Extracting anatomic reference from computed tomography (CT) images is a crucial step in fully automated image analysis solutions for CT and hybrid imaging of prostate cancer. Several deep learning-based applications can be used to perform segmentation of different anatomical structures in CT, but they might produce a false prostate segment for post-treatment scans of patients treated with prostatectomy. In this study, our aim was to both systematically assess the performance of a state-of-the-art CT segmentation tool, TotalSegmentator, for post-prostatectomy patients and to investigate the potential of using a convolutional neural network (CNN) to automatically detect prostatectomy from CT images. We collected a dataset of CT images from 542 patients, 269 of which were treated with robotic-assisted laparoscopic radical prostatectomy (RP), 194 of which were treated with radiation and/or androgen deprivation therapy only, and 79 of which were treatment-naive. We used TotalSegmentator to perform multi-organ segmentation for all the patients, computed the volumes of the greatest connected components of the prostate segments, and studied the use of a cut-off threshold for the resulting volumes to detect RP with five-fold cross-validation. Additionally, we trained and evaluated a light-weight CNN for classifying patients treated with and without RP. According to our results, TotalSegmentator produced a false prostate segment for 98.5 ^3 vs. 22.8 ± 8.9 cm ^3 and 22.4 ± 10.9 cm ^3 , p-values: 1.3e−17 and 2.1e−17). The use of cut-off thresholds for TotalSegmentator’s prostate volumes resulted in RP detection accuracy of 77.5 ± 1.3
We established a theoretical framework to analyze the individual noise components on computed tomography (CT) images and evaluated the relationship between CT doses and each component. The model was developed based on CT numbers and the relative noise standard deviation (SD) measured directly on the CT images, enabling noise decomposition solely from image-domain characteristics without relying on raw projection data. CT images of a quality control phantom were obtained at various CT doses using two CT scanners. We measured mean CT numbers and the relative noise SD at each rod position on the CT images and estimated each noise component using the proposed framework. The relative noise SD was expressed as a function of the transmitted dose corresponding to the number of photons through the phantom, where the respective terms in the formula represent the contributions of structural, quantum, and electronic noise components. Results showed that the proportion of quantum noise exceeded 70
The diagnostic accuracy of medical imaging is critically dependent on the performance of primary diagnostic displays. However, quality assurance (QA) programs for these displays are often limited by geographic dispersion, manual testing requirements, and inconsistent calibration practices. This study evaluates the implementation of a centralised, automated QA program aligned with AAPM Task Group 270 (TG-270) recommendations across a large health service. A four-stage process was used to assess and improve the performance of 53 diagnostic displays. This included baseline QC evaluation, internal photometer and ambient light sensor accuracy testing, implementation of a cloud-based automated QC system, and post-intervention performance testing. Internal photometers were cross-calibrated using a traceable external photometer. Initial testing revealed that 53
Accurate classification of brain tumors from MRI scans requires models capable of capturing both fine-grained structural details and broader contextual patterns. This study proposes a hybrid Wavelet–EfficientNetV2–Vision Transformer (ViT) framework that integrates multi-resolution frequency enhancement, hierarchical spatial encoding, and global dependency modeling into a unified representation. The Discrete Wavelet Transform (DWT) enriches directional and boundary-aware features, EfficientNetV2 extracts robust localized patterns, and the ViT captures long-range contextual relationships essential for differentiating tumor categories with overlapping visual characteristics. The method was evaluated on the Kaggle-MRI dataset consisting of four classes: glioma, meningioma, pituitary tumor, and no tumor. Experimental results show that the proposed model achieves 98.25
This study proposes a new method for evaluating distance accuracy on the reformatted oblique plane using the American College of Radiology (ACR) computed tomography (CT) phantom. Evaluations were performed on oblique images reformatted from axial, sagittal, and coronal images. Oblique images from axial images were reformatted through two radiopaque markers commonly used for distance accuracy on the axial image. Oblique images from sagittal and coronal images were reformatted through eight radiopaque markers at the edge of the phantom usually used to perform centering of the phantom in the x, y, and z directions. Axial images were scanned with two pitches of 0.86 and 1.0. Oblique images were reformatted using IndoQCT and RadiAnt. The distance accuracy was evaluated on the reformatted oblique plane using an electronic caliper available in these two platforms. The percentage differences of measured distances on the reformatted oblique image using IndoQCT were within 2
