Comparison of performance between an ${}^{18}\mathrm{F}$-FDG PET normal brain template and a commercial template using the MIMneuro software

Abstract.

Purpose

The aim of this study was to create and validate a normal brain template of ${}^{18}\mathrm{F}$-fluorodeoxyglucose (${}^{18}\mathrm{F}\text{-}\mathrm{FDG}$) uptake using the MIMneuro software to improve clinical practice.

Approach

One hundred and nine volunteers underwent an ${}^{18}\mathrm{F}\text{-}\mathrm{FDG}$ positron emission tomography/computed tomography scan. Sixty-three participants with normal Alzheimer’s disease (AD) biomarkers were used to create a template. A group of 23 participants with abnormal AD biomarkers and an additional group of 23 participants with normal AD biomarkers were used to validate the performance of the generated template. The MIMneuro software was used for the analysis and template creation. The performance of our newly created template was compared with that of the MIMneuro software template in the validation groups. Results were confirmed by visual analysis by nuclear medicine physicians.

Results

Our created template provided higher sensitivity, specificity, positive predictive value, and negative predictive value (NPV; 90%, 97.83%, 100%, and 100%, respectively) than did the MIMneuro template when using the positive validation group. Similarly, slightly higher performance was observed for our template than for the MIMneuro template in the negative validation group (the highest specificity and NPV were 100% and 100%, respectively).

Conclusions

Our normal brain template for ${}^{18}\mathrm{F}\text{-}\mathrm{FDG}$ was shown to be clinically useful because it enabled more accurate discrimination between aging brain and patients with AD. Thus, the template may improve the accuracy of AD diagnoses.

1. Introduction

The alteration of glucose metabolism in the brain is considered an important biomarker of Alzheimer’s disease (AD).¹ AD is a major problem worldwide, and dementia associated with AD affects the quality of life of both patients and their families. Thus, improving diagnosis and screening methods for the early detection of AD is crucial.^2,3

High sensitivity and specificity can be achieved using ${}^{18}\mathrm{F}$-fluorodeoxyglucose (FDG) positron emission tomography (PET) for differentiating patients with AD from non-demented healthy controls. In combination with clinical examinations, the ${}^{18}\mathrm{F}$-FDG radiotracer is used for the diagnosis of AD and other neurodegenerative diseases.⁴

The interpretation of brain PET scans is based on both qualitative (visual) and quantitative analyses. Both analysis methods are commonly used together for accurate and precise diagnoses. Improving quantitative analyses is essential for increasing the accuracy of diagnoses, particularly in patients with uncertain radiotracer hypoactivity, such as those with aging-related hypometabolism and mild cognitive impairment. One quantitative analysis method involves the comparison of data with a normal brain template. This procedure can be applied with high reliability and allows the detection of brain areas with abnormalities and radiotracer retention. Brain template comparisons can be performed with various software, such as the MIMneuro tool (MIM Software Inc., Cleveland, Ohio), the Senium Neuro database (Siemens, Munich, Germany), Statistical Parametric Mapping (SPM; The Wellcome Centre for Human Neuroimaging, UCL Queen Square Institute of Neurology, London, UK), and Cortex ID (General Electric Company; GE, Boston, Massachusetts). However, the templates established by these programs are based on a normal database of different population groups. Therefore, the use of a specific local normal database may be a crucial tool for improving the accuracy and precision of interpretations of PET data.⁵

The aim of the present study was to create a normal brain template of FDG depositions using the MIMneuro software and validate the brain template for use in clinical practice with high reliability and reproducibility.

2. Methods

This study was approved by the Human Research Ethics Committee of Chulabhorn Research Institute. Written informed consent was obtained from all participants before the study.

