lNTRODUCTlON

Personalized medicine (precision medicine) is a developing medical practice that develops tailor-made approaches for individual patients, leading to increased reliability and a significant impact on preventative, diagnostic, and therapeutic pathways[1]. Artificial intelligence (AI) integration plays a significant role in achieving precision medicine in nuclear medicine[2]. It refers to a field of computer science aimed at imitating the performance of tasks typically requiring human intelligence[3]. Advancements in AI have allowed for precision medicine models to be developed for individual patients (Figure 1, Table 1). The advancements in AI have been in the order of machine learning (ML), deep learning (DL), artificial neural networks (ANNs), convolutional neural networks (CNNs), and generative adversarial networks (GANs)[4,5].

No sooner did it touch him than the Tanuki leapt yelling into the air, and the hare laughed, and ran to tell his friend the peasant what a trick he had played on their enemy

Al MODELS

Machine learning is not a singular algorithm, but a subset of AI. It processes a set of training data and constructs a model that carries the associations among the variables that are relevant to a particular outcome. It usually needs handcrafted features, requiring more human intervention, for data extraction and filtration[2]. There are many ML methods, some of which are supervised learning, unsupervised learning, semi-supervised learning, and reinforcement machine learning[5,6]. DL is a subset of ML, automating many parts of input extraction, enabling less human intervention. In contrast, ML requires more human intervention for data extraction and filtration[2,5,6].

Mrs. Katrinka didn t know what to do at first. But then she had an idea. She called the sanitation3 department in her town to come around and pick up the 60 foot tall crane. If you have an old couch4, an old table, an old refrigerator, or an old washing machine, you can call the sanitation department, and they ll come around and pick it up.

Artificial Neural Networks are a subfield of DL. ANNs are connected nodes with weighted paths. Each node has parent nodes that influence it, an activation function, firing threshold, and an output value. ANNs are analogous to neurons and their intercommunication[4,5].

The sisters called out, You with your two eyes will no doubt succeed! But Little Two-eyes climbed up, and the golden apples did not jump away from her, but behaved quite properly, so that she could pluck them off, one after the other, and brought a whole apron-full down with her

There is no conflict of interest associated with the senior author or other coauthors who contributed their efforts to this manuscript.

Tamam MO performed the majority of the writing, prepared the figures and tables; Tamam MC performed data accusation and writing; Tamam MC provided the input in writing the paper; Tamam MC designed the outline and coordinated the writing of the paper.

34 They rode together through the sweet-scented woods, where the green boughs97 touched their shoulders, and the little birds sang among the fresh leaves

APPLlCATlONS

The integration of AI into clinical practice will transform the medical profession and nuclear medicine imaging in particular. New abilities, such as clinical data science, computer science, and ML will be considered a necessity when AI is applied to medical imaging workflow and protocols. This could provide the opportunity to decrease the error rate of physicians and eventually lead to improved patient management.

For medical imaging planning, AI will automatically check for specific contraindications, such as allergies and drug interference, or eliminate needless repetition of exams by evaluating past examinations before any examination is done on a patient[10,11].

This article is an open-access article that was selected by an in-house editor and fully peer-reviewed by external reviewers. It is distributed in accordance with the Creative Commons Attribution NonCommercial (CC BYNC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is noncommercial. See: https://creativecommons.org/Licenses/by-nc/4.0/

Another hot topic for research is the enhancement of image quality; Hong

[12] improved the picture resolution and noise properties of PET scanners using large pixelated crystals with a deep residual convolutional neural network[12,13]. Kim

[14] demonstrated that Iterative PET reconstruction employing denoising CNNs and local linear fitting enhanced picture quality and robustness to noise-level disparities.

How beautiful it must be to sit under that tree with the student, but that he could not do; he had to content himself with the key-hole and be happy there!There he stood out on the cold landing, the autumn wind blowing through the cracks of the floor

For the interpretation of images, studies on an AI-based triage system for identifying artifacts have been published recently[15]. In the near future, similar systems will be able to detect directly using raw data, such as sinograms, and issue alarms throughout the scanning process, even before reconstruction, so that technicians can adjust or prolong the scheduled scan procedure to accommodate an unexpected discovery[16]. Automated identification of pathologies provides additional intriguing potential in identifying overlooked results and secondary discoveries, saving time and effort[17].