This paper provides both effortless augmentation of data and efficient creation of images according to the image input size of deep learning models by converting almost all numerical data into 24-bit images of particular standards. As cardiovascular disease is a cause of mortality, artificial intelligence-based architectures may play an important role here, and predicting survival from heart failure is a great challenge. For this purpose, we adjust the image input size according to different deep learning architectures by using the automated feature engineering technique in the approach, which can be easily used in different numerical datasets. Afterward, we horizontally augment the transpose of the new features obtained, thus creating the 8-bit rectangular coded images. After converting these images to 24-bit, we rotate them into multiples of 15° using different coefficients to augment the dataset. We train the ResNet18 and ResNet50 architectures using these images in the new data. According to the findings, the performance metrics (Accuracy, Sensitivity, Specificity, and F1 Score) are quantified as 0.9331–0.9610–0.8742–0.9512 for ResNet18 and 0.9617–0.9743–0.9349–0.9718 for ResNet50. Evaluated alongside current literature, these findings confirm that the proposed methodology represents an innovative, competitive, and highly applicable solution for heart failure classification. The proposed novel method allows numerical data to be evaluated as image representations and, accordingly, to be easily utilized in image processing, data augmentation, and similar techniques.
A dual-panel Positron Emission Mammography (PEM) scanner is a breast-dedicated device with better spatial resolution and sensitivity than conventional Positron Emission Tomography (PET). However, the main limitation of dual-panel PEM is the limited-angle artifacts in the cross-plane slices of the reconstructed images, arising from the acquisition geometry. This work proposes incorporating deep unrolled regularization into the image reconstruction algorithm to mitigate these artifacts. A U-Net was trained to identify and correct the artifacts in cross-plane slices. Dual-panel PEM and ring-shaped dedicated breast PET Monte Carlo (MC)-generated images served as the input and ground truth, respectively; the latter was selected because it provides an artifact-free reference due to its full-angle geometry. Data augmentation was employed to expand the training dataset to 3840 image pairs, thereby improving the model's generalization performance. The trained deep learning model was incorporated into the Forward-Backward Splitting Expectation-Maximization algorithm to regularize reconstruction and ensure agreement between the reconstructed images and the measured data. Images reconstructed with the proposed image reconstruction framework exhibited effective mitigation of limited-angle artifacts using MC-generated and experimentally measured data from a dual-panel PEM. Additionally, this framework outperformed OSEM and MAPEM as measured by metrics quantifying noise, contrast, peak-to-valley ratio, recovery coefficients, spillover ratios, and intensity profiles. To the best of our knowledge, this is the first study to address limited-angle artifacts in a dual-panel PEM incorporating deep learning-based regularization. At this stage, the present work constitutes a proof-of-concept; for clinical implementation, a model trained on anthropomorphic phantoms and further validation are required.
Regulating blood pressure (BP) effectively is vital for maintaining health and ensuring survival, yet elevation and fluctuation in BP can lead to significant health threats. Traditional methods for measuring BP, such as cuff-based and invasive procedures, can be cumbersome and do not allow for continuous measurements. In response to these challenges, our research focuses on enhancing non-invasive BP monitoring by leveraging photoplethysmography (PPG) signals in conjunction with sophisticated machine learning (ML) techniques. Our research analyzed PPG data from a diverse cohort of subjects, ranging in age from 21 to 86, including both individuals in good health and those with underlying health conditions. The analysis involved rigorous preprocessing and feature extraction processes. To enhance computational efficiency and mitigate the risk of overfitting, we applied four distinct feature selection strategies. The features identified by each method were then utilized in five ML classification models using k-fold cross validation to differentiate BP across four categories. Our findings indicate that the ensemble-based extra trees classifier (ETC) model, combined with the SelectFromModel feature selection approach, achieved remarkable accuracy of 91.13