2.1. Participants

One hundred and fifty-two enrolled participants provided their written informed consent to participate in the study. Each participant was provided a questionnaire, as well as the Thai version of the Montreal Cognitive Assessment (MoCA-Thai) for the assessment of mental status. All participants were assessed and diagnosed by clinicians using the National Institute on Aging Alzheimer’s Association (NIA-AA) criteria for probable AD.⁶ All participants underwent three radiotracers examination including ${}^{18}\mathrm{F}$-PI-2620, ${}^{18}\mathrm{F}$-FDG, and ${}^{18}\mathrm{F}$-florbetaben (${}^{18}\mathrm{F}\text{-}\mathrm{FBB}$). ${}^{18}\mathrm{F}\text{-}\mathrm{FBB}$ was a $\beta$-amyloid-targeting PET tracer.^7,8 Whereas, ${}^{18}\mathrm{F}$-PI-2620 was a PET tracer with a high binding affinity for aggregated tau.^9,10 Both radiotracers were obtained and approved by Life Molecular Imaging Company (LMI GmbH, Berlin, Germany). All images were interpreted according to the amyloid/tau/neurodegeneration classification (A/T/N) system.¹¹ Participants were excluded if they had concurrent underlying disease such as hypertension, dyslipidemia, diabetes mellitus, cardiovascular disease, pulmonary, or renal conditions. No participants had a history of psychological or neurological diseases, use of psychotropic drugs, or cancer found within the last 5 years.

Examples of normal and abnormal AD biomarkers were shown in Figs. 1 and 2, respectively. Data were then divided into three datasets as follows.

Fig. 1.

Characteristics of a cognitively normal participant according to the A/T/N classification. (a) ${}^{18}\mathrm{F}$-FDG PET interpreted as N−. (b) ${}^{18}\mathrm{F}$-FBB PET interpreted as A−. (c) ${}^{18}\mathrm{F}$-PI-2620 PET interpreted as T−. All images show normal AD biomarkers.

Fig. 2.

Characteristics of a participant with AD, according to the A/T/N classification. (a) ${}^{18}\mathrm{F}$-FDG PET with hypometabolism of the bilateral parieto-temporal lobes, posterior cingulate gyrus, and bilateral precuneus, interpreted as N+. (b) ${}^{18}\mathrm{F}$-FBB PET with abnormal uptake in the cortical gray matter, including the lateral temporal lobe, frontal lobe, posterior cingulate gyrus, precuneus, and parietal cortex, interpreted as A+. (c) ${}^{18}\mathrm{F}$-PI-2620 PET with abnormal uptake in the bilateral occipital, frontal, and parieto-temporal cortices, interpreted as T+. All images show abnormal AD biomarkers.

2.1.1. Template creation dataset

There were 63 volunteers with (1) normal AD biomarkers in all three PET scans, (2) a score of over 25 on the MoCA-Thai, (3) preserved activities of daily living, (4) absence of significant levels of impairment in other cognitive domains, (5) no sign and symptom of mild cognitive impairment or dementia, and (6) no diagnosis of probable AD according to criteria from NIA-AA workgroups.

2.1.2. Positive validation dataset

A total of 23 volunteers were included with abnormal AD biomarkers in all three PET scans and diagnosed with probable AD by using criteria from NIA-AA workgroups.

2.1.3. Negative validation dataset

Another 23 volunteers comprised the same conditions as Sec. 2.1.1.

Datasets and participants’ characteristics are shown in Tables 1 and 2. The groups of AD pathologic change, AD with N-, AD and concomitant suspected non-AD pathologic change, and non-AD pathologic change were not included to the template creation and evaluation datasets since they associate with other dementia diseases, which can cause variations in the analysis.

Table 1.

Participants characteristics are described by NIA-AA criteria and A/T/N classification.

NIA-AA criteria	A/T/N classification			Participants
Normal	A−	T−	N−	86
AD pathologic change	A+	T−	N−	23
AD with N−	A+	T+	N−	12
AD and concomitant suspected non AD pathologic change	A+	T−	N+	3
Non AD pathologic change	A−	T−	N+	5
AD	A+	T+	N+	23
Total				152

Table 2.

Characteristics of participants in each dataset.

