Perspectives for Generative AI-Assisted Art Therapy for Melanoma Patients

Abstract

Digital technologies are making their mark in medicine, and especially also in art therapy, offering innovative therapeutic interventions for patients, including those with melanoma skin cancer. However, the integration of novel technologies, such as AI-generated art, brings along ethical, psychological, and technical challenges that are viewed differently among therapists. We aim to gauge art therapists’ views on the ethical, application, and challenge facets of utilizing AI-generated art from medical images in therapy. The focus is on assessing its applicability and limitations for melanoma patients. Art therapists were surveyed via a questionnaire focusing on their experience, digital tool familiarity, and views on AI in therapy, encompassing ethics, benefits, challenges, and applicability for melanoma. Art therapists have already implemented digital technologies and acknowledged potential therapeutic benefits of creating personalized artworks with generative artificial intelligence. Attention needs to be given to technological hurdles and the necessity for supplementary interventions. Views on the method’s adaptability varied, underscoring a need for tailored, patient-focused applications. Art therapists are welcoming AI-generated art as a promising creative therapeutic tool and acknowledge potential therapeutic benefits. There are ethical, technical, and psychological challenges that must be addressed for application in therapeutic sessions. Therapists should navigate AI integration with sensitivity, adhering to ethical norms around consent and privacy. Future studies should show the therapeutic benefit in practice with emphasis on equipping therapists to manage the technical complexities effectively. Furthermore, it is important to ensure that patients can influence the AI output, allowing for creative moments in the process.

1. Introduction

The COVID-19 pandemic accelerated the adoption of digital technologies in many areas of medicine including art therapy [1], ushering in an era where AI [2] and virtual reality [3] could play significant roles despite existing biases against machine-generated visualization and art [4].

Melanoma accounts for the most deaths from any skin cancer [5] and develops from the pigment-producing cells known as melanocytes [6]. The incidence has been increasing over the past 30 years [7]. While suffering from physical pain, patients are also vulnerable to mental illnesses, as about 30% of all patients diagnosed with melanoma report psychological issues [8]. Follow-up research indicates that mental health problems can reduce the effectiveness of cancer medication [9,10]. Therefore, it is essential to care for the mental health of the patient during treatment. Even after successful treatment, melanoma survivors face psychological problems, as they may feel anxious or depressed due to fear of living with the disease or cancer recurrence [11]. In conclusion, adequate psychological help, such as art therapy, among others, is necessary to achieve better coping with the illness. Art therapy has already proven effective in various medical treatment approaches and shown to improve the patient’s expression, self-consciousness and resilience to pressure [12,13,14]. Visual art appreciation of famous paintings, such as Starry Night or Sunflowers by Vincent Van Gogh, is a common form of art therapy, and cancer patients feel relaxed in their emotional state through visual art appreciation, potentially leading to more in depth and long-standing healing [15,16].

In June 2014, a team led by Goodfellow introduced an advanced deep learning system called the Generative Adversarial Network (GAN) [17]. After that, other experts began creating different versions of GANs to solve a variety of challenges in many areas. For example, Zhu et al. proposed the CycleGAN model utilizing cycle consistency loss to constrain training and achieve cross-domain image transformation with unpaired datasets [18]. In this work, we propose the transformation of melanoma images into art paintings based on a CycleGAN variant. CycleGANs have been broadly applied in various scientific fields, including liver medical image generation [19], synthetic CT generation from MRI [20], road dataset generation for urban mobility [21], virtual immunohistochemical staining image generation [22], and CT synthesis from MRI for head-and-neck radiation therapy [23]. We aim to provide a tool based on generative artificial intelligence to be used in art therapy with melanoma patients. The tool can convert medical images of melanoma into artworks based on certain themes. In this work, a flower theme was used to show the concept. GANs are ideal for this concept since, following the initial network training, image production is safe, as the potential for glitches, such as the generation of unappealing images, is minimized. Digital art therapy has been shown to be effective in improving mental health and well-being by providing a dynamic and contemporary approach to therapeutic intervention. For instance, digital tools enable therapists to share and create images easily, which can be particularly beneficial in distance therapy or hybrid formats that combine in-person and online sessions [24]. Moreover, digital art therapy can address various therapeutic goals, such as emotional expression, stress reduction, and cognitive engagement. The incorporation of digital media allows for a broader range of expressive possibilities, helping clients to explore and process their emotions in innovative ways [24]. The survey conducted in this work, administered to a broad spectrum of art therapists, was designed to gather insights on the integration of AI-generated art, particularly derived from medical images, into therapeutic practice. It encapsulated themes ranging from ethical considerations, therapeutic applications, and challenges to the adaptability of this approach for patients across various medical conditions, offering a panoramic view of prevailing thoughts in the art therapy community.

