Since its introduction in 2015, the U-Net architecture used in Deep Learning has played a crucial role in medical imaging. Recognized for its ability to accurately discriminate small structures, the U-Net has received more than 2600 citations in academic literature, which motivated continuous enhancements to its architecture. In hospitals, chest radiography is the primary diagnostic method for pulmonary disorders, however, accurate lung segmentation in chest X-ray images remains a challenging task, primarily due to the significant variations in lung shapes and the presence of intense opacities caused by various diseases. This article introduces a new approach for the segmentation of lung X-ray images. Traditional max-pooling operations, commonly employed in conventional U-Net++ models, were replaced with the discrete wavelet transform (DWT), offering a more accurate down-sampling technique that potentially captures detailed features of lung structures. Additionally, we used attention gate (AG) mechanisms that enable the model to focus on specific regions in the input image, which improves the accuracy of the segmentation process. When compared with current techniques like Atrous Convolutions, Improved FCN, Improved SegNet, U-Net, and U-Net++, our method (U-Net++-DWT) showed remarkable efficacy, particularly on the Japanese Society of Radiological Technology dataset, achieving an accuracy of 99.1%, specificity of 98.9%, sensitivity of 97.8%, Dice Coefficient of 97.2%, and Jaccard Index of 96.3%. Its performance on the Montgomery County dataset further demonstrated its consistent effectiveness. Moreover, when applied to additional datasets of Chest X-ray Masks and Labels and COVID-19, our method maintained high performance levels, achieving up to 99.3% accuracy, thereby underscoring its adaptability and potential for broad applications in medical imaging diagnostics.

Positron emission tomography (PET) using F-FDG is a well-known modality for the diagnosis of various diseases in patients of different ages, sexes, and states of health, which implies that internal radiation dosimetry is highly desired for different phantom anatomies. In this study, we validate "DoseCalcs," a new Monte Carlo platform that combines personalized internal dosimetry calculations with Monte Carlo simulations. To achieve that, we used the specific absorbed fraction (SAF) calculated by DoseCalcs and those from ICRP publication 133 to estimate the absorbed dose per injected activity (AD/IA) and effective dose per injected activity (ED/IA) for F-FDG. The investigation focused on various voxelized phantoms representing different age groups, including adult male and female, and pediatric phantoms of various ages, from newborn to 15 years old. Using the DoseCalcs Monte Carlo platform, we have simulated the emission of F-FDG positrons based on the energy spectrum provided in ICRP publication 107. The results demonstrated the impact of anatomical differences and different organ/tissue compositions on radiation absorption, with significant variations in the AD/IA across different phantoms. Interestingly, organs/tissues near the emission source showed higher AD/IA, highlighting the anatomical dependence on the phantom. When our results were compared to established reference data, especially from ICRP128, most organs/tissues had good agreement. Still, some cases have shown differences. This shows how important it is to use accurate radionuclide data and biokinetic modeling in internal dosimetry calculations. Furthermore, we compared AD/IA and ED/IA values calculated in newborns by DoseCalcs with those derived from alternative codes, MCNP and EGSnrc. While the results generally exhibited consistency, subtle variations underscored the influence of biokinetics modeling choices and computational methodologies. Overall, this research contributes valuable insights into the precision of internal dosimetry calculations using "DoseCalcs-Gui" by providing one platform for Monte Carlo simulation and personalized internal dosimetry in nuclear medicine. The DoseCalcs platform is free for research and available for download at www.github.com/TarikEl/DoseCalcs-Gui .