Template creation dataset
Age (years)
Range	56–78
Mean	65.97 ± 5.77
Sex ($n$, %)
Male	19 (30.16)
Female	44 (69.84)
Education (years)
Range	6–16
Mean	12.87
Median	12
Positive validation dataset
Age (years)
Range	59–78
Mean	68.87 ± 4.98
Sex ($n$, %)
Male	8 (34.78)
Female	15 (65.22)
Education (years)
Range	4–16
Mean	12.78
Median	12
Negative validation dataset
Age (years)
Range	55–70
Mean	63.65 ± 3.62
Sex ($n$, %)
Male	7 (30.43)
Female	16 (69.57)
Education (years)
Range	6–16
Mean	15.04
Median	16
MIM’s template database
Age (years)
Range	41–80
Mean	63.79 ± 9.98
Sex ($n$, %)
Male	19 (44.19)
Female	24 (55.81)

We used a normal FDG-PET database provided by the MIMneuro software. According to the MIMneuro manual,¹² the database was created from 43 individuals who fulfilled the eligibility criteria as determined by an expert neurologist. All brain scans were reviewed by a radiologist to confirm participants were healthy before inclusion in the database. Subjects were excluded if they had any unfavored condition, such as brain tumors, brain metastasis, and history of a stroke. The MIM’s dataset characteristics are also shown in Table 2.

2.2. Image Acquisition

All participants underwent ${}^{18}\mathrm{F}$-FDG-PET using a Siemens Biograph Vision or Siemens Biograph 16 scanner (Siemens Healthcare GmbH, Erlangen, Germany) in three-dimensional (3D) mode.

2.2.1. Biograph Vision ${}^{18}\mathrm{F}$-FDG acquisition protocol

The PET scan was performed 50 min after intravenous injection of 2.5 mCi ${}^{18}\mathrm{F}$-FDG. A static brain scan was acquired for 10 min, during which brain computed tomography (CT) images for attenuation correction were also collected. The reconstruction parameters were: matrix size = 440, zoom = 2, all-pass filter using the TrueX + time-of-flight (ultrahigh-definition-PET) reconstruction method with eight iterations, and five subsets. Forty-three participants underwent an ${}^{18}\mathrm{F}$-FDG scan on the Biograph vision scanner.

2.2.2. Biograph 16 ${}^{18}\mathrm{F}$-FDG acquisition protocol

Static imaging was performed 50 min after intravenous injection of 2.5 mCi ${}^{18}\mathrm{F}\text{-}\mathrm{FDG}$. A static brain scan was acquired for 10 min, during which brain CT images for attenuation correction were also collected. The reconstruction parameters were: matrix size = 336, zoom = 1, and a Gaussian filter with a full-width at half-maximum of 5.0. Iterative reconstruction was performed using the ordered subset expectation maximization with four iterations and 16 subsets. Fourteen participants underwent an ${}^{18}\mathrm{F}$-FDG scan on the Biograph 16 scanner.

2.2.3. MIM ${}^{18}\mathrm{F}$-FDG acquisition protocol

The PET scan was performed for a minimum of 20 min after intravenous injection. Brain PET data acquisition was started after a minimum of 30 min after the injection and had a duration of at least 5 min for a 3D acquisition and 10 min for a two-dimensional acquisition. The PET brain images were attenuation-corrected and iteratively reconstructed to an in-plane voxel size of $\sim 2.5\mathrm{mm}$ or less. The acquired FDG-PET scans were registered to a common template space using both linear and nonlinear registration for inclusion in the database.

2.2.4. Biograph vision ${}^{18}\mathrm{F}$-FBB acquisition protocol

The PET scan was performed 90 min after intravenous injection of 8.0 mCi ${}^{18}\mathrm{F}$-FBB. A static brain scan was acquired for 20 min, during which brain computed tomography (CT) images for attenuation correction were also collected. The reconstruction parameters were matrix size = 440, zoom = 2, all-pass filter using the TrueX + time-of-flight (ultra-high-definition-PET) reconstruction method with eight iterations, and five subsets.

