Page 31 - ARNM-2-2

Advances in Radiotherapy
& Nuclear Medicine Generative AI enhances radiotherapy

including supervised learning, where algorithms are creation, and data augmentation. It has the potential to
trained on labeled data to classify or predict new data; revolutionize creative processes, automate content creation,
unsupervised learning, which involves identifying hidden and generate realistic simulations for training other AI
structures in unlabeled data; and reinforcement learning, models. In the health sector, this technology holds promise
which focuses on making optimized decisions through for tasks, such as generating synthetic medical images for
interaction with an environment. 10,11 training diagnostic algorithms, creating simulated patient
Deep learning, an advanced form of machine learning, data for testing health-care systems, and generating diverse
utilizes artificial neural networks to model complex medical texts for research and educational purposes.
data representations and discover correlations in large 2.3. Generative CS
12
datasets. Inspired by the human brain, neural networks
consist of interconnected layers of artificial neurons. Generative CS extends the principles of GAI into the realm
13
Deep learning excels in processing high-dimensional data of CS, which involves the use of computational models
across various domains, from signals and texts to images, and techniques to address complex environmental and
video, and audio. 13 sustainability challenges. Generative approaches within
this context focus on creating solutions, scenarios, or
Deep generative models (DGM), fueled by deep data that contribute to sustainable practices and the
learning techniques, have emerged to generate new content conservation of natural resources. 23
based on existing data, opening up new possibilities for AI
applications. These models aim to understand complex CS is described as an interdisciplinary research
7,14
data distributions and produce outputs resembling real- field aiming to develop computational models,
world data. Unlike discriminative models, which focus on methodologies, and tools to facilitate the management
15
the relationship between input features and output labels, of the equilibrium among environmental, economic, and
23
generative models learn data structure and generation societal requirements for sustainable progress. It involves
employing computational technologies and computational
processes, aiming to create new data samples resembling
the underlying class of the training dataset. 14,16 reasoning to advance sustainability objectives and address
the issue of mitigating the adverse environmental effects
One of the key architectures within generative AI is the of computing technologies themselves. Attaining
23
GAN. GANs consist of two neural networks – a generator sustainability targets necessitates addressing various
and a discriminator – which are trained simultaneously interconnected concerns, and it is crucial for researchers
through an adversarial process. The generator creates and practitioners in this domain to thoroughly grasp
18
new data, attempting to mimic real data, while the interactions among diverse systems operating at a
discriminator evaluates the authenticity of the generated societal level concerning sustainability, with a focus on
content. 17,18 This dynamic creates a feedback loop, leading interdisciplinary investigations. Furthermore, CS can only
24
to the generation of increasingly realistic and high-quality be meaningfully explored and applied when recognized
output. 17,18 as an enterprise encompassing both computing and
Other DGMs, including the variational autoencoder, socioeconomic dimensions; it scarcely holds significance
transformer, and latent diffusion model, are sophisticated as a purely technical or mathematical discipline. 24
algorithms within the realm of AI. 19,20 They enable the In the context of GAI and sustainable computing, AI
generation of new content based on existing data, thus models are utilized to enhance radiological diagnosis and
expanding the scope of AI applications. These models therapy while promoting energy efficiency. This approach
20
25
excel at understanding complex data distributions and involves the generation of advanced diagnostic algorithms,
generating outputs that closely resemble real-world the simulation of various therapeutic scenarios, or the
data. Through techniques, such as statistical analysis and creation of synthetic datasets for training models related
20
exploration of latent space, they learn high-dimensional to medical imaging and treatment planning. For example,
probability distributions from training data and produce generative models might be used to improve the accuracy
new samples that approximate the underlying data of radiological diagnoses, simulate the outcomes of
structure. 21,22 Unlike discriminative models, which focus different treatment strategies, or optimize the allocation
on input-output relationships, DGMs aim to understand of computational resources to reduce energy consumption
the inherent structure and generation processes of data, and minimize waste. 25
thus paving the way for innovative AI advancements. 21
The combination of GAI and CS aims to leverage
Generative AI has found applications in various advanced technologies to address complex challenges in
domains, including image synthesis, text generation, artistic environmental conservation and sustainable development,

Volume 2 Issue 2 (2024) 3 doi: 10.36922/arnm.3523