956 and went through cycles of optimism, followed by periods of disappointment. Interest resurged with the development of deep learning based on artificial neural networks with representation learning, and machine learning-based attention which intuitively mimics cognitive attention. This led to the AI boom of the early 2020s, with companies, universities, and laboratories pioneering the advances in AI worldwide with the lead of the USA. The growing use of AI in our century is influencing a shift toward increased automation, data-driven decision-making, and the integration of AI systems into various economic and social areas of life, impacting job markets, government, industry, healthcare, and education.

On the occasion of the European Academy of Dermatology and Venereology Spring Symposium 2024, reference was made to AI in dermatology, specifically in trichology.[1] According to the speaker, the promising implications of AI in trichology for the future include enhanced automation, faster diagnosis, and improved patient outcomes. Particularly in the domain of trichoscopy, AI’s contribution to automated image analysis stands out. Trichoscopic images, capturing intricate details of the scalp and hair, are swiftly processed and interpreted by AI algorithms. This precision allegedly surpasses traditional diagnostic methods, enabling AI to recognize a broad spectrum of trichoscopic patterns associated with specific conditions of alopecia.

Trichoscopy is the term coined for the use of a dermatoscope for the evaluation of hair and scalp. Trichoscopy has gained popularity as a tool in the differential diagnosis of hair and scalp disorders. However, as a diagnostic procedure, trichoscopy remains to be understood as representing an integral part of a more comprehensive dermatological examination.[2,3] As with any medical problem, a patient complaining of hair loss requires a comprehensive medical history, physical examination, and appropriate laboratory evaluation to identify the cause. By approaching the hair loss patient in a methodical way, commencing with the simplest and easiest to recognize objects, and ascending step by step to the knowledge of the more complex, a final diagnosis can be made as prerequisite to a treatment that is appropriate for that specific condition.[4] Listening attentively is the most powerful diagnostic procedure in the diagnosis of hair growth disorders: a doctor who takes a careful history reaches a correct diagnosis in 70% of cases. This is far more efficient than all currently available tests and technologies.[5] Touching is the oldest and most effective tool in doctoring:[5] The relationship with the patient often alters dramatically after the physical examination (scalp, complete skin, nails, mucous membranes, assessment of hair part width, hair pull, and hair feathering test). Today, doctors are trained to romance with technology. Healing is replaced with treating, caring is supplanted by managing, and the art of listening is taken over by technology.[5]

Using dermoscopy, signature patterns are seen in a range of scalp and hair conditions that have provided the foundation for diagnostic algorithms. Algorithms may be useful for the average diagnosis and treatment, but they fail when doctors need to think outside of their boxes, when symptoms are vague, multiple, or confusing, and when test results are inconclusive. Algorithms discourage physicians from thinking independently and with creativity. Doctors who turn down their thinking on the authority of algorithms have a statistical way of looking at people.[6] However, statistics embody averages, not individuals. Numbers can only complement a physician’s personal experience. There is never certainty as to where the individual fits on the normal statistical distribution curve. Statistics shroud souls and obscure individuality.[5] Moreover, lateral thinking that breaks out of the ordinary is sometimes vital. Creativity and imagination, rather than adherence to the obvious, are needed in situations where the data and clinical findings do not all fit neatly together.[6]

AI enhances the quantification by automated measurements of parameters such as hair density and shaft diameter, providing objective and standardized assessments for accurate diagnoses in a facilitated manner, and enabling dermatologists to monitor changes over time, assess intervention efficacy, and track treatment progress in alopecia. Due to the fairly standardized imaging and limited amount of diagnoses compared to clinical dermatology, trichoscopic images have indeed become a center of interest for automated medical image analysis and its commercial utilization. Yet, costly, multi-parameter computer-assisted trichoscopic image analysis is the least effective in clinical practice.[3]