2.2.5. Biograph vision ${}^{18}\mathrm{F}$-PI2620 acquisition protocol

The PET scan was performed 30 min after intravenous injection of 5.0 mCi ${}^{18}\mathrm{F}\text{-}\mathrm{FBB}$. A static brain scan was acquired for 45 min, during which brain computed tomography (CT) images for attenuation correction were also collected. The reconstruction parameters were: matrix size = 440, zoom = 2, all-pass filter using the TrueX + time-of-flight (ultrahigh-definition-PET) reconstruction method with eight iterations, and five subsets.

2.3. Creating the Normal Template

Sixty-three reconstructed ${}^{18}\mathrm{F}\text{-}\mathrm{FDG}$ images were imported into the MIMneuro software. Subsequently, the data were processed following a predesigned workflow for creating templates according to the MIMneuro software manual.

The first step involved creating a new or custom database. The imported image was registered using the Neuro Registration Tool, which is based on linear and nonlinear registration. Information regarding the participant, such as age, sex, radiotracer, and ethnicity, was entered manually. This step was repeated for all images. The second step involved creating a comparison set, which can be either (1) a static comparison set, which is a selection of images from the normal database to be used as normal when performing the analysis, or (2) a dynamic set, which uses the entire normal database, with the option to set certain matching criteria (i.e., sex, age, camera, and reconstruction matching) so that only images in the normal database that meet the criteria will be used as normal for analysis. In this study, we created a static comparison set to be used for the custom database.

The anatomic information used for the template creation and validation was provided by MIM and was called the MIM single brain atlas, as shown in Fig. 4. It was developed from a high-resolution T1-weighted magnetic resonance imaging (MRI) scan acquired on a 1.5 Tesla scanner of an asymptomatic 57-year-old male subject with no evidence of neuropsychiatric disease or significant past medical history. The MRI scan was reviewed by a radiologist for suitability. Regions were defined in conjunction with a neuroanatomist and expert physicians in the field using the MIM software’s standard contouring tools. The atlas was defined according to four levels. The first atlas level referred to lobe-level structures, the second to sublobar structures, the third to structures at the gyral level, and the fourth level included individual structures that did not fall into the previous three levels (e.g., the hippocampus and amygdala).¹³

Fig. 4.

(a) MIM’s template space consisted of an FDG intensity map fused with a single brain atlas, which is shown with a pink contour. (b) An FDG-PET image of a cognitively normal participant fused with the template space. The imported PET image did not fit perfectly with the template space, which caused a degree of misregistration.

2.4. Template Validation

The validation datasets were analyzed using the FDG neuro-analysis workflow in the MIMneuro software analysis tools using a custom template. The processes of the workflow are shown in Fig. 3. Results were displayed as $z$-score and 3D stereotactic surface projection (3D-SSP) images. The regions were selected and used in the analysis. All images were then reanalyzed using MIM’s template, and the outcomes were compared and confirmed by visual analysis.

Fig. 3.

The processes of the FDG MIMneuro software analysis workflow. In the first step, the imaging data underwent postprocessing by removing the unnecessary image background. Next, the data were automatically compared with the selected comparison set using Boolean logic. The radiotracer was then addressed. The data were registered and contoured using brain atlases and reoriented to the anterior commissure-posterior commissure plane. Atlases used in the registration and contouring steps comprised four levels of single-brain atlases and required user verification before proceeding. In the next step, the data were subjected to voxel analysis, which produced in $z$-scores that were normalized across the whole brain. The data were then analyzed using region analysis, and all results were subsequently displayed.

The z-score of each structure was the number of standard deviations the structure was away from the mean of the database for that structure. Areas of reduced uptake were considered significant when they were $-1.65$ standard deviations or more from the mean of the database (corresponding to a 95% statistical significance level). Regions with a $z$-score of $-1.65$ or lower were labeled as hypometabolism These regions were then compared with those derived from the visual analysis by nuclear medicine physicians.

The voxel-based analysis results were displayed as registered PET images as 3D-SSPs, with a color table corresponding to the tracer $z$-score. Increased and decreased uptake were defined as $z$-scores above and below the $z$-score threshold of 1.65. Examples are shown in Figs. 5 and 6.

