Добавил:
Sekretar
kiopkiopkiop18@yandex.ru
t.me/Prokururor I Вовсе не секретарь, но почту проверяю
Опубликованный материал нарушает ваши авторские права? Сообщите нам.
Вуз:
Предмет:
Файл:Ординатура / Хирургия / Библиотека им академика М.И. Перельмана / Книга_5545_Библиотеки_им_академика_М_И_Перельмана.pdf
X
- •Preface
- •Contents
- •1: Structure of Matter
- •2: Radioactive Decay
- •2.1 Spontaneous Fission
- •1.1.1 Radiation
- •1.2 The Atom
- •1.2.3 Nuclear Binding Energy
- •1.3 Nuclear Nomenclature
- •1.5 Questions
- •Suggested Readings
- •2.2 Isomeric Transition
- •2.2.1 Gamma (γ)-Ray Emission
- •2.2.2 Internal Conversion
- •2.2.2.1 Problem 2.1
- •2.2.2.2 Answer
- •2.3 Alpha (α)-Decay
- •2.4 Beta (β−)-Decay
- •2.5 Positron (β+)-Decay
- •2.6 Electron Capture
- •2.7 Questions
- •Suggested Readings
- •3.1 Radioactive Decay Equation
- •3.1.1 General Equation
- •3.1.2 Half-Life
- •3.1.3 Mean Life
- •3.1.4 Effective Half-Life
- •3.2 Units of Radioactivity
- •3.3 Specific Activity
- •3.4 Calculation
- •3.5 Successive Decay Equations
- •3.5.1 General Equation
- •3.5.2 Transient Equilibrium
- •3.5.3 Secular Equilibrium
- •3.6 Questions
- •Suggested Readings
- •4.5 Poisson Distribution
- •4.6 Gaussian Distribution
- •4.7 Chi-Square Test
- •4.8 Minimum Detectable Activity
- •4.10 Questions
- •Suggested Readings
- •5.1 Cyclotron-Produced Radionuclides
- •5.2 Reactor-Produced Radionuclides
- •5.2.1 Fission or (n, f) Reaction
- •5.2.2 Neutron Capture or (n, γ) Reaction
- •5.6 Radionuclide Generators
- •5.8 Questions
- •Suggested Readings
- •6.1.1 Specific Ionization
- •6.1.2 Linear Energy Transfer
- •6.1.3 Range
- •6.1.4 Bremsstrahlung
- •6.1.5 Positron Annihilation
- •6.2.1.1 Photoelectric Effect
- •6.2.1.2 Compton Scattering
- •6.2.1.3 Pair Production
- •6.2.1.4 Raleigh Scattering
- •6.2.1.5 Photodisintegration
- •6.3.2 Half-Value Layer
- •6.5 Questions
- •Suggested Readings
- •7: Gas-Filled Detector
- •7.1 Principles of Gas-Filled Detector
- •7.2 Ionization Chamber
- •7.2.1 Ion Chamber Survey Meter
- •7.2.2 Dose Calibrator
- •7.2.2.1 Constancy
- •7.2.2.2 Accuracy
- •7.2.2.3 Linearity
- •7.2.2.4 Geometry
- •7.2.3 Pocket Dosimeter
- •7.3 Proportional Counter
- •7.4 Geiger–Müller Counter
- •7.5 Questions
- •Suggested Readings
- •8.1 Scintillation Counter
- •8.4.3 Characteristic X-Ray Peak
- •8.4.4 Backscatter Peak
- •8.4.5 Iodine Escape Peak
- •8.2 Solid Scintillation Detector
- •8.2.1 NaI (Tl) Detector
- •8.2.2 Bismuth Germanate Detector
- •8.2.3 Barium Fluoride Detector
- •8.2.4 Lutetium Oxyorthosilicate Detector
- •8.2.5 Gadolinium Oxyorthosilicate Detector
- •8.2.6 Yttrium Oxyorthosilicate Detector
- •8.2.7 Yttrium Aluminum Perovskite Detector
- •8.2.8 Lutetium Yttrium Oxyorthosilicate Detector
- •8.2.9 Lanthanum Bromide Detector
- •8.3 Solid-State Detector
- •8.3.2 Cadmium–Zinc–Tellurium Detector
- •8.3.3 Cesium Iodide (CsI(Tl)) Detector
- •8.3.4 Solid Scintillation Counter
- •8.3.4.1 NaI(Tl) Detector
- •8.3.4.2 Photomultiplier Tube
- •8.3.4.3 Preamplifier
- •8.3.4.4 Linear Amplifier
- •8.3.4.5 Pulse-Height Analyzer
- •8.3.4.6 Display or Storage
- •8.4 Gamma-Ray Spectrometry
- •8.4.1 Photopeak
- •8.4.6 Positron Annihilation Peak
- •8.4.7 Coincidence Peak
- •8.5 Liquid Scintillation Counter
- •8.5.1 Quenching
- •8.6.1 Energy Resolution
- •8.6.2 Detection Efficiency
- •8.6.2.1 Intrinsic Efficiency
- •8.6.2.2 Photopeak Efficiency or Photofraction
- •8.6.2.3 Geometric Efficiency
- •8.6.3 Dead Time
- •8.7 Gamma Well Counter
- •8.8 Thyroid Probe
- •8.8.1 Thyroid Uptake Measurement
- •8.9 Questions
- •Suggested Readings
- •9: Gamma Camera
- •9.1 Gamma Camera
- •9.1.2 Detector
- •9.1.3 Collimator
- •9.1.4 Photomultiplier Tube