ETHlCAL CONSlDERATlONS, DATA PROTECTlON, REGULATlONS, AND PRlVACY

Despite the improvements that the field of AI brings to nuclear medicine, there are drawbacks. Ethical considerations, data protection, legal regulations, privacy, and education are among these problems. According to Hagendorf, the ethical concerns of AI in healthcare can be summarized in the “fairness, accountability, and transparency paradigm of AI ethics”[18,19]. Moreover, AI requires considerable sensitive data in healthcare, thus standards for data protection and privacy raise issues that must be dealt with. Furthermore, for AI to generalize large numbers, large amounts of data with variability are needed. This raises more questions about consent, data anonymization, and de-identification[19]. There are promising techniques being developed on top of DL algorithms such as federative learning that might mitigate some of these issues[20]. Additionally, traditional regulatory pathways are lagging behind the recent advancements, creating difficulties regarding regulations and laws. Lastly, insufficient education about AI both from patients, physicians, and academia causes mistrust of AI applications in healthcare. Physicians and academia need familiarity with AI and the rudimentary knowledge necessary to provide patients with the necessary information[19].

CONCLUSlON

AI advancements in the last decade have improved AI’s application in medical imaging. The myriad of applications of AI in nuclear medicine includes all steps of a typical medical imaging workflow: planning, image acquisition, and interpretation. In the future, even patient admission and payment could be included[7-9].

FOOTNOTES

Generative Adversarial Networks are made up of two networks, a generator, and a discriminator, that are in a zero-sum game. Generators generate fake input data to minimize the difference between counterfeits and real inputs. The discriminator classifies the real and counterfeit inputs, attempting to maximize efficiency. Over time, the generator will be good at generating input data and the discriminator will be good at classification[5].

Convolutional Neural Networks are made up of convoluting series of pooling layers. CNNs apply a neural-network layer to a part of an image and systematically traverse over the image. CNNs downsample and summarize features by alternating convolutional layers with pooling layers. Their computational requirements are much lower because they operate on a small subset of an image[4,5].

In nuclear medicine, attenuation maps and scatter correction remain relevant topics for image scanning, thus AI research focuses on these topics intensively. Hwang

[10] generated attenuation maps for whole-body positron emission tomography/magnetic resonance imaging (PET/MRI) using a modified U-Net, a specialized convolutional network architecture for biomedical image segmentation. They compared the CT-derived attenuation map to the Dixon-based 4-segment technique[10,11].

There passed through the forest a poor innocent girl; her heartwas pure, and her understanding increased by her faith. Her chiefinheritance had been an old Bible, which she read and valued. From its pages she heard the voice of God speaking to her, and telling her to remember what was said of Joseph s brethren when persons wished to injure her. They imagined evil in their hearts, but God turned it to good. If we suffer wrongfully, if we are misunderstood or despised, we must think of Him who was pure and holy, and who prayed for those who nailed Him to the cross, Father forgive them, for they know not what they do.

Turkey

Muge Oner Tamam 0000-0002-3793-0178; Muhlis Can Tamam 0000-0002-1327-8947.

Liu JH

A

Liu JH

1 National Human Genome Research Institute. Personalized Medicine. Available from: https://www.genome.gov/geneticsglossary/Personalized-Medicine

2 Chartrand G, Cheng PM, Vorontsov E, Drozdzal M, Turcotte S, Pal CJ, Kadoury S, Tang A. Deep Learning: A Primer for Radiologists.

2017; 37: 2113-2131 [PMID: 29131760 DOI: 10.1148/rg.2017170077]

3 Currie G, Rohren E. Intelligent Imaging in Nuclear Medicine: the Principles of Artificial Intelligence, Machine Learning and Deep Learning.

2021; 51: 102-111 [PMID: 33509366 DOI: 10.1053/j.semnuclmed.2020.08.002]

4 Herskovits EH. Artificial intelligence in molecular imaging.

2021; 9: 824 [PMID: 34268437 DOI: 10.21037/atm-20-6191]

5 Castiglioni I, Rundo L, Codari M, Di Leo G, Salvatore C, Interlenghi M, Gallivanone F, Cozzi A, D'Amico NC, Sardanelli F. AI applications to medical images: From machine learning to deep learning.