Indeed, measurement of the effects of treatment needs to be quantified reliably. The method should be more capable of analyzing relevant parameters of hair growth, which are: hair density, hair diameter, hair growth rate, and anagen/telogen ratio. For this purpose, computer-assisted image analysis has been proposed; some patents have been filed and publications followed since the 1980s. However, it soon became clear that hair is a tricky material for automated computer-assisted image analysis, and that numbers might not all be considered as reflecting hair measurements. The physical properties of hair, that is, the object, the variability of the skin, and their background are very complex. The multilayered fiber is composed of a nonpigmented cuticle, a cortex with the presence or absence of pigment granules, and a medulla filled with proteinaceous material or air cavities. On top, its organization and orientation at the exit point from the skin must also be taken into account. A follicular unit comprising a number of hair follicles (occasionally up to 5) may exit from a single orifice at the skin surface and it may be difficult to count individual hair fibers. Some attempts have suggested that the use of fully automatic systems may be an option but this has not been made available to the public. Computerized methods require further optimization. Ease of use and fast image processing, as pointed out by others are certainly appreciated. Nevertheless, although speed is considered smart in our culture, customers, that is, clinicians, patients, and pharmaceutical or cosmetic companies, deserve the highest standard and a better service than merely a fast one. All should be given the best possible and clinically most relevant information about hair measurements – both qualitatively and quantitatively – that have diagnostic, prognostic, and therapeutic relevance.[7]

While AI may perform some tasks with remarkable precision, it cannot fully replicate human intelligence and creativity. AI lacks consciousness and emotions, limiting its ability to understand complex human experiences and totally lacking empathy. Computers are incapable of the intuitive reasoning physicians employ when evaluating information from multiple sources to make a diagnosis and create a treatment plan. The power of technology, particularly computer-based, may shake the confidence of a specialist in his initial diagnosis. Nevertheless, machines cannot replace the doctor’s mind, and his thinking about what he sees and what he does not see.

The top-ranking challenges of AI are:

• Lack of quality medical data. Clinicians require high-quality datasets for the clinical and technical validation of AI models

• Clinically irrelevant performance metrics

• Methodological research flaws.

The drawbacks of AI are:

• Lack of creativity

• Absence of empathy

• Possible overreliance on technology and skill loss in humans

• Risk of commercial abuse.

Ultimately, common AI mistakes in health care are often not too different from mistakes born out of natural human error.

Upon testing AI with a photograph of a Xolo pug [crossbreed between the Mexican hairless dog and the pug: Figure 1], AI identified the pooch as a Neapolitan Mastiff [Figure 2], a massive, powerful guarder with profuse hanging wrinkles and folds, and pendulous lips, whose astounding appearance has intimidated intruders since the days of ancient Rome.

Figure 1.

Xolo pug (crossbreed between Mexican hairless dog Xoloitzcuintle and pug)

Figure 2.

Neapolitan mastiff

Taking all of this into account, it does not seem wise to precociously hail an evolving technology in public without the respective experimentation and experience to back up its factual utility in our practice.

Despite what pretentious publications would suggest us to believe, effective trichological practice is not a closed art, to be mastered only after years of perfecting and struggling with the mysteries of hair growth and shedding. In every art, there are many techniques, but few principles. The only way to achieve success is to have a firm foundation of principles to build upon and the right attitude about achieving one’s goals.

We have recently proposed the term trichiatry to describe physicians trained in the evaluation and management of patients with hair loss.[8] The quality and stringency of the trichiatrist’s graduate medical training are identical to that of fellow physicians of any other discipline, allowing the trichiatrist to be comprehensive in counseling patients, prescribing medication, conducting physical examinations, ordering laboratory tests, and participating with the other medical disciplines in the diagnosis and treatment of hair problems as they may relate to systemic disease. Prerequisites for successful management of hair loss are twofold: on the technical and the psychological level.

On the technical level, prerequisites are a specific diagnosis, a profound understanding of the underlying pathophysiology and its potentially multitudinous causes, and the best available evidence gained from the scientific method for clinical decision-making. Evidence-based medicine (EBM) seeks to assess the strength of the evidence of risks and benefits of diagnostic tests and treatments, using techniques from science, engineering, and statistics. EBM aims at the idea that healthcare professionals should make conscientious, explicit, and judicious use of the best available evidence gained from the scientific method to clinical decision-making. However, the limited success rate of evidence-based therapies in managing hair loss suggests that the complex nature of hair loss may be inadequately served by the present levels of evidence for diagnostics and therapies. It lies in the intelligence, experience, intuition, and creativity of the physician and beyond pure technology, such as AI, to design the proper management plan for the individual patient at hand.

On the psychological level, one must be sure that the patient’s key concerns have been directly and specifically solicited and addressed: Acknowledge the patient’s perspective on the hair loss problem, explore the patient’s expectations from treatment, and educate patients into the basics of the hair cycle, and why patience is required for effective cosmetic recovery. Finally, patients do not understand the use of AI in medicine, and three out of four patients do not trust AI in a healthcare setting.

Conflicts of interest

There are no conflicts of interest.

Funding Statement

Nil.