Recent advances in deep learning have impacted various scientific and industrial fields. Due to the rapid application of deep learning in biomedical data, molecular imaging has also started to adopt this technique. In this regard, it is expected that deep learning will potentially affect the roles of molecular imaging experts as well as clinical decision making. This review firstly offers a basic overview of deep learning particularly for image data analysis to give knowledge to nuclear medicine physicians and researchers. Because of the unique characteristics and distinctive aims of various types of molecular imaging, deep learning applications can be different from other fields. In this context, the review deals with current perspectives of deep learning in molecular imaging particularly in terms of development of biomarkers. Finally, future challenges of deep learning application for molecular imaging and future roles of experts in molecular imaging will be discussed.
Biomarkers ; Clinical Decision-Making ; Learning ; Machine Learning ; Molecular Imaging ; Nuclear Medicine ; Precision Medicine ; Statistics as Topic

Recent advances in deep learning have impacted various scientific and industrial fields. Due to the rapid application of deep learning in biomedical data, molecular imaging has also started to adopt this technique. In this regard, it is expected that deep learning will potentially affect the roles of molecular imaging experts as well as clinical decision making. This review firstly offers a basic overview of deep learning particularly for image data analysis to give knowledge to nuclear medicine physicians and researchers. Because of the unique characteristics and distinctive aims of various types of molecular imaging, deep learning applications can be different from other fields. In this context, the review deals with current perspectives of deep learning in molecular imaging particularly in terms of development of biomarkers. Finally, future challenges of deep learning application for molecular imaging and future roles of experts in molecular imaging will be discussed.

Metastatic disease involving the thyroid gland is uncommon. Thyroid metastases has been previously described from several primary cancers of lung, breast, and kidney. Because of the lower incidence and ambiguous clinical significance, it is not easy to consider thyroid metastasis and decide the optimal time for performing diagnostic examination. Here, we reported two cases of metastatic diseases of thyroid in patients who had underlying Hashimoto’s thyroiditis: a 39-year-old woman who had thyroid metastasis of breast cancer with underlying Hashimoto’s thyroiditis, and a 44-year-old woman with metastatic lung cancer.

Radiomics is a medical imaging analysis approach based on computer-vision. Metabolic radiomics in particular analyses the spatial distribution patterns of molecular metabolism on PET images. Measuring intratumoral heterogeneity via image is one of the main targets of radiomics research, and it aims to build a image-based model for better patient management. The workflow of radiomics using texture analysis follows these steps: 1) imaging (image acquisition and reconstruction); 2) preprocessing (segmentation & quantization); 3) quantification (texture matrix design & texture feature extraction); and 4) analysis (statistics and/or machine learning). The parameters or conditions at each of these steps are effect on the results. In statistical testing or modeling, problems such as multiple comparisons, dependence on other variables, and high dimensionality of small sample size data should be considered. Standardization of methodology and harmonization of image quality are one of the most important challenges with radiomics methodology. Even though there are current issues in radiomics methodology, it is expected that radiomics will be clinically useful in personalized medicine for oncology.

68Ga-DOTATOC PET/CT is widely used as a functional imaging technique in the detection and characterization of neuroendocrine tumors (NETs). Pancreatic NET and intrapancreatic accessory spleen (IPAS) have similar radiologic characteristics in anatomical imaging and usually show high uptake of 68Ga-DOTATOC. Thus, it is challenging to make a differential diagnosis between NET and IPAS when the tumor-like lesion is located in the pancreatic tail. Here, we present a case of 68Ga-DOTATOC PET-positive pancreatic tail lesion with high arterial enhancement on CT and MRI. Since 99mTc-labeled damaged red blood cell does not accumulate on NET, a negative spleen scan finding was a crucial diagnostic step to decide surgical resection, which was histologically proven as insulinoma. Our case shows a promising role of additional use of spleen scan with SPECT/CT for the differential diagnosis of 68Ga-DOTATOC PET-positive pancreatic NET from the accessory spleen.

Radiomics is a medical imaging analysis approach based on computer-vision. Metabolic radiomics in particular analyses the spatial distribution patterns of molecular metabolism on PET images. Measuring intratumoral heterogeneity via image is one of the main targets of radiomics research, and it aims to build a image-based model for better patient management. The workflow of radiomics using texture analysis follows these steps: 1) imaging (image acquisition and reconstruction); 2) preprocessing (segmentation & quantization); 3) quantification (texture matrix design & texture feature extraction); and 4) analysis (statistics and/or machine learning). The parameters or conditions at each of these steps are effect on the results. In statistical testing or modeling, problems such as multiple comparisons, dependence on other variables, and high dimensionality of small sample size data should be considered. Standardization of methodology and harmonization of image quality are one of the most important challenges with radiomics methodology. Even though there are current issues in radiomics methodology, it is expected that radiomics will be clinically useful in personalized medicine for oncology.
Diagnostic Imaging ; Humans ; Metabolism ; Population Characteristics ; Positron-Emission Tomography and Computed Tomography ; Precision Medicine ; Sample Size