2. Methods

2.1. Study Design

The study was designed to gain insights into the perspectives of art therapists regarding the integration of AI-generated art in therapy. A survey, hosted on Google Forms, was developed to address the potential applications and challenges associated with this therapeutic approach. The 56 art therapists who participated were affiliated with associations including the British Association of Art Therapists (BAAT), American Association of Art Therapists (AATA), College for Educational and Clinical Art Therapy (CECAT), and Canadian Art Therapy Association (CATA).

Figure 1 shows the array of GAN-generated transformation options applied to three melanoma skin lesions, as included in the survey.

The full survey comprised of 18 questions can be found in the Appendix A. It touched upon ethical considerations, potential therapeutic applications, challenges, and future prospects of AI-generated art in therapy. Additionally, the survey encompassed demographic queries to gather data on the therapists’ years of experience, their familiarity with digital tools, and their experience working with melanoma or cancer patients.

2.2. Participants

Figure 2 shows the analysis of the respondents’ years of experience in art therapy (left), their duration of implementing digital technologies in therapy (center), and their experience in working with melanoma or cancer patients (right).

The experience of the art therapists ranged from novice practitioners to veterans in the field. A significant majority, comprising approximately 55% of the respondents, possessed a decade of experience in practicing art therapy. This underlines the depth of expertise present in the feedback. Furthermore, there was a diverse representation from those in the early and mid-stages of their careers, with 5% having 0–2 years, 9% with 2–4 years, 13% with 4–8 years, and 18% with 8–10 years of experience, respectively. Such a distribution ensures a balanced composition of the study group.

Fortunately, 46% had engaged with cancer patients in their practice, with a subset of 13% even having specialized experience with melanoma patients, offering valuable insights into the needs and therapeutic approaches for melanoma patients.

The results indicate a distinct divide in the integration of digital tools: 59% of participants have not incorporated digital means in their practice. On the other hand, 41% had employed digital art tools, with 16% utilizing them for 0–2 years, 9% for 2–4 years, 7% for 4–8 years, and a notable 9% had over 8 years of experience, suggesting a significant integration of technology in therapeutic art practices.

2.3. Procedures and Ethics

The study was conducted with adherence to ethical standards to ensure the confidentiality and anonymity of the participants. Each art therapist was contacted via email, with an invitation containing detailed information about the study’s objectives, the voluntary nature of participation, and the assurance of data confidentiality. Therefore, we can ensure that they were fully aware of the study’s purpose and their rights as participants.

The questionnaire, hosted on Google Forms, was designed to be straightforward and respectful of the participants’ time and sensitivity to the subject matter. Participants were free to withdraw from the study at any stage without any data being stored. All data collected were anonymized and stored securely to uphold privacy standards. Identifying information was not collected, and the findings were reported in aggregate to further ensure anonymity.

2.4. Data Analysis and Reflexivity

Quantitative data were statistically analyzed using descriptive statistics to comprehend the trends in the responses. Qualitative comments were thematically categorized and analyzed.

The research team engaged in continuous reflection on their assumptions, beliefs, and values to mitigate bias and enhance the integrity of the study. We acknowledged the potential influences of our technical backgrounds, perspectives, and preconceptions on the research and aimed to maintain an objective stance throughout the study. This reflexive approach ensured that the findings were representative of the participants’ perspectives and not influenced by our biases or preconceived notions. Through this balanced methodology, the analysis aimed to provide a balanced and comprehensive understanding of the opportunities and challenges associated with AI-generated art in the therapeutic context of melanoma patients.

2.5. Basic Cycle-Consistent Generative Adversarial Network (GAN) Principle

The style transfer model used in this work is based on the cycle-consistent GAN framework proposed by Zhu et al. [18]. The structure consists of two GANs, as shown in Figure 3.

$X$ and $Y$ represent images from two different domains. In this work, the source domain $X$ is the melanoma image, and the target domain $Y$ is the flower image. $G$ and $F$ are the two generators. Here, $G$ converts the melanoma image to the flower image, and $F$ is the opposite conversion. In addition, there are two discriminators, $D_X$ and $D_Y$, employed to determine whether an image belongs to a specific domain. In this work, $D_Y$ is utilized to judge whether an image is an image of the flower, whether it is a fake flower image generated from $G$ or a real flower image that is already in the flower dataset $Y$. The real flower images as well as the synthetically generated flower images were judged by $D_Y$. Both generators have the same network structure.

Like cycleGAN, many residual blocks are built in the generator network to avoid degradation in deep networks [25]. Differently from other discriminators, which map the input to a real number as a representation of the probability of belonging to the real image, the patchGAN used in this work maps the input to a matrix of size N × N, i.e., the patch [26]. Each value in the matrix represents the discriminative result of a small receptive field of the image. Finally, the global average pooling layer reduces the large number of parameters [27], making the model more robust and resistant to overfitting.

We adjusted the cycleGAN network structure with the integration of a sub-pixel convolution, an attention mechanism, and a spectral normalization.