Compared to photon radiotherapy (RT), proton RT is less widely available and more costly. To maximize the normal-tissue complication probability (NTCP) benefit of limited proton resources at a population level, this work introduces a novel NTCP-optimized combined proton–photon treatment (NTCP-CPPT) approach that can achieve the OAR-sparing benefits in terms of reaching desired NTCP thresholds with the minimal proton fraction. NTCP-optimized intensity-modulated proton therapy (NTCP-IMPT) and NTCP-optimized intensity-modulated radiation therapy (NTCP-IMRT) plans are generated via the NTCP-optimized treatment planning based on physical dose and NTCP objectives. For each patient, a minimal proton fraction is determined for the NTCP-CPPT plan as a combination of NTCP-IMPT and NTCP-IMRT plans, while still retaining the sum-NTCP threshold for the patient to be justified for the proton access. For comparison, conventional CPPT (CONV-CPPT) plans are obtained by a combination of conventional IMRT (CONV-IMRT) and conventional IMPT (CONV-IMPT) plans that are solely optimized based on physical dose optimization. Compared to CONV-CPPT, NTCP-CPPT reduced the average proton fractions from 18 to 3 without compromising physical dose objectives, while both NTCP-CPPT and CONV-CPPT plans maintained a sum-NTCP below 20
Hypofractionated stereotactic body radiotherapy (SBRT) for pancreatic cancer has shown improved local control but is challenging due to the proximity of several organs at risk (OARs). This study investigated possible dosimetric variations over a simulated SBRT course for pancreatic cancer using daily MR images considering breath hold (BH) and internal target volume (ITV) motion management techniques for three treatment regimes. Ten healthy volunteers were scanned on a Siemens Skyra 3T MRI. Each volunteer had 2 sets of scans acquired per day, 3 h apart, over five days, to represent a daily simulation planning scan and a pre-treatment scan. A hypothetical GTV was generated, with a BH GTV and ITV expanded to generate two different planning target volumes (PTV). Treatment plans for each were generated using the Raystation TPS on each bulk density corrected scan. Plans were propagated and accumulated to simulate conventional, daily adaptive and online adaptive treatment regimes, with DVH comparison of PTV and OAR doses. The BH PTV D95
Healthcare workers are often exposed to external radiation from patients who have been administered radionuclides for nuclear medicine (NM) studies, particularly during subsequent procedures requiring close contact. Traditional dose rate estimates commonly assume patients to be unattenuated point sources and apply inverse square law calculations, which tend to overestimate exposures at clinical distances. This study presents an experimental approach to quantify dose rates more accurately from NM patients using point, line, and phantom source models of Tc-99m, F-18, I-131, and Lu-177. Measurements were performed with calibrated ionisation chambers under controlled low-background conditions. Radionuclide activities were selected to achieve detectable dose rates while minimising staff exposure. Point and line source measurements provided baseline external exposure data across 10–100 cm. Phantom models, including a homogeneous water-filled NEMA body phantom and a PMMA neck phantom, were used to replicate clinical patient geometries. Dose rates were decay-corrected, normalised to activity, and fitted with biexponential models to generate exposure curves. Gamma factors (GFs) were evaluated at 0.3 m and 1.0 m from the centre of the radioactive source and compared with internationally recognised literature values. Self-absorption factors (SFs) were derived by comparing phantom and point source curves across clinical distances. Measured GFs showed excellent agreement with published data and internationally recognised standards, differing by only 2
Ultrasound face phantoms are essential tools for cosmetic procedures, providing a realistic and anatomically accurate representation of the human face for training and practice. This research aims to create an ultrasound face phantom to simulate the skin, muscles, arteries and other structures of the face, allowing medical professionals to accurately assess and measure the impact of various treatments and aesthetic procedures using ultrasound imaging. The phantom was made using polyvinyl chloride mixed with additives. The ingredients were chosen to accurately replicate the physical and ultrasound properties of human tissues. To ensure alignment with the biomechanical and ultrasound characteristics of human facial tissues, compressive, tensile, and acoustic measurements were conducted. The ultrasound evaluation of the designed phantom was performed using a carrier frequency of up to 20 MHz. The phantom was 155 × 117 × 120 mm3 in size, weighed 1.1 kg and consisted of the models of a skull, muscle and soft tissues, salivary gland, lymph nodules, and blood vessels. The ultrasound images of the phantom closely mimicked those of the human face, allowing for accurate and realistic simulations. The tissue-mimicking materials showed a Young’s modulus ranging from 31 to 35 kPa through compression, a Young’s modulus of 80 to 170 kPa through tension, an ultrasound speed of 1440 to 1481 m/s, and an attenuation coefficient of 0.21 to 0.58 dB/cm/MHz, which were consistent with the properties of subcutaneous tissues. The designed face phantom allows for more accurate and realistic training in ultrasound-guided cosmetology. This study successfully developed a highly realistic ultrasound facial phantom with precise mechanical and acoustic properties. These phantoms can serve as a practical tool for training specialists in both diagnostic and surgical procedures, enabling simulations of injections, biopsies, and other minimally invasive interventions with accuracy close to clinical conditions.
