Using generative artificial intelligence in dermatological research

Any gen AI output needs verification for accuracy

The development and improvement of gen AIs is progressing very rapidly. Quite a few points mentioned in this article might be outdated by the time it is published, with newer models and tools being developed each day. Remember that you can create your own tools using gen AI too!

Gaps in research and literature review

While choosing a research topic, we need to look for research gaps that need to be addressed. This requires two things: looking for broad gaps and going deeper into the same by doing a detailed literature review. Gen AI can help in both these aspects. For example, suppose you are thinking of working on onychomycosis; a simple prompt of “What are the current research gaps in onychomycosis?” can give you broad areas to research. The model automatically classifies it into various aspects like diagnosis, epidemiology, treatment, etc.

https://chatgpt.com/share/676a429e-0c90-800a-9be8-476391d03932

If you want further information regarding any specific areas (for example, if you want to cover paediatric onychomycosis and you need to do a deeper literature review for the same, a simple prompt would be “Can you share some relevant resources regarding the treatment outcomes in paediatric onychomycosis?”)

https://chatgpt.com/share/676a43b5-f80c-800a-9e53-2da9267bdb9e

The response includes resources like journal links. For open-access articles, you can just paste the link and ask the model to summarise the entire article or ask for clarifications related to specific areas.

https://chatgpt.com/share/676a4499-3b50-800a-a81e-99a5704e981b

There are also many other interactive AI-based apps that allow you to upload entire .pdfs. They are capable of summarising specific aspects of the material or the entirety of it, thus saving researchers a lot of time.

There are quite a few powerful tools that can be used to find seeds and connected papers, create collections, and collaborate with your groups as part of the literature review (like https://www.researchrabbit.ai/).

Reference managers like Zotero have also embedded AI in them to make the process simpler and more intuitive.

NotebookLM is another useful tool which acts like a virtual research assistant. Besides helping to summarise data from uploaded material, NotebookLM helps in generating trends, brainstorming to develop connections and new ideas, and also serves to create presentations.

Consensus (https://consensus.app) is a powerful tool for various aspects of research, especially literature review. It can collect and summarize articles relevant to your research. For doctors, it can also serve as an easy go to tool for evidence based summaries for clinical decisions

Scite.ai combines the ease of using large language models (LLMs) with a large database of citations. It is useful in literature review, especially in choosing the best/ high quality citations for your work. Scite.ai can also help in creating blogs/ marketing material for your scientific work.

Formulating a research question

Gen AI can be used for formulating hypotheses as well as research questions. You can also specify the format (for example, in the PICO format).

https://chatgpt.com/share/676a808c-5eec-800a-a8d2-45e41268f0a6

Study design and statistics

It goes without saying that the researcher needs to be well-versed in the basics of study design, and that a biostatistician would be the best person to comment on any aspect of study design, including aspects like sample size and statistical analysis. However, gen AI can give a reasonable outline for the same

As an example for the previous research questions, a simple prompt gives an entire protocol including - sample size, inclusion/exclusion criteria, outcome measure, statistical analysis plan, and even ethical considerations

https://chatgpt.com/share/676a808c-5eec-800a-a8d2-45e41268f0a6

Some GenAI models also have the ability to comment on uploaded Excel data and the possible statistical analysis that can be used for its analysis. Also, software like Excel now has the option to do quick basic statistical analysis using prompts. For example, in a survey on the use of dermoscopy [Figure 1], the prompt was “How many respondents working in medical colleges responded ‘yes’ to the use of dermoscopy” after clicking the ‘analyse data’ tab in Excel

Close

Figure 1:

Automatic analysis in Excel.

Export to PPT

Gen AI can also be used to make the qualitative research process simpler and easier-right from creating questionnaires to detailed thematic analysis.

There are other powerful tools for data analysis, which can do not only statistical analysis, but also data cleaning and extensive visualisation (like https://julius.ai).

Writing up/presentation

Area where Gen AI can be useful include

• Formatting/correcting references. While a dedicated reference manager is best for this and has much more functionality, Gen AI can also be used to shift across reference formats (for example, changing a set of references from the Harvard to Vancouver format or vice versa). Some of the standard reference managers have AI capabilities which make the process of collecting related references, organising, and formatting much easier.

• Paraphrasing - Gen AI can be quite effective in paraphrasing, grammar and spelling checks. As an example, we are showing the paraphrasing of a section of the introduction from this article itself https://chatgpt.com/share/676b93c2-f3f8-800a-95da-b2e11047bb9c

Note that the conversation with the model is iterative, you can keep asking it to refine further till you are satisfied. Gen AI can also be used as a plagiarism checker.

• Writing abstracts/conclusions – Gen AI can effectively use the content of the manuscript to help draft smaller sections like abstracts and conclusions

• Creating PowerPoint presentations – while some Gen AI applications, as of now, cannot create .ppt directly, they can give you an outline for a slide, and there are indirect methods of transforming this content to a presentation that can be further edited and fine-tuned. Microsoft Copilot, however, has been integrated into PowerPoint now. By an iterative process, you can create editable presentation skeletons covering the topic as well as embed them with high-quality graphics.

• Images – Gen AI can be used for creating graphs/ representative images and video animations, as well as editing of clinical or histopathological images (within the limits of ethical editing). It can be especially useful for things like background correction and image denoising [Figures 2-5]. However, it is necessary for authors to explicitly mention if and how exactly AI was used in image editing, as well as keep the original images, with metadata, in hand in case requested by the journal editor.

Close

Figure 2:

Original unedited image.

Export to PPT

Close

Figure 3:

Background removal and image upscaling.