Fig. 5.

A participant with normal AD biomarkers. The 3D-SSP did not indicate any areas that had significantly lower uptake than those of our database. (b) A participant with abnormal AD biomarkers. The 3D-SSP indicated significantly lower uptake in the parietal cortex, precuneus, and posterior cingulate gyrus compared with our database.

Fig. 6.

(a) The same participant with normal AD biomarkers as Fig. 5. The 3D-SSP did not indicate any areas that had significantly lower uptake compared with MIM’s database. (b) The same participant with abnormal AD biomarkers as Fig. 5. The 3D-SSP indicated significantly lower uptake in the parietal cortex, precuneus, and posterior cingulate gyrus compared with MIM’s database.

2.5. Statistical Analysis

The results were analyzed using diagnostic tests, and sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated. All analyses were conducted in Stata version 16.1 (StataCorp LLC, College Station, Texas).

3. Results

As shown in Table 3, when the positive group was tested, our template provided better results overall than the MIM’s template in most regions as reflected by higher sensitivity, specificity, PPV, and NPV, with the highest values of 90%, 97.83%, 100%, and 100%, respectively).

Table 3.

Results of the positive validation set using MIM’s and custom templates.

Region	$n$		Visual		Sensitivity		Specificity		PPV		NPV
	Custom	MIM	Custom	MIM	Custom	MIM	Custom	MIM	Custom	MIM	Custom	MIM
Frontal lobe	3	1	6	6	33.33	16.67	97.5	100	66.67	100	90.7	88.89
Parietal lobe	10	7	26	26	38.46	26.92	100	100	100	100	55.56	51.28
Occipital lobe	11	5	10	10	90	50	94.44	100	81.82	100	97.14	85.71
Temporal lobe	11	19	21	21	38.1	72.22	88	66.67	72.73	68.42	62.86	70.59
Precuneus	18	17	27	27	59.26	55.56	89.47	89.47	88.89	88.24	60.71	58.62
Cuneus	7	3	6	6	83.33	33.33	95	97.5	71.43	66.67	97.44	90.7
Posterior Cingulate gyrus	27	24	32	32	81.25	75	92.86	83.33	96.29	87.5	68.42	68.18
Anterior Cingulate gyrus	10	7	4	4	50	75	80.95	90.48	20	42.86	94.44	97.44
Caudate	8	8	0	0	—	—	82.61	82.61	0	0	100	100
Putamen	1	0	0	0	—	—	97.83	100	0	—	100	100
Thalamus	1	0	0	0	—	—	97.83	100	0	—	100	100
Brain stem	0	2	0	0	—	—	100	95.65	—	0	100	100
Cerebellum	0	2	1	1	0	100	100	97.78	—	50	97.83	100

$n$ = Total number of hypometabolism regions in the dataset. Visual = Total number of regions interpreted as hypometabolism by visual analysis in the dataset.

Our template also produced slightly better results when applied to the negative group, as shown in Table 4 (the highest specificity and NPV were 100% and 100%, respectively).

Table 4.

Results of the negative validation set using MIM’s and custom templates.

Region	$n$		Visual		Sensitivity		Specificity		PPV		NPV
	Custom	MIM	Custom	MIM	Custom	MIM	Custom	MIM	Custom	MIM	Custom	MIM
Frontal lobe	0	0	0	0	—	—	100	100	—	—	100	100
Parietal lobe	1	0	0	0	—	—	97.83	100	0	—	100	100
Occipital lobe	1	1	0	0	—	—	97.83	97.83	0	0	100	100
Temporal lobe	0	2	0	0	—	—	100	95.66	—	0	100	100
Precuneus	0	0	0	0	—	—	100	100	—	—	100	100
Cuneus	0	0	0	0	—	—	100	100	—	—	100	100
Posterior Cingulate gyrus	2	4	0	0	—	—	95.66	91.30	0	0	100	100
Anterior Cingulate gyrus	2	2	0	0	—	—	95.66	95.65	0	0	100	100
Caudate	0	0	0	0	—	—	100	100	—	—	100	100
Putamen	0	0	0	0	—	—	100	100	—	—	100	100
Thalamus	2	1	0	0	—	—	95.66	97.83	0	0	100	100
Brain stem	0	8	0	0	—	—	100	82.61	—	0	100	100
Cerebellum	0	3	1	0	—	—	100	93.48	—	0	100	100

$n$ = Total number of hypometabolism regions in the dataset. Visual = Total number of regions interpreted as hypometabolism by visual analysis in the dataset.

