Radiological Physics and Technology最新文献

Cone-beam computed tomography-based online adaptive radiotherapy (CBCT-based online ART) is currently used in clinical practice; however, deep learning-based segmentation of CBCT images remains challenging. Previous studies generated CBCT datasets for segmentation by adding contours outside clinical practice or synthesizing tissue contrast-enhanced diagnostic images paired with CBCT images. This study aimed to improve CBCT segmentation by matching the treatment planning CT (tpCT) image quality to CBCT images without altering the tpCT image or its contours. A deep-learning-based CBCT segmentation model was trained for the male pelvis using only the tpCT dataset. To bridge the quality gap between tpCT and routine CBCT images, an artificial pseudo-CBCT dataset was generated using Gaussian noise and Fourier domain adaptation (FDA) for 80 tpCT datasets (the hybrid FDA method). A five-fold cross-validation approach was used for model training. For comparison, atlas-based segmentation was performed with a registered tpCT dataset. The Dice similarity coefficient (DSC) assessed contour quality between the model-predicted and reference manual contours. The average DSC values for the clinical target volume, bladder, and rectum using the hybrid FDA method were 0.71 ± 0.08, 0.84 ± 0.08, and 0.78 ± 0.06, respectively. Conversely, the values for the model using plain tpCT were 0.40 ± 0.12, 0.17 ± 0.21, and 0.18 ± 0.14, and for the atlas-based model were 0.66 ± 0.13, 0.59 ± 0.16, and 0.66 ± 0.11, respectively. The segmentation model using the hybrid FDA method demonstrated significantly higher accuracy than models trained on plain tpCT datasets and those using atlas-based segmentation.

This study evaluates the dosimetric impact of material and mass density misassignments in Acuros XB dose calculations using phantom simulations and clinical analysis in Eclipse TPS. The phantom study analyzed material and mass density misassignments in Acuros XB using virtual phantoms with a central insert assigned different materials and mass densities to simulate misassignment. A clinical analysis of 270 patient CT scans from three scanners assessed HU variations in sinonasal cavities, bladder, and liver. Dosimetric deviations were examined in 96 radiotherapy patients across these anatomical sites by comparing automatic and manual material assignments, with dose differences assessed using D_98%, D_mean, and D_2% for target volumes and misclassified structures. Material misassignment caused substantial dose differences, particularly in air-lung and cartilage-bone misassignments, with 12.1% and 2.8% deviations, respectively. Mass density misassignments led to dose variations of up to 5.5% for lung-air and 2% for bone. Combined misassignments amplified differences, reaching 18% for air-lung and 5.5% for cartilage-bone. Misassignment of non-biological materials such as biological tissues resulted in dose differences from 1 to 26.5%. Clinical analysis showed HU variations frequently led to material misassignment. Sinonasal air cavities were misclassified as lung, causing dose deviations of 11.8% for D_98%, 8.6% for D_mean, and 2.6% for D_2%. Bladder and liver were predominantly misclassified as muscle and cartilage, respectively, resulting in systematic dose deviations of approximately 1% and 0.5%. Accurate material assignment is critical for precise Acuros XB dose calculations. Material mischaracterization introduces significant dose differences, necessitating manual verification in cases where auto-assignment is prone to misassignment.

This study used AAPM Report 293 to estimate absorbed doses (EADs) from head and neck CT examinations in 100 patients (ages 18-90), leveraging its superior accuracy over Report 220. The study used data from a diagnostic CT scanner of model Canon Aquilion Lightning 16-row 32-slice, between years 2022 to 2024, with IndoseCT version 20b software to extract parameters, such as volumetric CT dose index (CTDI_Vol), dose length product (DLP), X-ray tube current (mAs), X-ray tube voltage (kVp) and Size Specific Dose Estimation (SSDE) based on water equivalent diameter (D_w). Two IndoseCT methods were employed: Z-axis Slice Number technique for D_w and Z-axis Slice range technique for EAD. A correlation analysis using Pearson Correlation Coefficient (PCC) and Principal Component Analysis (PCA) investigated relationships between CT dose and patient size parameters. The analysis was conducted using Microsoft Excel software embedded with XLSTAT 2024 version. Results showed CTDIvol values were higher than SSDE in most age groups, except 40-59 and 70-79 years. EADs ranged from 29.21 ± 8.12 mGy for (18-30) to 33.07 ± 5.81 mGy for (≥ 30) age groups. Conversion factors (CF) were varied, with notable impact from the 70-79 age group. this study found similar trends in CTDIvol to SSDE conversion factors (CF) as past work, with a mean CF < 1 indicating slight underestimation of radiation dose (SSDE). Notably, by including the 70-79 age group, CF can be shifted from < 1 to > 1; this suggests that patient size in the 70-79 age group may require protocol optimization.

