Substernal goiters can cause aerodigestive symptoms but are often asymptomatic. This study evaluated the relationship between symptoms, demographics, pathology, and anatomical severity in substernal goiter patients using standardized tools. A retrospective review of 201 patients surgically treated for substernal goiters over 11 years at a quaternary-care center was conducted. CT scan images were analyzed to measure tracheal deviation (TD), compression (TC), and depth. Symptoms included dysphagia, dyspnea, dysphonia, and coughing. Associations with demographics, anatomy, and pathology were analyzed. Dysphagia was the most common symptom (48.8%); 27% were asymptomatic. TC and larger specimen weights were significantly associated with dyspnea (p < 0.001, 0.013). TD, depth, age, and BMI showed no significant correlation with symptoms. Cough and vocal cord dysfunction were linked to aggressive pathology (p = 0.003, 0.017). CT imaging is essential in evaluating substernal goiters, as symptoms alone are insufficient. Symptom profiles may also aid in predicting pathological severity.

Predicting the risk of stroke is one of the critical problems in healthcare, which necessitates efficient solutions for providing accurate and prompt risk assessments while preserving data confidentiality. This work proposes a new framework using Federated Learning (FL) to combine Multi-Layer Perceptron (MLP) and Gated Recurrent Unit (GRU) models that are essential in analyzing multimodal data. Implemented in Python, the approach incorporates two datasets: Dataset 1, which consists of Demographic data medical history, and lifestyle data, and the second dataset, which includes the normal condition and the affected stroke condition CT scan images. Imputation of missing values, feature normalization by Min-Max scaling, and handling of imbalanced classes with SMOTE make the data pre-processing procedures exhaustive. In FL architecture three clients –Client A, Client B, and Client C – process a split multimodal dataset containing static and sequential information. Each client independently trains an MLP-GRU model. Each is applied with MLP handling static features from Dataset 1 and GRU handling sequential features from Dataset 2. To update models, Federated Averaging is used on a central server, to create a global model that is then returned to the clients for further refinement. The accuracy of the proposed method averages 99.00% and surpasses other models by 2.5% including CNN, LSTM, Random Forest, and SVM. By enhancing MLP with GRU and applying them to a privacy-preserving FL framework,The study addresses the fragmented use of multimodal medical data, where clinical records and imaging are generally evaluated separately, resulting in inadequate diagnostic support. The strategy integrates complementary modalities to create a more comprehensive perspective of patient health, enhancing healthcare predictive accuracy and decision-making. This incentive is essential for improving computational methods and linking technical advancement with medical objectives like fast diagnosis and therapy planning. The introduction emphasises the therapeutic necessity of harmonising organized and unstructured data to reduce diagnostic ambiguity. A translational approach is used to discuss how multimodal integration might improve clinical workflows, develop collaborative healthcare systems, and support sustainable medical practices. This repeated emphasis links methodological advances to real-world healthcare issues, boosting the study’s academic relevance sets.

We report a case of radiologically confirmed large vessel vasculitis (LVV) in a woman in her 70s undergoing neoadjuvant FOLFIRINOX (folinic acid, 5-fluorouracil, irinotecan, oxaliplatin) chemotherapy for borderline resectable pancreatic adenocarcinoma. The patient presented with fever and raised C reactive protein (CRP) shortly after completing her third cycle of chemotherapy. Infection and vasculitis screens were negative. CT and positron emission tomography CT scans showed shrinkage of the primary cancer but changes consistent with LVV. Granulocyte-colony stimulating factor (G-CSF), felt to be the most likely causative agent, was discontinued and the patient started high dose prednisolone, with a prolonged weaning course. At the 6-week follow-up, there was complete resolution of her LVV on CT and normalisation of CRP. She successfully completed her remaining cycles of chemotherapy without G-CSF, and her presentation did not recur. However, her pancreatic cancer remained inoperable, and she sadly died from cancer progression 18 months later.

Precision medicine aims to define subtypes of a heterogeneous disease, which can lead to more specific diagnosis, prognosis, and/or treatment. Chronic Obstructive Pulmonary Disease (COPD) is a heterogeneous disease and therefore appropriate for a precision medicine approach. The idea of COPD heterogeneity has been proposed for decades, initially with subtypes of emphysema and chronic bronchitis. Modern approaches include the use of chest CT scan imaging and omics biomarkers. One path forward is to start by identifying a clinical question, then try to understand the epidemiology, to describe clinical phenotypes and disease subtypes. One can then incorporate omics biomarkers to arrive at an endotype, a subtype with a shared biologic mechanism. Eosinophilic COPD and alpha-1 antitrypsin deficiency-related emphysema are endotypes that already have specific therapies. Subtypes such as patients with frequent exacerbations can be targeted with several treatments, but multiple biological processes are likely to be important. This review will highlight these approaches, using examples such as airway-predominant COPD, frequent exacerbations, and asthma-COPD overlap. These investigations have been conducted in large observational studies, including the multi-center U.S. COPDGene Study. The analyses have leveraged the wealth of data in COPDGene, including clinical information, pulmonary function tests, chest CT scans, and multi-omics such as genetics, RNA-sequencing and proteomics. Despite these advances, there are many challenges for COPD precision medicine, such as the requirement for large studies with longitudinal outcomes and available biospecimens. Clinical trials of targeted therapies will be needed for the ultimate application of precision medicine in COPD.

Improving Medical Imaging Interpretation through a Clerkship Anatomy Selective.