- •9.1.5 X-, Y-Positioning Circuit
- •9.1.6 Pulse-Height Analyzer
- •9.2 Digital Camera
- •9.2.1 Solid State Digital Camera
- •9.3 Questions
- •Suggested Readings
- •10.1.1 Spatial Resolution
- •10.1.1.1 Intrinsic Resolution
- •10.1.1.2 Collimator Resolution
- •10.1.1.3 Scatter Resolution
- •10.1.2.1 Bar Phantom
- •10.1.2.2 Line-Spread Function
- •10.1.2.3 Modulation Transfer Function
- •10.1.3 Sensitivity
- •10.1.3.1 Collimator Efficiency
- •10.1.4 Uniformity
- •10.1.5 Pulse-Height Variation
- •10.1.6 Nonlinearity
- •10.1.7 Edge Packing
- •10.2 Gamma Camera Tuning
- •10.4 Contrast
- •10.4.1 Count Density
- •10.4.2 Image Noise
- •10.4.4 High Count Rate
- •10.4.6 Patient Motion
- •10.5.1 Daily Checks
- •10.5.1.2 Uniformity
- •10.5.2 Weekly Checks
- •10.5.3 Monthly Checks
- •10.5.3.1 High-Count Uniformity Calibration
- •10.5.3.2 Collimator Integrity
- •10.5.4 Annual, Semiannual, or As-Needed Checks
- •10.6 Questions
- •References and Suggested Readings
- •11.1.1 Central Processing Unit
- •11.1.2 Computer Memory
- •11.1.3 External Storage Device
- •11.1.4 Input/Output Device
- •11.1.7 Digital-to-Analog Conversion
- •11.1.8 Digital Image
- •11.2.1 Digital Data Acquisition
- •11.2.2 Static Study
- •11.2.3 Dynamic Study
- •11.2.4 Gated Study
- •11.2.7 Display
- •11.3.1 PACS
- •11.4 Questions
- •Suggested Readings
- •12: Single Photon Emission Computed Tomography
- •12.1 Tomographic Imaging
- •12.2 Single Photon Emission Computed Tomography
- •12.2.1 Data Acquisition
- •12.2.2 Image Reconstruction
- •12.2.2.1 Simple Backprojection
- •12.2.2.2 Filtered Backprojection
- •12.2.2.3 The Convolution Method
- •12.2.2.4 The Fourier Method
- •12.2.2.6 Iterative Reconstruction
- •12.3 SPECT/CT Scanner
- •12.4 Factors Affecting SPECT
- •12.4.1 Photon Attenuation
- •12.4.2 Attenuation Correction Methods
- •12.5 Partial-Volume Effect
- •12.5.2 Sampling
- •12.5.3 Scattering
- •12.6.1 Spatial Resolution
- •12.6.2 Sensitivity
- •12.6.3 Other Parameters
- •12.7.1 Daily Tests
- •12.7.2 Weekly Tests
- •12.7.2.1 Spatial Resolution
- •12.9 Questions
- •References and Suggested Readings
- •13: Positron Emission Tomography
- •13.1 Introduction
- •13.2 PET Radiopharmaceuticals
- •13.3.2 Block Detector
- •13.5 Coincidence Timing Window
- •13.6 PET/CT Scanner
- •13.7 PET/MR Scanner
- •13.7.2 MR Scanner
- •13.7.3 Commercial PET/MR Scanner
- •13.8 Mobile PET or PET/CT Scanner
- •13.9 Micro-PET Scanner
- •13.11 Data Acquisition
- •13.12 Image Reconstruction
- •13.13 Factors Affecting PET
- •13.13.1 Normalization
- •13.13.2 Photon Attenuation Correction
- •13.13.4 Random Coincidences
- •13.13.5 Scatter Coincidences
- •13.13.6 Dead Time
- •13.13.7 Radial Elongation
- •13.14.1 Spatial Resolution
- •13.14.2 Sensitivity
- •13.14.2.1 Noise Equivalent Count Rate
- •13.15.1 Daily Tests
- •13.15.1.1 Sinogram Check
- •13.15.2 Weekly Tests
- •13.15.2.1 Normalization
- •13.18 Questions
- •References and Suggested Reading
- •14.1 Background
- •14.5 Artificial Neural Network
- •14.7 Machine Learning
- •14.7.1 Decision Tree
- •14.7.2 Random Forest
- •14.7.3 Support Vector Machine
- •14.7.4 Computer Vision
- •14.8 Deep Learning
- •14.8.1 Convolutional Network
- •14.8.2 Recurrent Neural Network
- •14.8.3 Generative Adversarial Network
- •14.8.4 Transfer Learning
- •14.9 Radiomics
- •14.10 Natural Language Processing
- •14.11 Large Language Model
- •14.12 Generative Artificial Intelligence
- •14.13.1 Prompt
- •14.13.2 Token
- •14.13.3 Hallucination
- •14.13.4 Deepfake
- •14.13.5 Overfitting
- •14.15 Chatbot
- •14.18 Legal Implication
- •14.20 Questions
- •References
- •15.1 Introduction
- •15.2.1 Scheduling
- •15.2.2 Image Acquisition