In recent years, generative AI has attracted significant public attention, and its use has been rapidly expanding across a wide range of domains. From creative tasks such as text summarization, idea generation, and source code generation, to the streamlining of medical support tasks like diagnostic report generation and summarization, AI is now deeply involved in many areas. Today's breadth of AI applications is clearly distinct from what was seen before generative AI gained widespread recognition. Representative generative AI services include DALL·E 3 (OpenAI, California, USA) and Stable Diffusion (Stability AI, London, England, UK) for image generation, ChatGPT (OpenAI, California, USA), and Gemini (Google, California, USA) for text generation. The rise of generative AI has been influenced by advances in deep learning models and the scaling up of data, models, and computational resources based on the Scaling Laws. Moreover, the emergence of foundation models, which are trained on large-scale datasets and possess general-purpose knowledge applicable to various downstream tasks, is creating a new paradigm in AI development. These shifts brought about by generative AI and foundation models also profoundly impact medical image processing, fundamentally changing the framework for AI development in healthcare. This paper provides an overview of diffusion models used in image generation AI and large language models (LLMs) used in text generation AI, and introduces their applications in medical support. This paper also discusses foundation models, which are gaining attention alongside generative AI, including their construction methods and applications in the medical field. Finally, the paper explores how to develop foundation models and high-performance AI for medical support by fully utilizing national data and computational resources.

This study evaluated whether temporal changes from the dynamic late phase to the hepatobiliary phase using gadolinium-ethoxybenzyl-diethylenetriamine penta-acetic acid supported liver tumor classification. A total of 147 patients with 165 lesions (hepatocellular carcinoma [HCC], metastases, and hemangiomas) underwent 3.0 T MRI. Quantitative liver lesion contrast (Q-LLC) and its rate of change (%) at 3, 10, and 15 min postcontrast were analyzed. Tumors were stratified according to albumin-bilirubin (ALBI) grade. For ALBI grade 1, the Q-LLC significantly differed at 3 min and 10 min (P < 0.05). Q-LLC increased over time, and the rate of change was the lowest in HCC, followed by metastases, and was the highest in hemangiomas. Significant differences in the rate of change were observed among tumor types for ALBI grade 1 (P < 0.01). These findings suggest that the Q-LLC rate of change may aid in liver tumor classification, particularly in patients with preserved liver function.

In acute ischemic stroke (AIS), where the shortest possible assessment is required to minimize time to mechanical thrombectomy (MT). With recent advancements in MRI reconstruction technology, MRI has also become valuable in the decision-making process for AIS treatment planning. In this study, we compared the examination times of our MRI protocol with those of a standard CT protocol for evaluating AIS through phantom simulations to obtain timing information directly relevant to treatment strategies, and evaluated the utility of MRI for MT. Ten radiological technologists performed scans using the same phantom for each modality. Evaluation items included time for hemorrhage detection, time for penumbra evaluation, and time for brain artery evaluation, and total examination time. The total examination time was slightly shorter with CT (696.2 ± 52.7 s) compared to MRI (701.8 ± 15.8 s), although this difference was not statistically significant (p = 0.4). For other parameters, MRI demonstrated significantly faster detection times: hemorrhage detection (CT, 80.9 ± 12.8 s; MRI, 66.3 ± 1.7 s; p = 0.0002), penumbra evaluation (CT, 696.2 ± 52.7 s; MRI, 262.1 ± 9.3 s; p = 0.0002), and brain artery evaluation (CT, 592.1 ± 32.3 s; MRI, 367.8 ± 8.3 s; p = 0.0002). The coefficient of variation (CV) was lower for MRI compared to CT, indicating less variability in examination times with MRI. This study demonstrates that MRI protocols, including perfusion imaging, can more rapidly visualize factors essential for MT decision-making and do not delay time to MT.