The flower-based GAN model utilized in this work is based on the CycleGAN framework, featuring two generators and two discriminators. The generators transform melanoma images into flower images and vice versa. Each generator starts with a convolutional layer (64 channels, 7 × 7 px kernel) followed by two downsampling layers and nine residual blocks to maintain image quality. Upsampling is done using sub-pixel convolution, which offers a larger receptive field and generates detailed, high-resolution images. An attention mechanism is incorporated after downsampling and residual blocks to capture global dependencies. The discriminators use a PatchGAN structure with a receptive field of 70 × 70 px, employing spectral normalization to stabilize training and the discriminator activator function leaky ReLU activation to prevent gradient disappearance [28]. Further details on the model architecture can be found in Jütte et al. [29].

We employ sub-pixel convolution for the upsampling in the generator to generate multiple channels by convolution and then reshape them [30]. Sub-pixel convolution has a large receptive field, providing more contextual information, and can generate more details [31]. This is useful for obtaining high-resolution results.

The attention mechanism [32] is utilized to obtain distant dependencies. In image-related tasks, the distant dependencies are the receptive fields formed by convolutional operations. Wang et al. proposed a self-attention mechanism for the problem in which traditional convolutional operations can only process local information to coordinate the details of the image relatively globally [33]. The softmax operation is performed on the attention matrix, which is derived from the query and the key, and then the resulting attention is applied to the value to obtain the attention feature [32].

It is challenging to train a GAN because, occasionally, the discriminator reaches the optimal state extremely quickly, making it impossible to train the generator more effectively. Spectral normalization can stabilize the training of discriminator networks. It can limit the Lipschitz constant of the discriminator to improve control of the discriminative network while limiting the upper gradient to make it less prone to gradient explosion [34].

We employ three loss functions in this work: adversarial loss, cycle-consistency loss, and identity loss. The purpose of using adversarial loss is to ensure that the image generated by the generator resembles the distribution of real data [17]. As opposed to the original CycleGAN proposed by Zhu et al. which used cross-entropy [18], we apply the least squares loss function, which has better robustness for outliers [35]. Cycle consistency ensures that the results generated by the two generators do not contradict each other. A melanoma image $X$ is converted to a flower image $Y_f$ by generator $G$. Afterwards, $Y_f$ is converted back to a melanoma image $X_r$ by generator $F$. The contents in the two images of $X$ and $X_r$ should ideally be the same. Generator $G$ converts melanoma images to flower images. If we use a flower image $Y$ as input to the generator $G$, the output $Y_f$ should not change. This is the purpose of identity loss.

In this work, we use two public datasets: the SIIM-ISIC Melanoma Classification Challenge 2020 [36] and the Oxford 102 Flower dataset [37]. Considering hardware limitations, we randomly selected 3661 melanoma images and 2079 flower images.

3. Results

In this section, we present the findings from the survey, summarizing the perspectives of art therapists regarding the integration of AI-generated artworks in their practice.

Figure 4 shows the distribution of therapists’ experience with digital technologies for art making, the likelihood of AI-generated art integration, and their art media preferences.

The adoption of digital technologies in art therapy remains varied among practitioners. A significant majority (70%) of the art therapists surveyed never or only rarely incorporate digital tools in their therapeutic processes. Of the participants, 12% frequently utilize digital tools for art making in their practice already. However, no therapist claimed to always use digital art tools.

The responses regarding the incorporation of AI-generated artworks into therapeutic practices indicate a spectrum of readiness and acceptance among art therapists. A small minority (2%) are firmly opposed to the idea, expressing a definite inclination not to integrate such technology into their practice. On the other hand, a substantial portion is open to the possibility, with 21% likely and an additional 16% definitely considering the incorporation of AI-generated art. Interestingly, the largest group, comprising 38% of respondents, occupies a neutral stance, suggesting a level of uncertainty or a wait-and-see approach. This could be attributed to a lack of information and experience on this emerging tool. The 23% who are unlikely to adopt AI artworks underscore existing reservations in a sub-group of the respondents.

The survey highlights the intuitive preference for traditional art media among therapists, with drawing (87%) and painting (73%) leading the choices. Sculpture is preferred by 64%, indicating the value of tactile, three-dimensional creative expression. Digital art media has achieved a significant 25% adoption rate, highlighting the growing interest and gradual integration of technology in art therapy. However, only 9% of respondents choose the medium based on individual patient needs and preferences.

Figure 5 shows the study results regarding the expectation of art therapists for therapeutic benefits for patients viewing the presented AI-generated artworks.

Generally, the majority of participants expect positive changes for melanoma patients with the integration of AI-generated artworks. A notable majority (57%) believe that there are therapeutic benefits to be derived from viewing these AI-generated artworks. When examining the potential impact on a patient’s perception of their diagnosis, the majority expects a positive shift. This indicates optimism among therapists about the role of AI artworks in reshaping a patient’s perspective. In terms of aiding patients in coping with their diagnosis, the majority also feels that these artworks can play a constructive role.