- •15.2.3 Image Processing
- •15.2.4 Interpretation
- •15.2.5 Reporting
- •15.3.1 Oncology
- •15.3.2 Cardiovascular Disease
- •15.3.3 Bone Scintigraphy
- •15.3.4 Thyroid Imaging
- •15.5 Drug Development
- •15.6 Questions
- •References and Suggested Reading
- •16: Internal Radiation Dosimetry
- •16.1 Radiation Unit
- •16.1.1 Roentgen
- •16.1.2 Rad
- •16.1.3 Gray
- •16.1.4 Rem
- •16.1.5 Radiation Weighting Factor
- •16.1.6 Quality Factor
- •16.1.7 Sievert
- •16.2 Dose Calculation
- •16.2.1 Radiation Dose Rate
- •16.2.2 Cumulative Radiation Dose
- •16.2.3 Factors Affecting Ã
- •16.2.4 The S Values
- •16.4 Pediatric Dosage
- •16.5 Questions
- •References and Suggested Readings
- •17: Radiation Biology
- •17.1 The Cell
- •17.2.1 DNA Molecule
- •17.2.2 Chromosome
- •17.5 Cell Survival Curves
- •17.6 Factors Affecting Radiosensitivity
- •17.6.1 Dose Rate
- •17.6.2 Linear Energy Transfer
- •17.6.4 Chemicals
- •17.7 Radiosensitizer
- •17.7.1 Oxygen
- •17.7.2 Pyrimidine
- •17.7.3 Others
- •17.8 Radioprotector
- •17.9 Apoptosis
- •17.13.1 Hematopoietic Syndrome
- •17.13.2 Gastrointestinal Syndrome
- •17.13.3 Cerebrovascular Syndrome
- •17.14.1 Somatic Effects
- •17.14.1.1 Carcinogenesis
- •17.14.1.3 Dose–Response Relationship
- •17.14.1.5 Leukemia
- •17.14.1.6 Breast Cancer
- •17.14.1.7 Other Cancers
- •17.14.1.10 Nonspecific Life-Shortening
- •17.14.1.11 Cataractogenesis
- •17.14.2 Genetic Effects
- •17.14.2.1 Spontaneous Mutation
- •17.14.2.2 Doubling Dose
- •17.14.2.3 Genetically Significant Dose
- •17.17 Questions
- •References and Suggested Readings
- •18.1 Introduction
- •18.2 Radiation Protection
- •18.2.3 Occupational Dose Limits
- •18.2.4 ALARA Program
- •18.2.5.1 Time
- •18.2.5.2 Distance
- •18.2.5.3 Shielding
- •18.2.5.4 Activity
- •18.2.6 Personnel Monitoring
- •18.2.6.1 Film Badge
- •18.2.6.2 Thermoluminescent Dosimeter
- •18.2.6.3 Optically Stimulated Luminescence Dosimeter
- •18.3 Radiation Regulations
- •18.3.1 License
- •18.3.1.1 General License
- •18.3.1.2 Specific License of Limited Scope
- •18.3.1.3 Specific Licenses of Broad Scope
- •18.3.2 Radiation Safety Committee
- •18.3.3 Radiation Safety Officer
- •18.3.4.3 Supervision
- •18.3.4.4 Mobile Nuclear Medicine Service
- •18.3.4.5 Written Directives
- •18.4 Bioassay
- •18.6 Radioactive Waste Disposal
- •18.6.2 Release into Sewerage Systems
- •18.6.4 Other Disposal Methods
- •18.7 Radioactive Spill
- •18.8 Recordkeeping
- •18.10 Dirty Bombs
- •18.11 Types of Accidental Radiation Exposure
- •18.12 Protective Measures in Case of Explosion of a Dirty Bomb
- •18.13 Verification Card for Radioactive Patients
- •18.14 Radiation Phobia
- •18.15 European Regulations Governing Radiation
- •18.16 Questions
- •References and Suggested Readings
- •Index

Application ofArtificial
Intelligence inNuclear Medicine
15.1 Introduction
Articial intelligence (AI) is applied to many sectors of enterprises, namely business, healthcare, nance, education, marketing, etc., for efcient and time-saving
operations. In the healthcare sector, nuclear medicine is a small entity compared to
radiology, orthopedics, oncology, etc., because of the difference in the number of
patient visits in these departments. However, the application of AI in nuclear medicine has been progressing exponentially over the past decade, and the results are
extremely encouraging. The essence of AI application in nuclear medicine is to
systematize all the tasks in the department, from patient booking to imaging to
interpretation of the study. Since these tasks are carried out by computers with no or
very minimal involvement of the caregivers, they can afford to mind other essential
chores in the department.