Export to PPT

Close

Figure 4:

Image denoising.

Export to PPT

Close

Figure 5:

Final edited image after background correction upscaling and denoising.

Export to PPT

Gen AI can also be used for creating all kinds of images. This could be useful in many ways, like creating patient awareness material for research patients. Patient image analysis using Gen AI as part of research, while possible, is fraught with concerns regarding data confidentiality.

ChatGPT has also been shown to be useful in the automation of systematic reviews.² There are also many other tools that can be used to simplify the process of systematic reviews and meta-analysis. The King’s College, London has an updated webpage showing the collection of AI tools available for systematic reviews and meta-analysis (https://libguides.kcl.ac.uk/systematicreview/ai)

One of the most commonly used ones of these is https://www.rayyan.ai/

Authorship and citing AI

Regarding the use of AI in manuscript preparation, the International Committee of Medical Journal Editors (ICMJE) categorically mentions that AI models cannot be mentioned as co-author, but need to be mentioned in the acknowledgement section (as well as the covering letter) and also specifically in sections where they was used (For example if used for data analysis section, mention specifically that it was used here)

https://www.icmje.org/recommendations/browse/roles-and-responsibilities/defining-the-role-of-authors-and-contributors.html#four

Many journals have now made it a policy to include an author declaration on whether and how AI was used in a manuscript. Authors are recommended to be as scrupulous as possible in giving a complete and honest declaration. Regarding the citing of AI-generated material, a simple way is to share the conversation link, as we have done in this article (earlier this could be specifically done for ChatGPT using shareGPT, but now with OpenAI, including the link sharing option within the same frame, it has become easier). For other models it is recommended to mention the following for the citation -the prompt used, the specific tool used, date of accessing, developer details, citation details, and the URL, with specific variations as per the citation style of the journal.

https://www.turnitin.com/blog/when-and-how-to-cite- chatgpt-and-ai-in-mla-apa-formats

https://apastyle.apa.org/blog/how-to-cite-chatgpt

Copyright issues

Ownership of any AI-generated content depends on the tool or platform used. It is important to go through the fine print before using any such content. For example, for ChatGPT, the ownership is with the user who used a particular prompt.

Ethics and other aspects of using AI in research

One of the biggest burdens investigators must face while conducting research is the paperwork. Gen AI is helpful in all such secretarial work, right from drafting emails and research scheduling to developing patient education leaflets and consent forms (including cross translation across languages). In fact, most major gen AI platforms now have voice-controlled conversational models, which help to converse with trial patients, overcoming language barriers.³

The focus of AI in dermatology continues to be in diagnostics. In this aspect some limitations continue to exists- including standardization and expansion of quality data-sets (lack of which in turn can lead to misclassification and various biases, specifically in the context of darker skin types)

As far as Gen AI is concerned, the concern of accuracy is still paramount, which is why cross-checking output is very important. The questions that arises is whether this will end up increasing the workload on the researcher, instead of decreasing it!

Ethical issues in the use of AI in research – limits and how to detect unethical use are two areas which will continue to be debated. As of now the key for authors to maintain scientific integrity would be to declare any form of AI use in detail, no matter how small it is.

WHO in its AI guidance document focuses on six core principles – protection of human autonomy, promoting human well-being, ensuring transparency, fostering responsibility, ensuring inclusiveness and equity, and promoting responsive and sustainable AI. It also encourages creating good regulatory frameworks, involvement of all stakeholders and ensuring minimization of risk associated with AI studies.

As far as protocols and reporting of trials in which AI based interventions were used, two relevant guidelines are recommended – the CONSORT-AI (for reporting) and the SPIRIT-AI (for protocols). These are extensions from corresponding guidelines for clinical trials in general, and these were developed because of the perceived need for more rigorous protocol designs and reporting templates to glean meaningful conclusions from trials involving AI intervention.⁴

Another concern is overreliance on gen AI for research purposes might dilute deeper understanding of basic research methodology, especially in early career researchers. Ensuring adqueate training on research methodology before adding on AI tools would help in this aspect.

The future

The limits of gen AI keep getting redefined. The key change would probably be the automation of any task related to research that is mundane. A palpable shift towards virtual clinical trials, especially in Dermatology, Chatbots for data entry, robots for eliciting consent/counselling of trial patients, automated data cleaning/analysis, and automated patient follow-up/appointment scheduling. The possibilities are endless (and many of these are already happening right now). Big data analysis will become much more common and easier. Entire medical record systems could be scanned in seconds and could provide prevalence data/health trends/red flags. In the context of Dermatology, quick analysis of images in research studies – clinical, histopathological, and dermoscopic (and correlation between them) would all gradually become automated.

Gen AI tools now have improved accuracy in terms of search capabilities, and more importantly, the process of how they think/reason is now visible for most tools. This visible reasoning is likely to become more and more refined. Understanding this reasoning process might actually help researchers gain valuable insight in their area of research and help them look at things from a different perspective.

In terms of research on AI in Dermatology, beyond the capabilities of AI, there is a need to shift to a more inclusive database for developing algorithms that would cover all skin types. Improvement in the quality of databases in terms of quantity and quality would be another aspect. Focus on patient-generated image databases, which will have more validity for patient-initiated teledermatology, is another area which needs to be looked into.

The one big question will be – will the dermatologist/dermatology researcher be redundant in the future? As of now, the answer would be a resounding NO. AI is far from perfect, and ultimately, no matter what level of automation we reach, it still requires reconfirmation of data validity. Also, ultimately, accountability is still with the human researcher; AI is just another tool.

Simply put, things will definitely become easier for the researchers who use AI, and they will become more effective and efficient.