The study shows that the majority of participants (51.8%) believe that all age groups would benefit from artificially generated art in therapy. Therefore, the integration of this technology is not limited by age. However, specific age groups were also highlighted, with adolescents (14–21 years) and young adults (21–40 years) receiving the most affirmation, at 33.9% and 32.1%, respectively. This suggests a particular interest in applying AI-generated art in therapy for these demographics. The age groups 7–14 and 40–60 both had a 25% selection rate, indicating that a significant portion of therapists also see the application value for children and middle-aged adults. The least-selected categories were 0–7 years at 5.4% and above 60 years at 14.3%, indicating a lower, yet existing, interest in these age groups. Only a small portion of participants (8.9%) believed that no age group would benefit.

The participants were asked how they would integrate AI-generated artworks into therapy sessions. The great potential of AI-generated digital artworks within traditional art therapy are illuminated through the broad spectrum of methodologies endorsed by art therapists, as detailed in

Table 1. Distribution of art therapists’ selected possible methods for integrating AI-generated digital artworks into traditional art therapy sessions. An explanation for each answer (e.g., ‘Journaling’) can be found in Appendix A (Q8).

Integration into Art Therapy Session	Counts	Percentage [%]
Introduction & Contextualization	23	41.1
Physical Art Creation	34	60.7
Storytelling & Symbolism	39	69.6
Group Discussions	34	60.7
Projection & Reflection	33	58.9
Journaling	36	64.3
Digital-Physical Fusion	28	50
Feedback & Evolution	27	48.2
Closing Reflection	23	41.1

.

The survey reveals a variety of ways art therapists would integrate AI-generated digital artworks into traditional art therapy sessions. The majority (69.6%) is inclined towards the ‘Storytelling & Symbolism’ approach, where patients are encouraged to construct narratives around the transformation of the melanoma lesion into a flower, facilitating a process of meaning-making and emotional connection to their healing journey. The physical interaction with art is also deemed critical, with ‘Physical Art Creation’ and ‘Digital-Physical Fusion’ chosen by 60.7% and 50% of respondents, respectively, underscoring the importance to maintain a tactile, hands-on experience in therapy. This demonstrates that novel technologies, such as artificially generated art and conventional art therapies are not meant to be used in isolation but can be combined to create successful synergies. Group dynamics and individual reflections are equally valued in the integration process, with 60.7% of participants preferring ‘Group Discussions’ as an option to integrate AI-generated art in therapy sessions. This underscores the significance of collaborative analysis for collective understanding and mutual support. The ‘Projection & Reflection’ and ‘Journaling’ methods, selected by 58.9% and 64.3%, respectively, emphasize the personal, introspective journey enabled by an immediate interaction with the digital artworks. Furthermore, the engagement process is seen as iterative, as indicated by the 48.2% of therapists opting for ‘Feedback & Evolution’, suggesting the adaptability of the therapeutic process based on ongoing patient feedback. The ‘Introduction & Contextualization’ and ‘Closing Reflection’ approaches, each chosen by 41.1% of therapists, frame the therapy session, ensuring a guided entry and reflective conclusion to the emotional exploration.

Art therapists think that the integration of AI-generated artwork yields a spectrum of potential impacts on the patients’ perception of their diagnosis, as shown in

Table 2. Distribution of art therapists’ anticipated impact of AI-generated artworks on melanoma patients’ diagnosis perception. An explanation for each answer (e.g., ‘Cathartic Release’) can be found in the Appendix A (Q10).

Impact on Patients’ Perception of Their Diagnosis	Counts	Percentage [%]
Empowerment	31	55.4
Distress	25	44.6
Cathartic Release	25	44.6
Increased Connection	25	44.6
Aestheticizing Illness	20	35.7
Educational Tool	20	35.7
Therapeutic Potential	35	62.5
Increased Anxiety	14	25
Reframing Perspective	34	60.7
Mixed Emotions	31	55.4
Sense of Ownership	27	48.2
Potential for Misunderstanding	25	44.6

.

The majority (62.5%) of participants highlighted the therapeutic potential of such artworks, referring to their capacity to enable patients to confront, articulate, and cope with emotions associated with their diagnosis. About 60.7% believed that the art could change the patients’ view, focusing on resilience, beauty, or hope. Similarly, 55.4% felt that the AI-generated art might make patients see their illness and perspective in a more positive way.

Nonetheless, art therapists play a crucial role in determining which patients could benefit from incorporating AI-generated art into their treatment. Around 44.6% of respondents pointed out the potential for distress, also shown by the 25% who anticipated a potential rise in anxiety as patients engage with the artistic representation of their illness. Furthermore, 44.6% acknowledged the duality of reactions, where patients might appreciate the artwork’s beauty while also being reminded of their medical condition.