15
15.2 Workflow inNuclear Medicine
The workow involved in a patient study in a nuclear medicine department includes
four steps (Fig.15.1): Scheduling, Imaging, Reading, and Reporting (Nensa etal.
2019). While nuclear medicine professionals are familiar with this sequence of
steps in carrying out the daily tasks of patient management, the application of AI to
carry out these steps is desirable for reasons of time-saving and accuracy of the
ultimate results (diagnosis, treatment, etc.). The following is a discussion of the
application of AI in each step.
© The Author(s), under exclusive license to Springer Science+Business Media, LLC,
part of Springer Nature 2025
G. B. Saha, Physics and Radiobiology of Nuclear Medicine,
https://doi.org/10.1007/978-1-0716-4816-2_15
277

278
Fig. 15.1 Medical Imaging workow in a typical nuclear medicine department. (This research
was originally published in JNM.Nensa etal. (2019). © SNMMI)
15 Application ofArticial Intelligence inNuclear Medicine
15.2.1 Scheduling
The scheduling of a patient for an imaging study is generally made by a designated
scheduler based on the order of the patient’s physician at an institution. However,
prior to imaging, the nuclear physician reviews the patient’s history and appropriateness of the study. All essential information of the patient may not be available at
the time of scheduling, prompting a need to contact the referring physician or a
search in in electronic health record (EHR) of the patient that contains information
on history, diagnosis, treatments and laboratory results of the patient. The application of AI using NLP can offer a helpful solution by automatically retrieving the
missing information from the EHR. In this process, AI can also annotate any hypersensitivities of the patient to medication or other agents (allergy) and also if the
study is a duplicate of a previous study. The study protocol remains the same or is
modied if needed.
On arrival, the patient is apprised by the technologist-in-charge of the procedural
sequence and any adverse effects expected with the procedure. A report by Srinivas
and Ravindran (2018) showed that the outpatient appointment system can be optimized by using the DL algorithm, which is trained on the dataset of patient appointments to recognize patterns to come up with a prediction. Curtis etal. (2018) have
shown that the patient wait times and delays can be predicted by the ML algorithm.
A major challenge to regular scheduling is the “no show” of the patient on scheduled time and date, which happens, though infrequently, disrupting the routine operation of the nuclear medicine department. This is more problematic in nuclear
medicine because of the availability of radionuclides, which is constrained by halflives and decay. In a study of 54,652 patients’ radiology appointments from

15.2 Workow inNuclear Medicine
279
researchers at Massachusetts General Hospital, USA, the AI model based on EHRdata predicted the incidence of “no-shows” with reasonable accuracy, with an area
under curve (AUC) of 0.75 (Harvey et al. 2017). With recent advancements in
machine learning and continuous progress in massive data collection, an accurate
prediction of “no-shows” by AI is quite possible in the near future.
15.2.2 Image Acquisition
Nuclear imaging of a patient is carried out on a scanner (SPECT or PET) using established settings for optimal representation of the spatiotemporal distribution of radioactivity. Different factors involved in this process are the mechanism of the detection
process (photon interaction with detector material, creation, amplication, and detection of a pulse, etc.), image noise, and body or organ motion during scanning, etc.
As the data for PET or SPECT imaging are amassed in the dataset in suitable
formats, AI algorithms are developed and trained to learn from the dataset the pattern of iterative methods to reconstruct the images. Machine learning (ML) and
Deep learning (DL) are the two most common AI methods employed for this purpose, depending on the type of data available. If the database is structured, ML can
make initial input of data followed by the “training” process, after which the algorithm learns to recognize the pattern of data to predict a correct answer to a question. In follow-up questions, the ML algorithm does not need the initial input of data
anymore and performs the search for an answer to a question based on prior training. Similarly, other AI models like DL and CNN are equally applicable to nuclear
medicine studies for image analysis.
Various parameters, such as scan time, positioning of the patient on the scanner
(FOV), energy window setting of the camera, etc., for routine scanning of patients in
nuclear medicine, are a priori set for a diagnostic study, although there may be slight
nuances among different imaging centers. Millions of nuclear medicine studies are
performed globally, and the above parameters are available from these studies to
generate databases for AI applications. To repeat, datasets are fed into the model to
train it to learn the pattern, which then results in a response (output) to a query.