The Internet of Things (IoT) has provided a lot of support for patient care by connecting technologies and clinicians through a high-speed network. Data management in IoT healthcare requires a lot of support, since most of the networks are designed for continuous monitoring through the sensor devices, and this leads to challenges in managing sensor data. As the resources in IoT are limited, it is necessary to explore data management strategies so as to achieve an energy-efficient IoT healthcare setup. The classical transmission networks implemented for human-aided applications have various problems, such as limited battery life and resources. Therefore, an effective healthcare data management system using IoT-aided wearable sensor devices is implemented in this work to address the challenges encountered in traditional models. To facilitate robust communication for an IoT-assisted healthcare device, a cloud layer is established, which is used for health data collection and transmission and health monitoring. Here, optimal resource allocation is done using the proposed Hybrid Tasmanian Devil Archery Algorithm (HTDAA), and it is developed from the existing concepts of Tasmanian Devil Optimization (TDO) and Archery Algorithm (AA) to achieve high multi-objective constraint balance and convergence power. This optimal resource allocation process helps in reducing energy consumption in the usual data transmission. For an effective resource allocation process, an objective function is formulated based on the normalized energy ratio value, the buffer memory ratio, and the packet arrival data rate. The proposed HTDAA is established in the IoT setup and used for allocating the optimal channel resources to be involved in the health data transmission, and minimizes the energy requirement. The performance of the suggested HTDAA is estimated by comparing its numerical outcomes with recent optimization techniques. According to the analysis, the packet arrival ratio of HTDAA is 5.6
Radiochromic films are widely used in radiotherapy dosimetry for their high spatial resolution and near-water-equivalence. Nonetheless, these films are frequently purchased in bulk, leading clinical departments to accumulate stock beyond the manufacturer’s expiration date. Assessing the dosimetric performance of expired films may confer economic and environmental advantages. This work evaluated whether expired EBT-XD radiochromic films remain suitable for specific radiotherapy dosimetry applications. Two expired batches were compared with a non-expired batch. Dose–response curves were generated using a 6 MV photon beam from a clinical linear accelerator. The films were tested for small-field output factor measurements and patient-specific quality assurance (PSQA) in volumetric-modulated arc therapy (VMAT) and stereotactic radiosurgery (SRS), employing gamma-index analysis. Results indicated that expired films exhibit reduced sensitivity relative to unexpired films, particularly at lower dose levels. However, acceptable dosimetric agreement with unexpired films was observed for small-field output factors and high-dose SRS PSQA cases, provided appropriate recalibration was applied. More pronounced discrepancies emerged in low-dose VMAT PSQA measurements, where diminished film sensitivity may have influenced the film’s response. These findings suggest that expired EBT-XD radiochromic films may still provide useful dosimetric information as relative dosimeters for selected radiotherapy measurements, especially when measurements fall within the film’s recommended dose range, and films are properly stored and recalibrated.