Positron emission tomography (PET) measurements in the cerebrospinal fluid (CSF) region may be overestimated because of spillover artifacts from surrounding radioactivity. In this study, we proposed a simple spillover correction method (subtraction method) and evaluated its validity. A cylindrical phantom simulating brain ventricles was used to compare the subtraction method with the geometric transfer matrix (GTM) correction approach. And the subtraction method was applied to dynamic PET images of [¹⁸F]fluorodeoxyglucose (FDG), [¹⁸F]fluorodopa (FDOPA), and [¹¹C]raclopride (RAC), and [¹⁵O]H₂O (H₂O). The effects of spillover correction on CSF measurements were assessed. Both methods effectively reduced spillover artifacts in the phantom study. In dynamic PET images, after spillover correction, time-activity curves for FDG, FDOPA, and RAC approached near-zero levels in the CSF, whereas H₂O continued to show increasing activity over time. This approach effectively reduces artifacts and offers the advantages of simpler volume-of-interest settings and straightforward calculation procedures.

In plain chest radiography (CXR), automatic exposure control (AEC) is generally used to standardize image quality. In contrast, dynamic chest radiography (DCR) systems preliminarily require manual setting of tube current-time products (mAs value). Body mass index (BMI) of patients is one of the indexes to determine the mAs value; however, standardization is limited because the anatomical differences are not considered. In this study, for further standardization, we propose a practical procedure to determine the individual mAs value of DCR using data obtained from CXR. To evaluate its effectiveness, we retrospectively analyzed 97 patients who underwent both CXR and DCR on the same day. DCR was performed in the following procedures: (1) obtain the relationship between the mAs value and the exposure indicator (S value, Konica Minolta, Inc.) obtained in CXR acquisition, (2) calculate the mAs value of DCR for the target S value of 2500, and (3) record the S value in DCR. The tube voltages for CXR and DCR were set to 120 kV and 100 kV with a copper filter, respectively. The differences in exposure doses were corrected by measuring the air kerma using a CdTe detector. As a result, the S values of CXR and DCR were 133 ± 13 (Coefficient of Variation (CV) = 9.9%) and 2629 ± 207 (CV = 7.9%), respectively, which were not dependent on the patient size based on evaluating the S values of five classified BMI groups. In conclusion, our proposed procedure enables standardization of the image quality in DCR by optimizing the patient-specific exposure conditions.

In ²⁰¹Tl myocardial perfusion single-photon emission computed tomography (SPECT), gastric wall uptake can impact the inferior wall. This study aimed to evaluate the effectiveness and usefulness of the masking on un-smoothed image (MUS) method for ²⁰¹Tl myocardial perfusion SPECT. A hemispherical gastric wall phantom was created to simulate the gastric fundus located closest to the myocardium, and the activity was enclosed to achieve an SPECT count ratio against the myocardium equivalent to that observed in clinical practice. The minimum values of the defect chip in the circumferential profile curve were compared for six SPECT count ratios and seven gap distances. In the conventional method, increasing SPECT count ratios or gap distances interfered with myocardial perfusion SPECT evaluation. Artifacts were less apparent when the MUS method was applied. The MUS method effectively suppressed the gastric wall uptake on ²⁰¹Tl myocardial perfusion SPECT.

We aimed to develop a technique to precisely measure the shrinkage of immobilization sheets (ISs) using a three-dimensional (3D) tracking of multiple points on IS using a near-infrared camera. A thermoplastic sheet and an elastomer sheet were used in this study. The inter-marker distance (IMD) of neighboring marker pairs and the triangular area (TA) formed by neighboring three markers were analyzed as a function of time since molding each IS. Thermal distance ratio (TDR), IMD normalized to IMD after 48 h, and thermal area ratio (TAR), TA normalized to TA after 48 h, were analyzed using an exponential function. The 3D visualization of the initial shrinkage amplitude (ISA) was created for each IS. The mean ISA and the time constant (TC) in the exponential function (ISA, TC) for horizontal and vertical pairs were (0.34 $\pm$ 0.06, 3.9 $\pm$ 0.9) and (0.60 $\pm$ 0.05, 7.6 $\pm$ 0.9) for HFT and (0.13 $\pm$ 0.02, 25.0 $\pm$ 4.7) and (0.06 $\pm$ 0.03, 4.7 $\pm$ 3.5) for SF, respectively. The mean (ISA, TC) for HFT and SF were (0.76 $\pm$ 0.07, 7.1 $\pm$ 0.9) and (0.22 $\pm$ 0.03, 17.7 $\pm$ 4.2), respectively. Horizontal pairs showed smaller ISA and shorter TC than vertical pairs for HFT, while horizontal pairs showed larger ISA and longer TC than vertical pairs for SF, possibly due to different chemical characteristics of each material under the effect of mechanical force. The mean TDR and TAR are considered useful for evaluating the gross property of IS. The visualized distributions of ISA are considered useful to provide spatial information for investigating relationships between actual handlings and shrinkage of IS.