One of the many responsibilities of nuclear medicine technologists is to apply a
correct protocol for a nuclear medicine study prescribed by the nuclear physician.
AI can alert the NM technologist to any change or error in scanning parameters that
might adversely affect the study.
15.2.3 Image Processing
Following collection of images, they are processed by traditional computer software mostly designed by the manufacturer of scanners. As discussed in Chap. 12,
decades ago, ltered backprojection was the primary technique for reconstruction of images, but the iterative method is the current choice for image reconstruction. Now, AI is spearheading into this arena to provide better and accurate

280
15 Application ofArticial Intelligence inNuclear Medicine
diagnoses. Images are affected by several factors, namely, photon attenuation
and scatter, body motion, noise, contrast, etc., which are discussed in detail in
Chaps. 12 and 13. Of these, attenuation and scatter of photons affect the image
resolution the most and need corrections, which had been handled by conventional methods.
Currently, the use of AI models for these corrections on PET/CT or SPECT/CT
images has drawn considerable attention. Hwang etal. (2019) used a deep neural
network to generate a PET attenuation map for the whole body 18F-FDG PET/
MRI. The network was simultaneously trained with reconstructed activity and
attenuation maps. The attenuation map was compared with the Dixon-based
4- segment method. Comparison showed the network produced a more reliable
attenuation map than the Dixon method.
McMillan and Bradshaw (2021) presented a detailed discussion of the use of AI
based synthetic attenuation and scatter corrections for PET and SPECT images
without acquiring a CT scan, and showed excellent image reconstruction. Common
AI models CNN, UNet, and GAN have been used for these corrections in brain,
whole body, myocardial perfusion, and pelvis images, and the references are cited
in the article. Even simultaneous corrections for attenuation and scatter in PET and
SPECT image reconstruction using the AI algorithm have been performed with
excellent success (Shiri etal. 2020).
Reader and Schramm (2021) published a review on the application of AI in the
reconstruction of PET images. They elucidated a number of approaches on the subject, of which the rst one is a direct AI application without any assumption of
models. A well-known example is DeepPET, which learns an encoding from the raw
data and decodes to the desired image, but the result is not convincing to many pundits. The next approach involves integration of the learning paradigm of AI into an
iterative reconstruction method (Chap. 12). The iterative loop of the reconstruction
algorithm (e.g., OSEM) is unfolded into a deep network that performs a series of
operations like forward propagation and backpropagation, producing PET images
discussed in Chap. 14.
15.2.4 Interpretation
Interpretation of PET/CT, SPECT/CT or PET/MR images is at the core of nuclear
studies, and demands a strong background knowledge from nuclear physicians for
accuracy in interpretation. Currently, physicians make interpretations by visual
inspection of images and correlating with the patient’s clinical data. While the accurate interpretation depends on the expertise of the clinician, AI, along with a diversied and quality dataset, can assist and excel in the interpretation of complex and
obscure images of PET/CT, PET/MR, and SPECT/CT.Given all necessary parameters, AI can accomplish all these activities in a short period, releasing valuable time
for clinicians to do other tasks. Various models ML, DL, CNN, GAN, etc. are the
most common AI models used for interpretation.

15.2 Workow inNuclear Medicine
281
AI models, particularly deep learning models, can automatically identify and
classify abnormalities, such as tumors and plaques, more accurately than the human
eye without much human help. Success in accurate interpretation by AI is possible
with the availability of a large amount of patient data on human diseases, which are
collected from literature and internet search and stored in the database. These datasets are used to train, validate, and test AI models and provide an outcome related to
patients’ diseases. AI models often are capable of detecting small abnormalities that
would otherwise not be detected by clinicians. The output from the AI model is
expected to approach the ground truth. While human interpretation is vulnerable to
error due to limitations in human perception, AI applies its versatile capability to
offer a more accurate output. As correlation with previous studies of a patient is
essential, AI can retrieve the pertinent information on the history of the patient’s
disease and correlate it with the current image features for a conclusive interpretation, thus relieving the clinician of time time-consuming search of EHR.
15.2.5 Reporting
Application of AI in medical reporting has been in practice for a long time. After a
nuclear medicine study is completed on a patient, the physician reads it thoroughly
and dictates the results and impression vocally, from which a voice-to-text conversion AI algorithm generates a report. Now, natural language processing in the ML
algorithm has made a great impact in this respect, and several vendors provide software for speech-to-text translation, leading to automatic transcription without the
need for typing.
Reporting of a patient study should be clearly understandable to the referring
physician and others who are involved in the management of the patient. Besides
clinical impression, the patient’s demographic data should be included in the report.
AI models can be trained to learn the pattern of data on a patient’s history and the
current study, which can generate a reliable report, obviating the need for a clinician’s report. The report should be timely, accurate, and actionable to ensure quality
care and management of the patient. Medical diagnosis error (hence reporting error)
is common in medical practice, which is, for the most part, attributable to the
increasing number of complex patient studies, causing overburden to the clinicians.