Variability in plan quality of radiotherapy is commonly attributed to the planner's skill rather than technological parameters. While experienced planners can set reasonable parameters before optimization, less experienced planners face challenges. This study aimed to assess the quality of volumetric-modulated arc therapy (VMAT) in patients with left-sided breast cancer following breast-conserving surgery. Twenty-eight patients requiring whole-breast irradiation were randomly selected for inclusion. Each patient underwent two VMAT treatment plans: one optimized by an experienced planner (VMAT-EXP group) and the other designed by a less experienced planner using feasibility dose-volume histogram (FDVH) parameters from PlanIQ (VMAT-FDVH group). Both plans aimed to deliver a prescription dose of 50 Gy in 25 fractions to the planning target volume (PTV). Dosimetry parameters for the PTV and organs at risk (OARs) were compared between the two groups. Both the VMAT-EXP and VMAT-FDVH groups met the clinical plan goals for PTV and OARs. VMAT-FDVH demonstrated a PTV coverage and homogeneity comparable to those of VMAT-EXP. Compared to VMAT-EXP plans, VMAT-FDVH plans resulted in a significant reduction in the mean ipsilateral lung dose, with an average decrease of 0.9 Gy (8.5 Gy vs. 7.6 Gy, P < 0.001). The V5Gy and V20Gy of the ipsilateral lung were also reduced by 3.2% and 1.8%, respectively. Minor differences were observed in the heart, contralateral lung, breast, and liver. Personalized objectives derived from the feasibility DVH tool facilitated the generation of acceptable VMAT plans. Less experienced planners achieved lower doses to the ipsilateral lung while maintaining adequate target coverage and homogeneity. These findings suggest the potential for the effective use of VMAT in in patients with left-sided breast cancer following breast-conserving surgery, especially when guided by feasibility DVH parameters.

The shielding of computed tomography (CT) suites has commonly relied on the assumption that the primary beam has the same beam quality and thus penetrability as the scattered radiation. This report expands on a preliminary work that showed scattered radiation from patients having an overall reduced beam quality, with published transmission data for 120 kVp and 140 kVp through lead. Beam quality data of patient scatter spectra for 80 kVp and 100 kVp are uniquely provided herein using the same methodology, expanding the diagnostic energy range. The mean energy of scatter radiation spectra across this 80-140 kVp diagnostic range was found to have a reduction of 13.4-17.9% compared to a primary beam with a defined 9.8 mm Al added filtration. A DOSXYZnrc Monte Carlo program using the EGSnrc photon and electron transport code was subsequently used to simulate the transmission of scattered spectra of all 80, 100, 120 and 140 kVp beams through various commonly used shielding materials, including lead, concrete, steel, plate glass and gypsum wallboard. Transmission data and Archer fitting coefficients for this scattered radiation were calculated and show a reduction in transmission over the range of practical shielding thicknesses for these materials. Transmission through lead was significantly reduced in comparison to the National Council of Radiological Protection (NCRP) and British Institute of Radiology (BIR) methodologies using primary beam spectra, with transmissions reduced between 40.4 and 63.9% for 120 kVp and between 38.1 and 42% for 140 kVp beams over a 0.44-2.64 mm thickness range. The use of CT scatter spectra and their resultant transmission data is recommended for optimal shielding design.

Humans express their feelings and intentions of their actions or communication through emotions. Recent advancements in technology involve machines in human communication in day-to-day life. Thus, understanding of human emotions by machines will be very helpful in assisting the user in a far better way. Various physiological and non-physiological signals can be used to make the machines to recognize the emotion of a person. The identification of emotional content in the signals is crucial to understand emotion and the machines act with emotional intelligence at appropriate times, thus providing a better human machine interaction with emotion identification system and mental health monitoring for psychiatric patients. This work includes the creation of an emotion EEG dataset, the development of an algorithm for identifying the emotion elicitation segments in the EEG signal, and the classification of emotions from EEG signals. The EEG signals are divided into 3s segments, and the segments with emotional content are selected based on the decrease in correlation between the frontal electrodes. The selected segments are validated with the facial expressions of the subjects in the appropriate time segments of the face video. EEGNet is used to classify the emotion from the EEG signal. The classification accuracy with the selected emotional EEG segments is higher compared to the accuracy using all the EEG segments. In subject-specific classification, an average accuracy of 80.87% is obtained from the network trained with selected EEG segments, and 70.5% is obtained from training with all EEG segments. In subject-independent classification, the accuracy of classification is 67% and 63.8% with and without segment selection, respectively. The proposed method of selection of EEG segments is validated using the DEAP dataset, and classification accuracies and F1-scores of subject dependent and subject-independent methods are presented.