The feedback on the flower-type representation’s psychological significance and therapeutic potential is generally positive. Many participants appreciate its potential advantages, drawing parallels to growth, life, regeneration, and natural beauty. They view the transformation of melanoma into a flower as a positive shift, resonating with themes of rebirth and renewal that flowers embody. Only a few participants suggested being mindful of varying individual reactions, as personal and cultural connections to flowers might differ. Nevertheless, the flower metaphor offers considerable potential as a starting point. In later stages, other metaphors, like fire for anger, could be integrated to better address the patient’s diverse emotional journey.

The therapists were asked to identify the potential challenges they anticipate when introducing artworks, especially those derived from medical images, to patients in a therapeutic setting. The anticipated challenges are shown in

Table 3. Summary of respondents’ anticipated challenges in introducing artworks derived from medical images to patients. An explanation for each answer can be found in the Appendix A (Q13).

Potential Challenges Introducing Artworks to Patients	Counts	Percentage [%]
None	1	1.8
Emotional Triggers	32	57.1
Misunderstanding	21	37.5
Digital Art Acceptance	36	64.3
Technical Barriers	42	75
Integration with Traditional Therapy	24	42.9
Overemphasis on Aesthetics	24	42.9
Differing Cultural Perceptions	34	60.7
Identity and Personal Connection	24	42.9
Ambiguity in Therapeutic Outcomes	27	48.2

.

Emotional responses are a key consideration among the various challenges associated with presenting medical image-based artworks to patients. Yet, the therapeutic value of these artworks also hinges on their ability to evoke emotions. This reflects the understanding among art therapists that AI-generated artworks have the potential to emotionally engage patients in a significant way.

As expected naturally with the implementation of a novel technology, the art therapists expect some technical barriers. This includes a potential lack of necessary technology competence and infrastructure in therapy settings, which need to be minimized by the tool design. About 64% of the participants believe that some patients, based on their initial expectations of art therapy, may not resonate with digital art and might not feel completely connected to the medium. Moreover, differing cultural perceptions of symbolic representations underscore the nuanced approach required to respect individual patients’ backgrounds and interpretations.

The participants were asked to evaluate accompanying therapeutic discussions or interventions when presenting AI-generated artworks to patients. The responses are displayed in

Table 4. Respondents’ perspectives on accompanying therapeutic interventions when presenting AI-generated artworks to patients. An explanation for each answer can be found in the Appendix A (Q14).

Possible Accompanying Therapeutic Interventions	Counts	Percentage [%]
Art History Introduction	6	10.3
Art Interpretation Session	34	58.6
Narrative Therapy Integration	36	62.1
Guided Imagery Meditation	33	56.9
Comparative Analysis	22	37.9
Digital Art Creation Session	35	60.4
Group Sharing Sessions	33	56.9
Feedback Loop	26	44.8
Art and Emotional Journaling	33	56.9
Integrating Traditional Art	37	63.8

.

The survey participants see a range of approaches when introducing AI-created art to patients. The most popular choice was combining AI-generated and traditional art (63.8%). Narrative therapy integration (62.1%) and digital art creation sessions (60.4%) were also highly favored, underscoring the importance of personal storytelling and active patient involvement in the creative process. Art interpretation sessions (58.6%) and guided imagery meditation (56.9%) received significant support, indicating a belief in the utility of aiding patients in artwork interpretation and mental imagery for enhanced emotional well-being. Many respondents highlighted the importance of assessing the patients’ experiences and their ability to connect with the artwork to evaluate the success of using AI-generated art. Key indicators of success included an improvement in emotional well-being, a change in the patients’ approach to their diagnosis, and their ability to gain new insights or perspectives on their illness.

On a more general outlook for technology in art therapy, the therapists were asked about their vision for the evolution of digital tools within the field, anticipating developments that could shape the landscape of art therapy in the next five years beyond the AI-generated artworks presented in this work. The preferences are shown in

Table 5. Preferences for future advancements in digital tools within art therapy. An explanation for each answer can be found in the Appendix A (Q16).

Advancements in Digital Art Therapy Tools (Next Five Years)	Counts	Percentage [%]
None	1	1.8
Virtual Reality Integration	20	35.7
AI-guided Sessions	6	10.7
Digital Art Toolkits	34	60.7
Remote Art Therapy Platforms	39	69.6
Augmented Reality in Art Therapy	11	19.6
Therapeutic Gaming	17	30.4
Digital Art Display and Storage	28	50
Wearable Biofeedback Devices	21	37.5
Multimedia Integration	25	44.6
Online Art Therapy Communities	25	44.6
Training and Workshops	44	78.6
Accessibility Features	33	58.9
Interactive Digital Installations	19	33.9
Digital Storytelling	30	53.6

.