In this respect, AI can play an important role in making very accurate and genuine
reports and lessening the burden of the clinicians.
A report should include any abnormal ndings beyond the connes of patient’s
disease, alerting the care provider. It should contain recommendations for managing
the patient appropriately. Prognostic assessment of the patient’s disease needs to be
addressed in the report. While currently clinicians do these chores, AI models are
helpful in predicting the prognosis of a patient based on ndings. For example,
Milgrom etal. (2019) analyzed
or II Hodgkin lymphoma and extracted ve features from mediastinal sites. They
fed the features into an ML model and found the results highly predictive of primary
refractory disease.
18
F-FDG PET/CT scans of 251 patients with stage 1

282
15 Application ofArticial Intelligence inNuclear Medicine
15.3 Application ofAI inDifferent Diseases
inNuclear Medicine
In nuclear medicine, most studies are performed based on organ (heart, bone, brain,
prostate, etc.) or disease (cancer, hypertension, etc.) classication. It is beyond the
scope of the book to include a detailed description of AI applications on all organs
and diseases, so the following is an abbreviated discussion of a few entities of
importance in nuclear medicine, namely cancer, heart, and brain, and a referential
quote for other entities.
15.3.1 Oncology
AI has made a signicant inroad in the eld of nuclear medicine, particularly in the
detection and treatment of different cancers. ML, DL, CNN, RNN are different
algorithms commonly used in current cancer research. A number of benets from
the AI application in cancer are achievable, namely interpretation of scan and clinical data, early detection of cancer, offering patient-specic personalized treatment
from the analysis of patient’s data, calculation of optimal dosimetry in radiotherapy
of patients, predicting the response to therapy, and integration of data from other
modalities like CT, MRI, and Ultrasound for better assessment of cancer.
The use of 18FDG-PET/CT and 18FDG-PET/MR has been the gold standard for
decades in the detection of cancer in nuclear medicine. There are numerous clinical
and research reports on the subject, but it is beyond the scope of the book to include
them all, so the book cites only some pertinent reports on oncologic cases. In keeping with the aim of this book, articles on the application of AI models in a few
cancer detections will be discussed.
Ou etal. (2020) studied 44 patients with 18F‐FDG-PET/CT diagnosed with breast
carcinoma or lymphoma. They sorted out 65 breast nodules and applied radiomics
to extract features from PET and CT images using extraction software. Standard
uptake values (SUVs) were obtained. Training and validation using an ML model
were employed for image analysis and the authors claimed good results in differentiating breast carcinoma and lymphoma. Sollini etal. (2021) reviewed the radiomics
feature extraction from PET/CT images in breast cancer using an extensive literature search and analyzed the results on diagnosis, prediction of responses to chemotherapy, staging, and outcome. They concluded that the radiomics analysis of PET/
CT studies was at the feasibility stage and needs coherent efforts among researchers
synchronizing data acquisition, image processing, training, validation, and classication using an appropriate AI model.
Gandhi etal. (2023) have provided a comprehensive review on the application of
AI in the eld of lung cancer, addressing the issues of screening, diagnosis, management, and prognosis. ML, DL and radiomics stand out to be remarkably capable
algorithmic tools in the detection and prognosis of lung cancer. Zhao etal. (2024)
reported a retrospective study of 189 patients with non-small cell lung cancer
(NSCLC), who had preoperative
18
F‐FDG PET/CT.Following initial processing of

15.3 Application ofAI inDierent Diseases inNuclear Medicine
283
the images, the cohort was randomly grouped into training, validation, and testing
datasets in the ratio of 6:2:2. Different DL models such as VGG16, Googlenet,
Inception V3, Resnet50, Densenet201, and Mobilenet v2 were employed to train
and optimize the datasets, and the results were evaluated in the testing phase. DL
features were compared with tumor size and maximum standard uptake value
(SUVmax). One of the DL models, MobilenetV2, offered a better area under the
curve (AUC) 0.744, offering an improved reproducibility and stratication
of NSCLC.
In a separate report, a deep CNN model (Inception v3) was trained with 1634
randomly chosen histopathology whole-slide images of lung cancer patients and
employed to predict the pathophysiologic features. The results were consistent with
the reports of pathologists with an average AUC of 0.97 (Coudray etal. 2018).
Prostate cancer is one of the leading causes of human mortality, and much attention has been paid to early detection of primary and metastatic prostate cancer (PCa
and mPCa) by various imaging technologies, such as PET/CT, PET/MR, and ultrasound. The target molecule of these studies is a prostate-specic membrane antigen
(PSMA), a transmembrane glycoprotein, which is upregulated in PCa. Dozens of
reports on the topic are available in the literature, and only a couple of relevant ones
will be cited here. The US FDA has approved two tracers for PET/CT imaging:
68
Ga-PSMA-11 and 18F-piufolastat (Pylarify). Liu et al. (2024) and Belal et al.