Simulating the outcome of double eyelid surgery is a challenging task. Many existing approaches rely on complex and time-consuming 3D digital models to reconstruct facial features for simulating facial plastic surgery outcomes. Some recent research performed a simple affine transformation approach based on 2D images to simulate double eyelid surgery outcomes. However, these methods have faced challenges, such as generating unnatural simulation outcomes and requiring manual removal of masks from images. To address these issues, we have pioneered the use of an unsupervised generative model to generate post-operative double eyelid images. Firstly, we created a dataset involving pre- and post-operative 2D images of double eyelid surgery. Secondly, we proposed a novel attention-class activation map module, which was embedded in a generative adversarial model to facilitate translating a single eyelid image to a double eyelid image. This innovative module enables the generator to selectively focus on the eyelid region that differentiates between the source and target domain, while enhancing the discriminator's ability to discern differences between real and generated images. Finally, we have adjusted the adversarial consistency loss to guide the generator in preserving essential features from the source image and eliminating any masks when generating the double eyelid image. Experimental results have demonstrated the superiority of our approach over existing state-of-the-art techniques.

Increasingly, interventional thoracic workflows utilize cone-beam CT (CBCT) to improve navigational and diagnostic yield. Here, we investigate the feasibility of implementing free-breathing 4D respiratory CBCT for motion mitigated imaging in patients unable to perform a breath-hold or without suspending mechanical ventilation during thoracic interventions. Circular 4D respiratory CBCT imaging trajectories were implemented on a clinical robotic CBCT system using additional real-time control hardware. The circular trajectories consisted of 1 × 360° circle at 0° tilt with fixed gantry velocities of 2°/s, 10°/s, and 20°/s. The imaging target was an in-house developed anthropomorphic breathing thorax phantom with deformable lungs and 3D-printed imaging targets. The phantom was programmed to reproduce 3 patient-measured breathing traces. Following image acquisition, projections were retrospectively binned into ten respiratory phases and reconstructed using filtered back projection, model-based, and iterative motion compensated algorithms. A conventional circular acquisition on the system of the free-breathing phantom was used as comparator. Edge Response Width (ERW) of the imaging target boundaries and Contrast-to-Noise Ratio (CNR) were used for image quality quantification. All acquisitions across all traces considered displayed visual evidence of motion blurring, and this was reflected in the quantitative measurements. Additionally, all the 4D respiratory acquisitions displayed a lower contrast compared to the conventional acquisitions for all three traces considered. Overall, the current implementation of 4D respiratory CBCT explored in this study with various gantry velocities combined with motion compensated algorithms improved image sharpness for the slower gantry rotations considered (2°/s and 10°/s) compared to conventional acquisitions over a variety of patient traces.