The responses suggest a strong interest in the continued evolution and integration of technology within the field of art therapy over the next five years. Most respondents (78.6%) express a desire for enhanced training and workshops to ensure that art therapists are equipped with the skills and knowledge to utilize emerging digital tools effectively. The findings indicate a strong inclination towards the development of remote art therapy platforms (69.6%), highlighting a trend towards virtual, accessible therapeutic encounters. Additionally, digital art toolkits are favored by 60.7% of respondents, underscoring the perceived benefits of diverse and innovative tools in therapeutic expression and exploration. Moreover, digital storytelling and accessibility features are underscored. Interestingly, a notable portion of respondents is also open to exploring the potentials of virtual reality and AI-guided sessions, pointing towards a future where art therapy is enriched by immersive, personalized, and data-informed digital experiences. Some art therapists (37.5%) even foresee the potential applications of wearable biofeedback devices.

The study design also gave participants the opportunity to provide additional comments. While most respondents express a general excitement about the potential of generative AI, others feel not fully qualified due to limited experience with digital processes. The comments underscore the sensory limitations of digital work and the importance of physical interactions in therapy. Combining technology and therapy is seen as a progressive step and there is a recognition of the utility of online platforms and digital storytelling, especially highlighted during the COVID pandemic, yet in-person interactions offer a richer therapeutic experience. The art therapists appreciate the modern touch, possibly inspiring a novel patient demographic to embrace art therapy, and potential benefits for specific patients, like cancer survivors. Furthermore, there is a belief that this method could aid memory reconsolidation, reducing traumatic responses and enhancing mental well-being. Moreover, AI art making might be particularly engaging for autistic individuals, who sometimes connect better with digital interfaces.

Most responses accept the novel approach but express caution with transforming medical images into art. There is a strong consensus on the necessity of a full understanding of patients about the process and purpose of this transformation. Respondents highlighted potential therapeutic benefits, including narrative reframing and offering new perspectives on illness. However, concerns were raised about the potential for misinterpretation, aestheticization of illness, privacy, and the necessity of ethical guidelines.

In the remainder of this section, we will explore the art therapists’ views on a different technique based on generative AI. Figure 6 presents examples of AI-generated artworks shown in the survey, each crafted via prompt-based image-to-image translation with Runway (2023 Runway AI, Inc., New York, NY, USA). This approach empowers the patient’s creativity by transforming the medical image according to a prompt they craft. For instance, a patient might choose the prompt: “A cinematic bird’s-eye view of a flourishing river delta”, as exemplarily shown in Figure 6 with a prostate cancer scan. In this approach, patients have more agency over their expressions compared to the work introduced above. Figure 6 shows different medical conditions, offering a visual exploration into the adaptability of AI-generated artworks across a range of illnesses.

About half of the respondents envision a potential application of prompt-based image-to-image translations to patients with brain tumors (50.4%) and breast cancer (48.6%), indicating a perceived adaptability of this therapeutic approach. Furthermore, the possible application to lung scans of lung cancer patients is acknowledged by 43.2% of the participating art therapists. However, 14.4% of respondents express uncertainty, marking an area for further exploration. These mixed responses open a dialogue on the universal application of AI-generated art within therapeutic contexts, calling for empirical studies to delineate its boundaries and potentials.

4. Discussion

While the majority of the participating therapists spotlight the innovation’s potential to enrich the therapeutic process, some concerns are raised about the missing physical aspects of art making. However, AI-generated art could assist in art therapy sessions with potential therapeutic benefit, without replacing the traditional sessions. Many therapists express optimism for the application across various types of cancer. The transformation of sensitive medical data into art necessitates a robust framework to safeguard patient dignity, autonomy, and confidentiality. Developing comprehensive guidelines that encompass informed consent protocols, data handling, and sharing practices, and ethical considerations in therapy delivery can help in navigating these challenges. The integration of AI and digital art in therapy is not without its technological hurdles, with challenges such as secure storage, ethical sharing, and intuitive creation interfaces forming critical areas of focus. On the positive side, emerging technologies like AI and Virtual Reality are enabling highly personalized, immersive, and adaptive therapeutic points of contact. The introduction of AI-generated artworks to patients is expected to benefit them but requires thoughtful incorporation of therapeutic discussions or interventions, ensuring a sensitive approach to the unique emotional and psychological dynamics each artwork could evoke. Integrating insights from art therapists, ethical guidelines, and empirical research is vital in developing a dynamic framework that adapts to the evolving nature of AI-generated art in therapy. AI’s role in art therapy is poised to evolve aiming for a balanced integration that upholds the core values of therapy. Future innovations should maintain the sensory and emotional connections intrinsic to traditional art therapy, ensuring that the human touch is preserved amidst the incorporation of digital elements. A possible integration of AI and digital tools in art therapy necessitates a training framework for therapists to ensure that the technology is utilized effectively and ethically. Education should enable therapists to navigate the complex landscapes of digital intervention. Additionally, establishing the infrastructure and support systems is crucial to facilitate the widespread and sustainable incorporation of these innovations, ensuring accessibility, efficiency, and adaptability in diverse therapeutic settings. The infusion of AI-generated art in therapy underscores a pressing need for comprehensive empirical studies to authenticate its efficacy and safety. We identified several key studies that illustrate the potential applications and effectiveness of AI-generated art in art therapy. Lee et al. [38] highlighted the use of AI-generated artwork to reflect emotions, suggesting its value in facilitating creativity and emotional expression in therapy. Yoo et al. [39] developed “Mind Palette”, a mobile app integrating AI and art therapy to foster emotional discussions and self-expression, demonstrating practical application for melanoma patients. Du et al. [40] introduced DeepThInk, an AI-infused art-making system, which, by lowering the expertise threshold for art making, could benefit patients facing physical or emotional barriers. Liu et al. [41] showed how AI can enhance family expressive art therapy by providing diverse materials that facilitate communication. Choe et al. [42] discussed the Expressive Therapies Continuum alongside AI to design interventions addressing the unique creative needs of patients. Finally, Hu et al. [43] examined AI-generated imagery in digital mental health services, emphasizing its potential to aid emotional processing and self-expression.