(2024) have provided two separate comprehensive reviews on the application of AI
models on PET/CT images to analyze and detect PCa and mPCa. Readers are
referred to these reviews for detailed information.
Trägårdh etal. (2022) employed a CNN model in the analysis of 660 patients
suspected of PCa and recurrence, who underwent 18F-PSMA PET-CT scans. The
model was trained on 420 patients’ images and validated on 120 patients’ images.
Testing was performed on a separate dataset of 120 patients. Manual segmentations
of images made by several NM experts were compared with the values given by the
CNN model. The model yielded, on average, a sensitivity of 79% for the detection
of PCa/recurrence and 79% for lymph node metastases, and 62% for bone metastases, compared to 78%, 78%, and 59%, respectively, given by NM physicians.
Chen etal. (2019) introduced a hybrid model using a three-dimensional convolutional neural network (3D-CNN) and many objective radiomics (MaO-radiomics)
to predict lymph node metastasis (LNM) in head and neck cancer (HNC). The
MaO-radiomics extracted various features from the lymph nodes PET/CT images
(volume, diameter, perimeter, texture features including homogeneity, cluster shade,
etc.). The CNN model was designed to automatically learn features from images for
LNM prediction. The authors included 59 patients suspected of HNC in the study,
of which the lymph node dataset of 41 patients was used to train the 3D-CNN model
and the remaining 18 patients’ data for validation. Both the MaO-radiomics and the
3D-CNN model were applied individually as input, and the resultant two outputs
were then fused to obtain the nal output. The hybrid model yielded an AUC of
0.97in differentiating normal nodes from abnormal ones, supporting the usefulness
of AI applications in predicting HNC.

284
15 Application ofArticial Intelligence inNuclear Medicine
The successful application of AI in the management of many other cancers, such
as gastrointestinal, pancreatic, bone, thyroid, and multiple myeloma have been
reported, and readers are referred to the literature for reference.
15.3.2 Cardiovascular Disease
Like many other diseases, cardiovascular diseases have received their fair share of
attention in terms of their diagnosis, treatment, and prognosis by articial intelligence. Many review articles referring to articles on the application of AI in cardiovascular diseases are available in the literature. A few examples are alluded to here.
Popescu etal. (2022) presented an excellent review of the application of AI in
cardiovascular diseases, with many citations with their pros and cons of each article.
Wang et al. (2020) retrospectively analyzed 88 patients’ data who had PET/CT
scans with 13N-ammonia for myocardial perfusion and 18F-FDG for myocardial
metabolism. They employed ML models to various parameters, namely, perfusion
defects, myocardial blood ow, etc. to create an optimized multivariate model,
which was then combined with the support vector machine (SVM). The method
apparently had a much higher accuracy (AUC=0.897) in the detection of obstructive coronary artery diseases.
The prediction of major adverse cardiac events (MACE) has been a challenge
to nuclear medicine physicians, as myocardial perfusion imaging (MPI) parameters such as stress myocardial blood ow (sMBF) or myocardial ow reserve
(MFR) are difcult to measure accurately. Bors etal. (2024) have applied a pipeline of AI models to leverage 82Rb PET imaging on 234 patients to predict MACE
and compared with the performance of global, regional, radiomics, and CNN
models. They observed that regional models outperformed the global model with
the best AUC of 0.739 for the CNN model. The regional AI model applied to 17
segments yielded an AUC of 0.734, and radiomics model also outperformed the
global model.
In a report by Nakajo etal. (2024), the Ml method used
features from images to predict adverse clinical events (ACEs) in patients with cardiac sarcoidosis. Out of a cohort of 47 retrospective patients, 38 patients were
assigned to training and 9 to testing. Radiomic features were extracted from FDGPET images, which were fed to different ML models (such as decision tree, random
forest, and support vector machine). With minor variations among the models, overall, all models provided good prediction of ACEs in patients with sarcoidosis, with
AUC of >0.80.
18
F-FDG-based radiomic
15.3.3 Bone Scintigraphy
Metastatic survey of bone in various cancers, namely breast cancer, prostate cancer,
lung cancer, etc., using
patient’s disease. Traditionally, a bone scan index (BSI) is expressed as the amount
99m
Tc-MDP is an important step to evaluate the status of the

15.4 Use ofChatGPT inNuclear Medicine
285
of cancer involvement in bone as a percentage of the total skeletal mass of a reference man and is used to assess the degree of metastasis. Now, AI models, particularly the DL method, have been successfully used to diagnose bone metastatic
lesions more accurately (Zhao etal. 2020). The authors accumulated 12,222 bone
scintigraphy cases of breast, prostate, lung, and other cancers, of which 9776 cases
were assigned for training, 1223 cases for validation and optimization, and 1223
patients for testing. Pretrained CNN (ResNet-50) was applied for feature extraction,
and fully connected layers were used for classication of metastasis or nonmetastasis spots. The AI results were compared with the dictations of three nuclear
physicians. The AI model yielded higher AUC values (0.955 for prostate cancer,
0.988 for breast cancer, and 0.957 for lung cancer), supporting the usefulness of the
AI paradigm in bone imaging to detect cancer spread.