Our lab has been developing a deuterium-deuterium (DD) neutron generator-based neutron activation analysis (NAA) system to quantify metals and elements in the human body in vivo. The system has been used to quantify metals such as manganese, aluminum, sodium in bones of a living human. The technology provides a useful way to assess metal exposure and to estimate elemental deposition, storage and biokinetics. It has great potential to be applied in the occupational and environmental health fields to study the association of metal exposure and various health outcomes, as well as in the nutrition field to study the intake of essential elements and human health. However, the relatively low sensitivity of the system has greatly limited its applications. Neutron moderation plays an important role in designing an IVNAA facility, as it affects thermal neutron flux in irradiation cave and radiation exposure to the human subject. This study aims to develop a novel thermal neutron enhancement method to improve the sensitivity of the in vivo neutron activation analysis (IVNAA) system for elemental measurement but still maintain radiation dose. Utilizing a compact DD neutron source, we propose a new and practical moderator design that combines high density polyethylene with heavy water to enhance thermal neutrons by reducing thermal neutron absorption. All material dimensions are calculated by PHITS, a general-purpose Monte Carlo simulation program. The improvement of the new design predicted by the Monte Carlo simulation for the quantification of one of the elements, manganese was verified by experimental irradiation of manganese-doped bone equivalent phantoms. For the same radiation dose, a 67.9% thermal neutron flux enhancement is reached. With only 4.2% increase of radiation dose, the simulated thermal neutron flux and activation can be further increased by 84.2%. A 100% thermal neutron enhancement ratio is also achievable with a 20% dose increase. The experimental results clearly show higher manganese activation gamma ray counts for each specific phantom, with a significantly reduced minimum detection limit. Additionally, the photon dose was suppressed. The thermal neutron enhancement method can increase the number of useful neutrons significantly but maintain the radiation dose. This greatly decreased the detection limit of the system for elemental quantification at an acceptable dose, which will broadly expand the application of the technology in research and clinical use. The method can also be applied to other neutron medical applications, including neutron imaging and radiotherapy.

The treatment, planning, simulation, and setup of radiotherapy patients contain many processes subject to errors involving both staff and equipment. Cone-beam-CT (CBCT) provides a final check of patient positioning and corrections based on this can be made prior to treatment delivery. Statistical Process Control (SPC) techniques are used in various industries for quality management and error mitigation. The utility of SPC techniques to monitor process and equipment changes in our Head and Neck patient treatments was assessed by application to CBCT results from a quality-focused longitudinal study. Individuals and moving range (XmR) as well as exponentially-weighted moving average (EWMA) techniques were explored. The SPC techniques were sensitive to process changes and trends over the 12 years of data collected. A reduction in the random component of patient setup errors needing correction was observed. Systematic components of error remained more stable. An uptick in both datasets was observed correlating with the COVID-19 pandemic. Process control limits for use in prospective process monitoring were established. Challenges that arose from using SPC techniques in a retrospective study are outlined.

A novel phantom for measuring the 10% and 50% values of the modulation transfer function (MTF) for computed tomography scanners (CT) was investigated. The phantom was constructed by drilling rows of holes of different sizes and frequencies into a small block of polymethyl methacrylate (PMMA). The MTF at a given frequency was determined from the ratio of the range of Hounsfield units within the rows of holes at different frequencies, and the difference in Hounsfield units between air and PMMA. A MTF curve was plotted from measurements at different frequencies and the 10% and 50% MTF values were obtained from a cubic spline interpolation. The MTF results obtained with the drilled hole phantom method were compared to a conventional method - using a thin wire and Spice-CT ImageJ Plugin- and with identical acquisition and reconstruction parameters. The drilled hole phantom measured the 50% MTF with reasonable accuracy but underestimated the 10% MTF by 8.2% on average. MTF measurements were reproducible for repeated image acquisitions and with different users analysing the images, and the phantom was able to accurately measure the change in MTF when measured on images using different reconstruction kernels. The tool may find application as a cheap, easy to use method for routine QC testing of CT scanners.