Incorporating patient involvement in influencing the AI-generated art output can significantly enhance the therapeutic process. Practical strategies for implementing this include developing interactive interfaces that allow patients to input their preferences, facilitating collaborative sessions where patients and therapists co-create AI-generated art, and establishing feedback loops for continuous patient input. This active involvement can lead to increased engagement, a sense of empowerment, and deeper emotional expression. Furthermore, customizing the AI-generated art to reflect individual patient preferences can strengthen the therapeutic alliance and improve overall therapy outcomes. By allowing patients to influence the AI output, therapy becomes more personalized and responsive to their unique needs and experiences.

Future research should focus on diverse methodologies and implementation in sessions to evaluate the therapeutic impact. Therapists should consider the diverse symbolic, emotional, and aesthetic values attached to these artworks across different cultures and individual preferences, employing adaptive strategies to ensure that AI interventions are as resonant and therapeutically effective as possible for each patient.

5. Limitations of the Study

The method of contact so far was via email. It is possible that some art therapists with reservations about the research topic did not participate in the survey after reading the description. Those who responded might be more inclined towards innovative approaches, more receptive to integrating technology in their practice, or simply more proactive in engaging with academic research. Conversely, therapists with reservations or skepticism about the research topic might be underrepresented.

6. Implications for Policy, Practice, and Further Research

The emergence of AI-generated art within therapeutic settings necessitates the establishment of guidelines to address ethical and privacy concerns including informed consent, data protection, and the confidentiality of patients’ medical images and resultant artworks. Moreover, policies should delineate the boundaries between therapeutic and aesthetic objectives, ensuring that the integration of AI does not compromise the core goals of art therapy.

For practice, professionals need to be equipped with appropriate training to effectively incorporate AI-generated artworks in therapy. Specialized training modules can be developed to enhance therapists’ technological capabilities. The inclusion of patients in the creative process is central.

Future adaptations of AI applications in therapy could potentially involve interactive platforms where patients can influence the artistic transformation of their medical images. Offering themes beyond flowers, such as elements representing a broader range of emotions, could enrich the therapeutic experience, making it more responsive to individual emotional landscapes. Further research should explore the therapeutic impacts on patients’ mental and emotional states. Prompt-based image-to-image translations, where specific prompts guide the transformation of medical images to artworks, present a promising direction for exploration. Investigations can focus on how different prompts influence patients’ emotional and psychological responses and how they can be optimized to enhance the therapeutic journey. Experimental studies, incorporating a broader spectrum of diseases and diverse patient demographics, can provide insights into the universal and specific impacts of AI-generated art in therapy.

To effectively integrate AI-generated art into therapy sessions, therapists can employ a variety of strategies that balance digital tools with traditional therapeutic techniques. One approach is to combine AI-generated images with traditional art materials, allowing patients to modify and enhance digital art using paints or pencils, fostering creativity and deeper emotional engagement. Another strategy involves using AI-generated art to visualize abstract concepts, which can then be further explored through traditional methods like sculpting or drawing. Structured sessions that incorporate both AI and traditional techniques can also be beneficial, starting with digital exercises to stimulate discussion followed by hands-on activities. Maintaining a balance between AI and traditional methods, tailoring the approach to individual patient needs, and continuously gathering patient feedback are essential for maximizing therapeutic outcomes.