4. Discussion

Recently, PET brain images have been quantitatively analyzed using a database of radiopharmaceutical uptake in brain regions of healthy subjects as a normal template to improve diagnoses of neurodegenerative diseases. Although several commercial brain templates are available, differences in ethnicity may result in inaccurate diagnoses.^14–17 Thus, we created a normal brain template from ${}^{18}\mathrm{F}\text{-}\mathrm{FDG}$-PET brain images acquired from a cognitively normal Thai population.

Generally, the registration of multimodal images between PET and MRI is a fundamental task in 3D image analyses and is applied to the process of brain template comparisons. Most commercial brain templates are based on Caucasian populations. In the present study, the subjects were of Thai ethnicity. Differences in brain morphology can result in the misregistration of fusion images and contribute to variation in analyses.^18,19

In Thai subjects, our template for FDG-PET analysis had higher sensitivity and specificity than did MIM’s template. However, misregistration did occur, particularly around the ventricles and posterior brain regions because our generated brain template was based on anatomical data provided by the MIMneuro software, which were derived from a Caucasian population. Although such variations in brain shape and volume affect the performance of the template, physiological uptake may also vary across specific populations depending on various factors, such as age, sex, cognitive status, and ethnicity. We acquired data from the local population, which may be more suitable. Consequently, our template outperformed MIM’s template despite having registration problems. Remarkably, when using our template in patients with AD, we obtained better results in the posterior cingulate gyrus, parietal cortex, and precuneus, which are the significant brain areas significant to AD.^20,21

In our previous study, we created ${}^{18}\mathrm{F}\text{-}\mathrm{THK}$-5351 and 11C-labeled Pittsburgh compound-B PET normal brain templates in a Thai population using the SPM software, which is a widely applied analysis method. The software requires a PET image and a diffusion tensor image obtained using an MRI scanner of each subject to create a template. A reference point is then manually selected during the registration and reorientation process. If there are MRI data available, results appear to be more reliable and accurate than those of the current study.¹⁷ However, Wang et al. reported that even with the use of individual MRI scans in the creation of the template, there is still a degree of distortion. Nevertheless, both studies demonstrated that the application of a local template provides better results when used in a non-Caucasian population.²²

In our retrospective study, we only used subjects’ PET images. The analysis workflow was predesigned, with the ability to be easily modified by the user, which is convenient and less time-consuming in practice. However, compared with SPM, the in-house software requires more data, instruments, and time. Moreover, image analysis experts are necessary.²³ Nevertheless, despite the absence of individual MRI scans, our template performed relatively well, which is in concordance with previous studies. Partovi et al. evaluated a quantitative software-aided (MIMneuro) approach for the diagnosis of early-stage AD using FDG-PET and reported that the fully automated quantitative software improved physicians’ specificity, especially those who overemphasize minimal physiological changes that contribute to intersubject variability. Furthermore, combining the software approach with qualitative visual interpretation could significantly improve current clinical diagnoses of early-stage AD.²⁴ Piper compared the reduction in anatomical variability between SPM and MIMneuro when applied directly to PET brain volumes and found significantly better correlation results for MIMneuro than for SPM.²⁵