In recent years, eye lens exposure among radiation workers has become a serious concern in medical X-ray fluoroscopy and interventional radiology (IVR), highlighting the need for radiation protection education and training. This study presents a method that can maintain high accuracy when calculating spatial dose distributions obtained via Monte Carlo simulation and establishes another method to three-dimensionally visualize radiation using the obtained calculation results for contributing to effective radiation-protection education in X-ray fluoroscopy and IVR. The Monte Carlo particle and heavy ion transport code system (PHITS, Ver. 3.24) was used for calculating the spatial dose distribution generated by an angiography device. We determined the peak X-ray tube voltage and half value layer using Raysafe X2 to define the X-ray spectrum from the source and calculated the X-ray spectrum from the measured results using an approximation formula developed by Tucker et al. Further, we performed measurements using the "jungle-gym" method under the same conditions as the Monte Carlo calculations for verifying the accuracy of the latter. An optically stimulated luminescence dosimeter (nanoDot dosimeter) was used as the measuring instrument. In addition, we attempted to visualize radiation using ParaView (version 5.12.0-RC2) using the spatial dose distribution confirmed by the above calculations. A comparison of the measured and Monte Carlo calculated spatial dose distributions revealed that some areas showed large errors (12.3 and 24.2%) between the two values. These errors could be attributed to the scattering and absorption of X-rays caused by the jungle gym method, which led to uncertain measurements, and (2) the angular and energy dependencies of the nanoDot dosimetry. These two causes explain the errors in the actual values, and thus, the Monte Carlo calculations proposed in this study can be considered to have high-quality X-ray spectra and high accuracy. We successfully visualized the three-dimensional spatial dose distribution for direct and scattered X-rays separately using the obtained spatial dose distribution. We established a method to verify the accuracy of Monte Carlo calculations performed through the procedures considered in this study. Various three-dimensional spatial dose distributions were obtained with assured accuracy by applying the Monte Carlo calculation (e.g., changing the irradiation angle and adding a protective plate). Effective radiation-protection education can be realized by combining the present method with highly reliable software to visualize dose distributions.

Theory and modelling suggest that detection of neuronal activity may be feasible using phase sensitive MRI methods. Successful detection of neuronal activity both in vitro and in vivo has been described while others have reported negative results. Magnetic resonance electrical properties tomography may be a route by which signal changes can be identified. Here, we report successful and repeatable detection at 3 Tesla of human brain activation in response to visual and somatosensory stimuli using a functional version of tissue conductivity imaging (funCI). This detects activation in both white and grey matter with apparent tissue conductivity changes of 0.1 S/m (17-20%, depending on the tissue baseline conductivity measure) allowing visualization of complete system circuitry. The degree of activation scales with the degree of the stimulus (duration or contrast). The conductivity response functions show a distinct timecourse from that of traditional fMRI haemodynamic (BOLD or Blood Oxygenation Level Dependent) response functions, peaking within milliseconds of stimulus cessation and returning to baseline within 3-4 s. We demonstrate the utility of the funCI approach by showing robust activation of the lateral somatosensory circuitry on stimulation of an index finger, on stimulation of a big toe or of noxious (heat) stimulation of the face as well as activation of visual circuitry on visual stimulation in up to five different individuals. The sensitivity and repeatability of this approach provides further evidence that magnetic resonance imaging approaches can detect brain activation beyond changes in blood supply.

Radiochromic film, evaluated with flatbed scanners, is used for practical radiotherapy QA dosimetry. Film and scanner component effects contribute to the Lateral Response Artefact (LRA), which is further enhanced by light polarisation from both. This study investigates the scanner bed's contribution to LRA and also polarisation from the mirrors for widely used EPSON scanners, as part of broader investigations of this dosimetry method aiming to improve processes and uncertainties. Alternative scanner bed materials were compared on a modified EPSON V700 scanner. Polarisation effects were investigated for complete scanners (V700, V800, on- and off-axis, and V850 on-axis), for a removed V700 mirror system, and independently using retail-quality single mirror combinations simulating practical scanner arrangements, but with varying numbers (0-5) and angles. Some tests had no film present, whilst others included films (EBT3) irradiated to 6 MV doses of 0-11.3 Gy. For polarisation analysis, images were captured by a Canon 7D camera with 50 mm focal length lens. Different scanner bed materials showed only small effects, within a few percent, indicating that the normal glass bed is a good choice. Polarisation varied with scanner type (7-11%), increasing at 10 cm lateral off-axis distance by around a further 6%, and also with film dose. The V700 mirror system showed around 2% difference to the complete scanner. Polarization increased with number of mirrors in the single mirror combinations, to 14% for 4 and 5 mirrors, but specific values depend on angles and mirror quality. Novel film measurement methods could reduce LRA effect corrections and associated uncertainties.