15.3.4 Thyroid Imaging
Thyroid disease affects more than 300 million people worldwide. In clinical practice, physicians diagnose thyroid disorders using a number of tests, like thyroid
uptake of
roid imaging with a radioactive tracer, ultrasonography, and biopsy.
to classify scintigraphy thyroid images. They procured 3087 thyroid images from
one clinical center, of which 2468 were used for training and the remaining 619 for
validation. They collected an additional 302 images for external validation. Four
different pre-trained neural networks, namely, DenseNet169, ResNet50,
InceptionV3, and InceptionResNetV2 used to construct AI models, which were
trained separately with transfer learning. All models yielded accuracy greater than
90%, with InceptionV3 giving the highest accuracy of 92.73%. In internal validation, the AUC values were 0.986 for diffusely increased, 0.997 for diffusely
decreased, 0.998 for focal increased, and 0.945 for heterogenous uptake, whereas
the external validation yielded the corresponding values as 0.939, 1.000, 0.974, and
0.915, respectively. This study supports the use of AI for the classication of common thyroid diseases.
131
123
I‐ or
I‐sodium iodide, blood test (thyroid-stimulating hormone), thy-
Yang etal. (2021) employed deep convolutional neural network (DCNN) models
15.4 Use ofChatGPT inNuclear Medicine
Currently ChatGPT has become a popular LLM AI model in many applications,
including the healthcare sector. While different models of AI are used to analyze
nuclear medicine images with optimal success, nuclear physicians have recently
become increasingly interested in ChatGPT for clinical applications, for its simplistic approach to problem-solving. In Chap. 14, a detailed discussion on ChatGPT is
provided as to its characteristics and applications. Below, we cite a couple of references on the use of ChatGPT in nuclear medicine studies.

286
15 Application ofArticial Intelligence inNuclear Medicine
In 2023, Buvat and Weber published an article on the use of ChatGPT asking
several intuitive questions related to the practice of nuclear medicine imaging. The
rst question asked was to make a PET/CT report on a patient with suspected lung
cancer, based on scans showing regions of increased uptake. The report given in a
few seconds was quite impressive, suggesting lung cancer in the patient and recommending correlation of the ndings with biopsy. In another question, the authors
asked ChatGPT if it would report SUV
authors’ surmise, ChatGPTpicked SUV
correct answer SUV
, indicating that ChatGPT can make mistakes. However,
max
or SUV
peak
as the answer, which is wrong against the
peak
in a medical report. To the
max
since the publication of the article, a lot more data has been added to the database,
and the Chatbot strategy has improved signicantly, expecting more accurate
responses to queries in nuclear medicine.
Belge Bilgin etal. (2024) evaluated the accuracy, conciseness, and readability of
responses from OpenAI ChatGPT-4 and Google Bard to patient inquiries. Twelve
questions on
177
Lu-PSMA-617 therapy were prompted to ChatGPT and Google
Bard and the AI generated responses were blindly rated by eight experts. Overall,
ChatGPT provided more accurate responses than Bard, yet the former needs further
improvement.
ChatGPT has been upgraded to ChatGPT-4V(ision) in two versions—one as
“Default” and the other as “Advanced Data Analysis (beta)” version. Rogasch etal.
(2024) presented 11 scintigraphy and 4 PET images to these models, asking them to
identify the type of examination and tracer used, and any abnormalities found on the
images, with explanation. The examination and the tracer were accurately identied
by the Default version in most cases, while the beta version failed in all cases, indicating the limitation of ChatGPT in image analysis in nuclear medicine.
ChatGPT is a large language model of AI primarily used to write texts and
communicate with humans, with minimal application in medical imaging. One
of its main advantages is that patients can retrieve personal medical reports on
their personal cell phone anywhere and at any time they choose. Concerns of
security about data privacy and legal liability for medical images, however, have
limited the use of ChatGPT.Yet clinicians are increasingly getting interested in
utilizing ChatGPT in their clinical practice because of its easy accessibility on
personal phones. With time, ChatGPT is improving to the point to analyze medical images like other complex AI tools, and it will overcome the difculties it
faces now.
It should be emphasized that many chatbots are in current use, like Open AI’s
ChatGPT 5, Google.s Gemini, Microsoft’s Copilot, etc., and all of them provide
reasonable answers to prompts given to them.
15.5 Drug Development
Traditional drugs take years (10 to 15years) to develop and a few more years for
patients’ use. These methods suffer from the limitation of their reliance on the trialand- error technique, which results in uncertainty in the efcacy of the drug.
Соседние файлы в папке Библиотека им академика М.И. Перельмана