2021; 83: 9-24 [PMID: 33662856 DOI: 10.1016/j.ejmp.2021.02.006]

6 Machine Learning. Available from: https://www.ibm.com/cloud/Learn/machine-learning

7 Visvikis D, Cheze Le Rest C, Jaouen V, Hatt M. Artificial intelligence, machine (deep) learning and radio(geno)mics: definitions and nuclear medicine imaging applications.

2019; 46: 2630-2637 [PMID: 31280350 DOI: 10.1007/s00259-019-04373-w]

8 Currie GM. Intelligent Imaging: Artificial Intelligence Augmented Nuclear Medicine.

2019; 47: 217-222 [PMID: 31401616 DOI: 10.2967/jnmt.119.232462]

9 Currie G, Hawk KE, Rohren E, Vial A, Klein R. Machine Learning and Deep Learning in Medical Imaging: Intelligent Imaging.

2019; 50: 477-487 [PMID: 31601480 DOI: 10.1016/j.jmir.2019.09.005]

10 Hwang D, Kang SK, Kim KY, Seo S, Paeng JC, Lee DS, Lee JS. Generation of PET Attenuation Map for Whole-Body Time-of-Flight

F-FDG PET/MRI Using a Deep Neural Network Trained with Simultaneously Reconstructed Activity and Attenuation Maps.

2019; 60: 1183-1189 [PMID: 30683763 DOI: 10.2967/jnumed.118.219493]

11 Ronneberger O, Fischer P, Brox T. U-Net: convolutional networks for biomedical image segmentation. In: Navab N, Hornegger J, Wells WM, Frangi AF, eds. mMedical Image Computing and Computer-Assisted Intervention–MICCAI 2015.Cha, Switzerland: Springer International Publishing; 2015:234-241

12 Hong X, Zan Y, Weng F, Tao W, Peng Q, Huang Q. Enhancing the Image Quality

Transferred Deep Residual Learning of Coarse PET Sinograms.

2018; 37: 2322-2332 [PMID: 29993685 DOI: 10.1109/TMI.2018.2830381]

13 Orlhac F, Boughdad S, Philippe C, Stalla-Bourdillon H, Nioche C, Champion L, Soussan M, Frouin F, Frouin V, Buvat I. A Postreconstruction Harmonization Method for Multicenter Radiomic Studies in PET.

2018; 59: 1321-1328 [PMID: 29301932 DOI: 10.2967/jnumed.117.199935]

14 Kim K, Wu D, Gong K, Dutta J, Kim JH, Son YD, Kim HK, El Fakhri G, Li Q. Penalized PET Reconstruction Using Deep Learning Prior and Local Linear Fitting.

2018; 37: 1478-1487 [PMID: 29870375 DOI: 10.1109/TMI.2018.2832613]

15 Li W, Liu H, Cheng F, Li Y, Li S, Yan J. Artificial intelligence applications for oncological positron emission tomography imaging.

2021; 134: 109448 [PMID: 33307463 DOI: 10.1016/j.ejrad.2020.109448]

16 Noortman WA, Vriens D, Mooij CDY, Slump CH, Aarntzen EH, van Berkel A, Timmers HJLM, Bussink J, Meijer TWH, de Geus-Oei LF, van Velden FHP. The Influence of the Exclusion of Central Necrosis on [

F]FDG PET Radiomic Analysis.

2021; 11 [PMID: 34359379 DOI: 10.3390/diagnostics]

17 Nensa F, Demircioglu A, Rischpler C. Artificial Intelligence in Nuclear Medicine.

2019; 60: 29S-37S [PMID: 31481587 DOI: 10.2967/jnumed.118.220590]

18 Hagendorff T. The Ethics of AI Ethics: An Evaluation of Guidelines.

2020; 30: 99-120 [DOI: 10.1007/s11023-020-09517-8]

19 Aktolun C. Artificial intelligence and radiomics in nuclear medicine: potentials and challenges.

2019; 46: 2731-2736 [PMID: 31673788 DOI: 10.1007/s00259-019-04593-0]

20 Gajera J, Knipe H. Federated learning. Reference article, Radiopaedia.org. (accessed on 19 Apr 2022) [DOI: 10.53347/rID-81590]