Foundations for an artificial intelligence-mediated phototherapy system for psoriasis.

Background Narrowband ultraviolet B therapy (NB-UVB) is an effective treatment for psoriasis, a skin condition that affects 2.8% of the UK population.

Prior to NB-UVB, patients are 'Fitzpatrick skin-typed' to assess skin colour and then undergo minimal erythema dose (MED) testing to various NB-UVB doses (based on Fitzpatrick skin-type) to determine their propensity to sunburn. However, Fitzpatrick skin-typing and MED testing are subjective and often inaccurate.

Objective measurement of skin colour (skin tone) and MED using a spectrophotometer, and concomitantly training artificial intelligence (AI) to quantify these, would significantly improve MED testing thus optimizing NB-UVB doses for treatment of psoriasis.

Currently, NB-UVB is delivered to all the skin, which increases the risk of normal skin sunburning and skin cancer development.

This limits the NB-UVB doses that can be used and necessitates patients attending three times weekly for up to 10 weeks to complete treatment.

An AI-mediated, targeted NB-UVB approach would avoid these risks and allow use of higher NB-UVB doses for the psoriasis, thereby reducing the treatment course to 4 weeks.

I have trained AI to recognise/delineate psoriasis and now need to train AI to recognise/delineate psoriasis as it improves during NB-UVB treatment so that ultimately, a targeted NB-UVB system which recognises/delineates psoriasis at the start and during treatment, can be developed.

Engaging patients/stakeholders in medical research ensures the research is relevant and is considered vital for AI-based healthcare research to ensure uptake of AI technology by patients in the NHS. Aims and objectives To train (and test) AI to determine skin tone objectively and accurately. To train (and test) AI to objectively assess the MED during MED testing.

To train (and test) AI to recognise psoriasis as the psoriasis improves during NB-UVB treatment.

To analyse the views of patients and relevant stakeholders to inform the future development of targeted NB-UVB phototherapy. Methods Artificial neural networks (ANN) form the basis of the deep learning AI in this fellowship.

For determination of skin tone, photographs will be taken of participants alongside corresponding spectrophotometer readings, which will be inputted into an ANN that will be trained and tested for its accuracy.

To assess the MED, an ANN will be provided with photographs of MED test sites and their corresponding spectrophotometer readings to train the ANN to 'learn' how to 'read' the MED, with subsequent testing of its accuracy.

Photographs will be taken of individuals with psoriasis as they progress through their course of NB-UVB, and these images will be inputted into an ANN so that it can recognise/delineate psoriasis as it improves, with subsequent testing of its accuracy at delineating psoriasis and predicting response to therapy.

Semi-structured interviews with patients and relevant stakeholders will be conducted and the results analysed using thematic analysis. Dissemination and impact Presentation of results at local, national and international events. Sharing results with the Psoriasis Association so that they can inform their members.

Liaise with University of Southampton Press Office to inform the public (including people with psoriasis and their relatives). Publication of results in medical/scientific journals.