Knowledge-based planning (KBP) can improve the efficiency and consistency of volumetric-modulated arc therapy (VMAT); however, the extent to which treatment-machine characteristics influence KBP model behavior remains unclear. This study aimed to clarify the dosimetric impact of machine-specific KBP models by distinguishing the contributions of model libraries and plan-generation environment. Three RapidPlan models were independently trained using 30 manually generated plans created on NovalisTx (15X), TrueBeam (10X), and Halcyon (6FFF). The models were applied to an independent cohort of 15 patients using two complementary evaluation approaches: a clinical evaluation employing each machine’s native beam model and a controlled evaluation using a common TrueBeam 10X beam model with identical optimization settings. Dosimetric endpoints included planning target volume excluding the rectum (PTV-R) D2
CyberKnife stereotactic radiosurgery (SRS) treatment of multiple brain metastases (MBM) using fixed cone collimation is well established. This study aims to rigorously evaluate the commissioning of MBM SRS using feasible complex clinical scenarios. Sixteen gross tumour volumes (GTVs) were contoured on the CT of an in-house developed phantom. Six treatment plans, testing different levels of complexity, were generated using the Ray Tracing (RT) dose calculation algorithm and recalculated with the Monte Carlo (MC) algorithm. Respective quality assurance (QA) plans were generated on the same phantom and delivered on a CyberKnife M6 system. QA was performed with EBT4 film inserted in the GTVs under investigation. The films were scanned and analysed using in-house developed software for comparison with RT and MC calculated dose distributions in the film planes. For simple cases, RT and MC performed equally well. As the number and proximity of GTVs increased, with complex GTV geometrical distribution and multiple prescription dose levels (MPDLs), the plan gamma pass rate at 3
Dosimetry audits are a key component of radiotherapy quality improvement, however lack of appropriate tools and resources may be hampering the adoption of audit practices for brachytherapy in particular. In this study, an audit method was developed around a 3D printed reference air kerma rate (RAKR) measurement jig and a pilot high dose rate (HDR) postal audit was completed across eighteen HDR centres across Australia and New Zealand. This pilot study allowed the performance of the 3D printed jig and accompanying audit spreadsheet to be assessed, to inform the design of future iterations of the 3D printed jig tool and provide baseline data for future brachytherapy audits.
This study aimed to automate and standardize treatment planning by comparing the dosimetric performance of conventional volumetric modulated arc therapy (C-VMAT) with that of HyperArc optimized using the knowledge-based RapidPlan model trained with the HyperArc plan (RP-HA) for total scalp irradiation (TSI) in the treatment of angiosarcoma of the scalp (AS). Additionally, this study compared the dosimetric performance of RP-HA with that of non-coplanar VMAT optimized using the same model (RP-NC-VMAT). From 40 patients with AS who had undergone TSI, 20 patients were selected for HyperArc planning and RapidPlan model training. The remaining 20 patients had treatment plans generated using three different methods (C-VMAT, RP-NC-VMAT, and RP-HA). The prescription doses were 70 and 56 Gy in 35 fractions of the target volumes, including gross tumor and whole scalp, respectively, using the simultaneous integrated boost technique. The dose distribution, dosimetric parameters, and dosimetric accuracy of each treatment plan were evaluated and compared between methods. None of the three methods exceeded the acceptable limits for all constraint parameters of the target and organs at risk. The RP-HA plan provided a significantly lower mean brain dose (10.96 ± 1.54 Gy) than the C-VMAT and RP-NC-VMAT plans (18.06 ± 2.17 Gy and 12.36 ± 2.12 Gy). The dose received by 0.1 cm3 of the hippocampus was significantly lower in the RP-NC-VMAT and RP-HA plans than in the C-VMAT plan. The RapidPlan model trained with the HyperArc plan can be helpful in automating and standardizing treatment planning in TSI for AS.
Ophthalmology diseases are among the leading causes of vision loss worldwide. Glaucoma, diabetic retinopathy, and cataracts are the most common diseases and can lead to permanent vision loss if left untreated. In this paper, a new hybrid model has been proposed with the methods accepted in the literature used in the early diagnosis of these diseases. The relationships between imaging analyses and clinical evaluations performed in the diagnostic processes of glaucoma, diabetic retinopathy, and cataract are discussed, and the methods that help to identify diseases in the early stages are emphasized. In addition, the contributions of advanced technologies and imaging systems used in diagnosing these diseases to the developments in the field of eye health are discussed. This article proposes a hybrid model for eye disease detection that combines Convolutional Neural Networks (CNNs) with a metaheuristic optimization algorithm. This model uses ShuffleNet and ResNet101 models as feature extractors, while the Secretary Bird Optimization Algorithm (SBOA) is used for feature selection. Then, the extracted feature maps were combined with ShuffleNet and ResNet101 and optimized with SBOA. The feature fusion process aimed to improve the performance of the developed model by combining different features of the same image. The combined feature map optimized with SBOA was classified into six different classifiers so that the model could work faster and more effectively. Competitive results were produced in the developed model. Finally, explainable artificial intelligence methods were used to visualize the decisions of the developed hybrid model and understand the internal working principle of the model.