To measure the therapeutic benefits of AI-generated art, future research could employ a variety of research designs and methodologies. Randomized controlled trials (RCTs) could be conducted to compare the effects of AI-generated art therapy with traditional art therapy or control groups, using standardized assessments such as the Beck Depression Inventory (BDI) [44] and the State-Trait Anxiety Inventory (STAI) [45] to quantify changes in mental health. Qualitative approaches, such as case studies and thematic analysis of patient and therapist interviews, can provide deeper insights into individual experiences and the impacts of AI tools on therapy. These methodologies will help to show the therapeutic potential of AI-generated art and guide its effective implementation in clinical settings.

In addition to CycleGAN, several advanced AI generation technologies, e.g., Video GANs, StyleGAN, DALL-E and CLIP, hold potential for enhancing art therapy. For example, Video GANs (e.g., MoCoGAN [46], VideoGPT [47]) generate coherent video sequences rather than static images, providing a dynamic representation of therapeutic concepts. The ability to generate videos can create more immersive and engaging therapeutic experiences, allowing patients to interact with moving visuals that can better represent emotional journeys or healing processes. However, video generation requires significantly more computational power and data, making it less accessible for many therapeutic settings. Additionally, ensuring the therapeutic relevance and appropriateness of generated content poses challenges.

Furthermore, StyleGAN is known for its ability to generate high-quality, realistic images with fine-grained control over style and content [48]. It offers high flexibility and control over the artistic elements, enabling the creation of highly personalized therapeutic artworks. It can generate diverse styles, which may cater to individual patient preferences more effectively. However, in initial testing, the StyleGAN did not show promising results and was not investigated further. The complexity of StyleGAN may require more advanced technical skills to operate and integrate into therapy sessions. Additionally, the generation process can be slower compared to simpler GAN models.

Finally, DALL-E [49] generates images from textual descriptions, while CLIP [50] can understand and generate images based on natural language inputs. These models enable an intuitive interaction where therapists and patients can describe the desired therapeutic content in natural language, making the process more accessible and engaging. They offer vast creative possibilities, aligning well with narrative therapy approaches. Still, the generated content’s accuracy and relevance depend heavily on the quality of the textual input. Misinterpretations of complex therapeutic concepts can occur, and there may be ethical concerns regarding the representation of sensitive medical images. Future research should aim to evaluate these technologies’ therapeutic impacts, accessibility, and practicality in clinical settings.

7. Conclusions

In this work, we implement the transformation of melanoma images to art images utilizing a GAN which we designed and trained for this purpose. This states, to our knowledge, the first implementation of GANs trying to address patient mental health in melanoma disease management. Furthermore, we also demonstrate a potential application of prompt-based image-to-image translation models in art therapy.

The survey conducted in this work delves into the perspectives of art therapists on the integration of AI-generated art, derived from medical images, into therapeutic practices, particularly for melanoma patients. It explores the potential applications and challenges, and the prospect of extending this innovative approach to other patient demographics and medical conditions, offering a comprehensive insight into this emerging intersection of art, technology, and therapy.

AI-generated art in therapy could help to meet the needs of patients that are unable to perform classical art due to cognitive or physiological restrictions. The process of creating art is considered therapeutic, and AI-assisted art might not fit traditional definitions. While AI can offer innovative means of expression and exploration, the irreplaceable human elements of empathy, understanding, and adaptability must remain at the core of therapeutic practices. The study revealed a diversity of opinions among art therapists regarding the use of AI-generated art in therapy. The participants acknowledged potential therapeutic benefits. Concerns were raised about technological barriers, and the need for complementary therapeutic interventions was stressed. These findings align with existing literature that shows the challenges of integrating technology in therapy. A comprehensive ethical framework is needed to govern the integration of AI-generated art in therapy, ensuring that patient consent and privacy are paramount. Clinically, practitioners are encouraged to implement this innovative approach in accordance with the specific needs of each patient. The responses from therapists prompt a deeper exploration into the psychological impacts and ethical boundaries of incorporating AI-generated art, urging a reevaluation and expansion of existing theories. Future research should explore the incorporation of diverse thematic elements beyond e.g., flowers and include methodologies that enable the creative involvement of patients in the AI-based art creation process. Policy and practice should evolve to establish comprehensive ethical and procedural guidelines, ensuring the sensitive and personalized application of AI in art therapy, while addressing privacy, consent, and the therapeutic integrity of the intervention. To address the ethical and technical challenges associated with AI-generated art therapy, several strategies are recommended. Ensuring patient privacy and consent is paramount; this includes obtaining informed consent through clear communication about the use and implications of AI-generated art, implementing robust data anonymization techniques, and adopting stringent data security measures such as encryption. On the technical side, providing comprehensive training for therapists through online training and hands-on workshops, developing user-friendly AI tools to minimize the technical burden, and establishing a robust technical support system and comprehensive manuals are crucial steps. These strategies aim to empower therapists to effectively integrate AI-generated art into their practice while maintaining high ethical standards and overcoming technical barriers.