Purpose: EGFR-mutation (EGFR-mt) is a major oncogenic driver mutation in lung adenocarcinoma (ADC) and is more oftenobserved in Asian population. In lung ADC, some radiomics parameters of FDG PET have been reported to be associated withEGFR-mt. Here, the associations between EGFR-mt and PET parameters, particularly asphericity (ASP), were evaluated inAsian population. Methods: Lung ADC patients who underwent curative surgical resection as the first treatment were retrospectively enrolled.EGFR mutation was defined as exon 19 deletion and exon 21 point mutation and was evaluated using surgical specimens. OnFDG PET, image parameters of maximal standardized uptake value (SUVmax), metabolic tumor volume (MTV), total lesionglycolysis (TLG), and ASP were obtained. The parameters were compared between EGFR-mt and wild type (EGFR-wt) groups,and the relationships between these PET parameters and EGFR-mt were evaluated. Results: A total of 64 patients (median age 66 years, M:F = 34:30) were included in the analysis, and 29 (45%) patients showedEGFR-mt. In EGFR-mt group, all the image parameters of SUVmax, MTV, TLG, and ASP were significantly lower than inEGFR-wt group (all adjusted P< 0.050). In univariable logistic regression, SUVmax (P= 0.003) and ASP (P= 0.010) weresignificant determinants for EGFR-mt, whereasMTV was not (P= 0.690). Multivariate analysis revealed that SUVmax and ASPare independent determinants for EGFR-mt, regardless of inclusion of MTV in the analysis (P< 0.05). Conclusion In Asian NSCLC/ADC patients, SUVmax, MTV, and ASP on FDG PET are significantly related to EGFR mutationstatus. Particularly, low SUVmax and ASP are independent determinants for EGFR-mt.

Purpose: The precise quantification of dopamine transporter (DAT) density on N-(3-[18F]Fluoropropyl)-2β-carbomethoxy-3β-(4-iodophenyl) nortropane positron emission tomography ([18F]FP-CIT PET) imaging is crucial to measure the degree of striatal DAT loss in patients with parkinsonism. The quantitative analysis requires a spatial normalization process based on a template brain. Since the spatial normalization method based on a delayed-phase PET has limited performance, we suggest an early-phase PET-based method and compared its accuracy, referring to the MRI-based approach as a gold standard. Methods: A total of 39 referred patients from the movement disorder clinic who underwent dual-phase [18F]FP-CIT PET and took MRI within 1 year were retrospectively analyzed. The three spatial normalization methods were applied for quantification of [18F]FP-CIT PET-MRI-based anatomical normalization, PET template-based method based on delayed PET, and that based on early PET. The striatal binding ratios (BRs) were compared, and voxelwise paired t tests were implemented between different methods. Results: The early image-based normalization showed concordant patterns of putaminal [18F]FP-CIT binding with an MRI-based method. The BRs of the putamen from the MRI-based approach showed higher agreement with early image- than delayed image-based method as presented by Bland-Altman plots and intraclass correlation coefficients (early image-based, 0.980; delayed image-based, 0.895). The voxelwise test exhibited a smaller volume of significantly different counts in putamen between brains processed by early image and MRI compared to that between delayed image and MRI. Conclusion The early-phase [18F]FP-CIT PET can be utilized for spatial normalization of delayed PET image when the MRI image is unavailable and presents better performance than the delayed template-based method in quantitation of putaminal binding ratio.

Solid pseudopapillary tumor (SPT) of the pancreas is a neoplasm with low malignant potential. It is often challenging to diagnose SPT due to its nonspecific clinical and radiological features, and ­[18F]FDOPA is effective in diagnosing SPT, particularly in differentiating SPT from benign conditions such as splenosis. A 55-year-old woman underwent distal pancreatectomy and splenectomy for histologically confirmed SPT. She was also initially diagnosed with splenosis. During follow-up, sizes of multiple nodular lesions were increased, raising the possibility of peritoneal seeding of SPT. For diagnosis, a spleen scan and SPECT/CT were performed using 99mTc-labeled damaged red blood cells, which showed no uptake in the peritoneal nodules. Subsequent ­[18F]FDOPA PET/CT revealed ­[18F]FDOPA-avidity of the nodules. The patient underwent tumor resection surgery, and the nodules were pathologically confirmed as SPT.

Methods: 150 amyloid brain PET images were visually assessed by experts and categorized as negative and positive. Standardized uptake value ratio (SUVR) was calculated with cerebellum grey matter as the reference region, and receiver operating characteristic (ROC) and precision-recall (PR) analysis for BTXBrain-Amyloid were conducted. For comparison, same image processing and analysis was performed using Statistical Parametric Mapping (SPM) program. In addition, to evaluate the spatial normalization (SN) performance, mutual information (MI) between MRI template and spatially normalized PET images was calculated and SPM group analysis was conducted. Results: Both BTXBrain and SPM methods discriminated between negative and positive groups. However, BTXBrain exhibited lower SUVR standard deviation (0.06 and 0.21 for negative and positive, respectively) than SPM method (0.11 and 0.25). In ROC analysis, BTXBrain had an AUC of 0.979, compared to 0.959 for SPM, while PR curves showed an AUC of 0.983 for BTXBrain and 0.949 for SPM. At the optimal cut-off, the sensitivity and specificity were 0.983 and 0.921 for BTXBrain and 0.917 and 0.921 for SPM12, respectively. MI evaluation also favored BTXBrain (0.848 vs. 0.823), indicating improved SN. In SPM group analysis, BTXBrain exhibited higher sensitivity in detecting basal ganglia differences between negative and positive groups. Conclusion BTXBrain-Amyloid outperformed SPM in clinical performance evaluation, also demonstrating superior SN and improved detection of deep brain differences. These results suggest the potential of BTXBrain-Amyloid as a valuable tool for clinical amyloid PET image evaluation.