Our study has several limitations. We used a small number of subjects, which was insufficient for dividing into dynamic comparison sets, such as specific age-matched or sex-matched templates. PET images acquired from different machines may affect the performance of the template because of differences in efficiency and data acquisition parameters.²⁶ Misregistration impacts template performance particularly in small brain structures. Therefore, we only used 24 major regions of the brain that are less affected by spatial misregistration. Attenuation correction errors caused by patient motion influenced the intensity of PET data as well as the analysis. However, the template was validated using validation data for statistical comparisons and calculations. Moreover, most of the brain regions that showed a significantly lower uptake of ${}^{18}\mathrm{F}\text{-}\mathrm{FDG}$ as calculated using the template were compatible with the abnormal regions assessed and identified by visual analysis. Nevertheless, the combination of various methods, such as qualitative and quantitative analyses, and clinical information are important aspects that should be considered for making a final diagnosis.

Even though dynamic comparison sets were not applied to this study, our template demonstrated higher sensitivity and specificity for differentiating abnormal hypometabolism and a healthy aging brain.

5. Conclusion

We developed a normal brain template for ${}^{18}\mathrm{F}\text{-}\mathrm{FDG}\text{-}\mathrm{PET}$ analysis that may be clinically useful for the local population because it more accurately discriminates between aging brain and patients with AD. Our template produced better results than did the commercial template, which was created using a non-Asian population. The results of our study suggest that our created template is feasible for clinical use as a diagnostic tool to improve the accuracy of AD diagnoses.

Biographies

Chanisa Chotipanich is an associate professor and the director of the National Cyclotron and PET Centre, Chulabhorn Hospital. She received her MD degree from Thammasat University and her Diploma degree in nuclear medicine from Chulalongkorn University, in 1996 and 2000, respectively. She is the author of more than 51 journal papers and has written five books. Her current research interests include nuclear medicine, oncology, neurology, and cardiology.

Natdanai Hirata is a nuclear medicine radiographer at the National Cyclotron and PET Centre, Chulabhorn Hospital. He received his BS degree in radiological technology from Chiang Mai University in 2020. His current research interests include artificial intelligence, image analysis, data processing, and nuclear medicine.

Attapon Jantarato is a nuclear medicine radiographer at the National Cyclotron and PET Centre, Chulabhorn Hospital. He received his BS degree in radiological technology from Chiang Mai University in 2016. He is the author of more than 15 journal papers. His current research interests include image analysis, data processing, nuclear medicine, and medical imaging analysis.

Peerapon Kiatkittikul is a nuclear medicine physician at the National Cyclotron and PET Centre, Chulabhorn Hospital. He received his MD and his Diploma degree in nuclear medicine from Khon Kaen University, in 2015 and 2019. He is the author of more than nine journal papers. His current research interests include nuclear medicine, oncology, neurology, and cardiology.

Dheeratama Siripongsatian is a nuclear medicine physician at the National Cyclotron and PET Centre, Chulabhorn Hospital. He received his MD and Diploma degrees in nuclear medicine from Khon Kaen University, in 2016 and 2020. He is the author of more than 10 journal papers. His current research interests include nuclear medicine, oncology, neurology, and cardiology.

Anchisa Kunawudhi is an assistant professor and a nuclear medicine physician at the National Cyclotron and PET Centre, Chulabhorn Hospital. She received her MD and Diploma degree in nuclear medicine from Chulalongkorn University, in 2008 and 2011. She is the author of more than 25 journal papers. Her current research interests include nuclear medicine, oncology, neurology, and cardiology.

Chetsadaporn Promteangtrong is an assistant professor and a nuclear medicine physician at the National Cyclotron and PET Centre, Chulabhorn Hospital. She received her MD and Diploma degrees in nuclear medicine from Siriraj Hospital, in 2008 and 2011. She is the author of more than 23 journal papers. Her current research interests include nuclear medicine, oncology, neurology, and cardiology.

Saiphet Vanprom is a registered nurse at the National Cyclotron and PET Centre, Chulabhorn Hospital. She received her BNS degree from Mahidol University, in 2011. She is the author of more than three journal papers. Her current research interests include nuclear medicine, oncology, neurology, and cardiology.

Biographies of the other authors are not available.

Disclosures

The authors declare no conflicts of interest.