In this paper, we propose a new deep learning method based on Quaternion Wavelet Transform (QWT) phases of 2D echocardiographic sequences to estimate the motion and strain of myocardium. The proposed method considers intensity and phases gained from QWT as the inputs of customized PWC-Net structure, a high-performance deep network in motion estimation. We have trained and tested our proposed method performance using two realistic simulated B-mode echocardiographic sequences. We have evaluated our proposed method in terms of both geometrical and clinical indices. Our method achieved an average endpoint error of 0.06 mm per frame and 0.59 mm between End Diastole and End Systole on a simulated dataset. Correlation analysis between ground truth and the computed strain shows a correlation coefficient of 0.89, much better than the most efficient methods in the state-of-the-art 2D echocardiography motion estimation. The results show the superiority of our proposed method in both geometrical and clinical indices.

This study introduces a novel watermarking technique for electrocardiogram (ECG) signals. Watermarking embeds critical information within the ECG signal, enabling data origin authentication, ownership verification, and ensuring the integrity of research data in domains like telemedicine, medical databases, insurance, and legal proceedings. Drawing inspiration from image watermarking, the proposed method transforms the ECG signal into a two-dimensional format for QR decomposition. The watermark is then embedded within the first row of the resulting R matrix. Three implementation scenarios are proposed: one in the spatial domain and two in the transform domain utilizing discrete wavelet transform (DWT) for improved watermark imperceptibility. Evaluation on real ECG signals from MIT-BIH Arrhythmia database and comparison to existing methods demonstrate that the proposed method achieves: (1) higher Peak Signal-to-Noise Ratio (PSNR) indicating minimal alterations to the watermarked signal, (2) lower bit error rates (BER) in robustness tests against external modifications such as AWGN noise (additive white Gaussian noise), line noise and down-sampling, and (3) lower computational complexity. These findings emphasize the effectiveness of the proposed QR decomposition-based watermarking method, achieving a balance between robustness and imperceptibility. The proposed approach has the potential to improve the security and authenticity of ECG data in healthcare and legal contexts, while its lower computational complexity enhances its practical applicability.

Breathing conditions affect a wide range of people, including those with respiratory issues like asthma and sleep apnea. Smartwatches with photoplethysmogram (PPG) sensors can monitor breathing. However, current methods have limitations due to manual parameter tuning and pre-defined features. To address this challenge, we propose the PPG2RespNet deep-learning framework. It draws inspiration from the UNet and UNet + + models. It uses three publicly available PPG datasets (VORTAL, BIDMC, Capnobase) to autonomously and efficiently extract respiratory signals. The datasets contain PPG data from different groups, such as intensive care unit patients, pediatric patients, and healthy subjects. Unlike conventional U-Net architectures, PPG2RespNet introduces layered skip connections, establishing hierarchical and dense connections for robust signal extraction. The bottleneck layer of the model is also modified to enhance the extraction of latent features. To evaluate PPG2RespNet's performance, we assessed its ability to reconstruct respiratory signals and estimate respiration rates. The model outperformed other models in signal-to-signal synthesis, achieving exceptional Pearson correlation coefficients (PCCs) with ground truth respiratory signals: 0.94 for BIDMC, 0.95 for VORTAL, and 0.96 for Capnobase. With mean absolute errors (MAE) of 0.69, 0.58, and 0.11 for the respective datasets, the model exhibited remarkable precision in estimating respiration rates. We used regression and Bland-Altman plots to analyze the predictions of the model in comparison to the ground truth. PPG2RespNet can thus obtain high-quality respiratory signals non-invasively, making it a valuable tool for calculating respiration rates.

In this paper, we proposed a complete study method to achieve accurate aortic dissection diagnosis at the patient level. Based on the CT angiography (CTA) images, a classification model named DAT-DenseNet, which combined the deep attention Transformer module with the DenseNet architecture is proposed. In the first phase, two DAT-DenseNet are combined in parallel. It is used to accurately achieve two classification task at the CTA images. In the second stage, we propose a feature fusion module. It concatenates and fuses the image features output from the two classification models on a patient by patient basis. In the comparison experiments of classification model performance, DAT-DenseNet obtained 92.41 accuracy at the image level, which was 2.20 higher than the commonly used model. In the comparison experiments of model fusion method, our method obtained 90.83 accuracy at the patient level. The experiments showed that DAT-DenseNet model exhibits high performance at the image level. Our feature fusion module achieves the mapping from two classification image features to patient outcomes. It achieves accurate patient classification. The experiments' results in the Discussion section elaborate the details of the experiment and confirmed that the results were reliable.

To develop and assess an automated Sub-arc Collimator Angle Optimization (SACAO) algorithm and Cumulative Blocking Index Ratio (CBIR) metrics for single-isocenter coplanar volumetric modulated arc therapy (VMAT) to treat multiple brain metastases. This study included 31 patients with multiple brain metastases, each having 2 to 8 targets. Initially, for each control point, the MLC blocking index was calculated at different collimator angles, resulting in a two-dimensional heatmap. Optimal sub-arc segmentation and collimator angle optimization were achieved using an interval dynamic programming algorithm. Subsequently, VMAT plans were designed using two approaches: SACAO and the conventional Full-Arc Fixed Collimator Angle. CBIR was calculated as the ratio of the cumulative blocking index between the two plan approaches. Finally, dosimetric and planning parameters of both plans were compared. Normal brain tissue, brainstem, and eyes received better protection in the SACAO group (P < 0.05).Query Notable reductions in the SACAO group included 11.47% in gradient index (GI), 15.03% in monitor units (MU), 15.73% in mean control point Jaw area (A), and 19.14% in mean control point Jaw-X width (W), all statistically significant (P < 0.001). Furthermore, CBIR showed a strong negative correlation with the degree of plan improvement. The SACAO method enhanced protection of normal organs while improving transmission efficiency and optimization performance of VMAT. In particular, the CBIR metrics show promise in quantifying the differences specifically in the 'island blocking problem' between SACAO and conventional VMAT, and in guiding the enhanced application of the SACAO algorithm.

Accurate segmentation of ovarian cancer (OC) lesions in PET/CT images is essential for effective disease management, yet manual segmentation for radiomics analysis is labor-intensive and time-consuming. This study introduces the application of a 3D U-Net deep learning model, leveraging advanced 3D networks, for multi-class semantic segmentation of OC in PET/CT images and assesses the stability of the extracted radiomics features. Utilizing a dataset of 3120 PET/CT images from 39 OC patients, the dataset was divided into training (70%), validation (15%), and test (15%) subsets to optimize and evaluate the model's performance. The 3D U-Net model, especially with a VGG16 backbone, achieved notable segmentation accuracy with a Dice score of 0.74, Precision of 0.76, and Recall of 0.78. Additionally, the study demonstrated high stability in radiomics features, with over 85% of PET and 84% of CT image features showing high intraclass correlation coefficients (ICCs > 0.8). These results underscore the potential of automated 3D U-Net-based segmentation to significantly enhance OC diagnosis and treatment planning. The reliability of the extracted radiomics features from automated segmentation supports its application in clinical decision-making and personalized medicine. This research marks a significant advancement in oncology diagnostics, providing a robust and efficient method for segmenting OC lesions in PET/CT images. By addressing the challenges of manual segmentation and demonstrating the effectiveness of 3D networks, this study contributes to the growing body of evidence supporting the application of artificial intelligence in improving diagnostic accuracy and patient outcomes in oncology.