Lung cancer is still ranked as one of the most common causes of cancer related deaths all over the world, and this has necessitated early detection which will enhance the survival chances of patients. The process of interpreting the images of lung CT scan by hand is a complicated and time- consuming process and it relies heavily on the experience of radiologists. This weakness underscores the importance of having effective automated machinery that can help in correct and effective diagnosis. In this contribution, a smart deep learning-based system of lung cancer identification was created. The proposed system was designed to take the raw DICOM CT images and perform normalization of Hounsfield Unit (HU) to maintain critical information on radiology. The preprocessing of the images followed by noise reduction, intensity normalization and resizing of the images was done to enhance the image quality overall. The hybrid models of U-Net and SegNet architectures were used to segment the tumor regions, which permitted a more accurate reflecting of the affected region through the combination of both contextual and boundary-level features. Keywords: CT Scan Analysis , Deep Learning , Hybrid Segmentation , Image Preprocessing , Lung Cancer Detection , Medical Image Classification , SegNet , U-Net

Parathyroid four-dimensional CT (4DCT) involves repeated non-contrast and multiphase contrast-enhanced acquisitions, leading to increased radiation exposure. As the examination targets the anatomically complex structures of the neck, maintaining high image quality is particularly important. Photon-counting CT (PCCT) is a novel technology reported to improve dose efficiency and image quality compared to conventional energy-integrating detector CT (EIDCT). This study provides a systematic, quantitative comparison of CT scanner radiation output (CTDIvol and DLP) and image quality, assessed by signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR), between PCCT and EIDCT for multiphase parathyroid 4DCT exams. In this IRB-approved retrospective study, 96 patients who underwent a 4DCT for the workup of hyperparathyroidism (PCCT, n = 32; EIDCT, n = 64) were reviewed. Regions of interest (ROIs) were placed on the non-contrast, arterial, and venous phase images in the sternocleidomastoid muscles at the upper thyroid level, the anterior scalene muscles at the lower thyroid level, and in the extracorporeal air. Mean attenuation, SNR, and CNR were measured. Statistical analyses included unpaired t-tests and two-way ANOVA. PCCT had significantly lower radiation output across all 4DCT phases (total CTDIvol 30.6 ± 3.3 vs. 60.8 ± 15.7 mGy; DLP 859 ± 178 vs. 1523 ± 519 mGy·cm; P < .001). SNR and CNR were significantly higher on PCCT across all evaluated ROIs, with lower background noise (all P < .05). Two-way ANOVA confirmed the advantages of PCCT regardless of contrast injection side, with no significant interactions. PCCT substantially reduces CT scanner radiation output while improving quantitative image quality in multiphase parathyroid 4DCT, including technically challenging arterial-phase acquisitions at the lower thyroid level, supporting its technical advantage over conventional EIDCT.

This study aims to develop a framework for detecting and classifying lung diseases, including pneumonia, tuberculosis, and lung cancer, using standard X-ray and CT scan images augmented with volumetric datasets. Leveraging three deep learning architectures—Sequential, Functional, and Transfer Learning models— we trained these systems on publicly available datasets to explore their potential in advancing biomedical imaging. Deep learning, particularly convolutional neural networks (CNNs), offers transformative capabilities in feature extraction from medical images, enabling faster and more accurate disease diagnosis. Our work evaluates the efficacy of these models against existing approaches, addressing challenges such as poor performance in traditional networks when handling rotated, tilted, or atypically oriented images. The proposed Sequential model achieved an F1 score of 98.55%, accuracy of 98.43%, and recall of 96.33% for pneumonia detection, while demonstrating 97.99% F1 score, 99.4% accuracy, and 98.88% recall for tuberculosis. The Functional model excelled in lung cancer classification with 99.9% accuracy and 99.89% specificity, requiring fewer trained parameters and reducing computational costs compared to pre-trained benchmarks. Our results highlight the framework’s superiority in accuracy and efficiency, positioning it as a cost-effective, high performing solution for early disease diagnosis. By integrating state-of-the-art CNNs with diverse architectures, this study advances automated lung disease classification and underscores the clinical viability of deep learning in medical imaging.

598 Background: Follow-up of patients with germ-cell tumours (GCT) relies on monitoring serum markers (SM) including alpha-fetoprotein (AFP), human chorionic gonadotropin (hCG), and lactate dehydrogenase (LDH). Instances of benign, non-GCT related elevation of AFP have been reported as case reports. Identifying such occurrences is crucial to prevent misdiagnosis of residual disease and subsequent overtreatment, which may lead to increased morbidity. Methods: We conducted a national retrospective study in nine GETUG institutions. Patients with isolated persistent elevation of AFP levels above the normal range after receiving curative treatment for GCT without detectable disease on CT scan imaging were identified. Data on patient characteristics, treatment response, changes in SM values over time, and serum AFP levels from first- and second-degree relatives were collected. Familial AFP elevation was defined as AFP levels above the normal range in at least one first- or second-degree relative. Results: From June 1990 to April 2024, 31 patients were identified. Median age was 35 years (range, 19 to 49 years). 15 patients (48%) had pure seminoma, and 16 (52%) had non-seminomatous GCT. Overall, 19 patients (61%) had a clinical stage I disease, 17 (23%) had stage II, and 5 (16%) had stage III. No patient had a history of chronic alcoholism or consumption of medications with hepatic toxicity. Median AFP level after curative treatment was 12 ng/mL (range, 8.8 to 36 ng/mL). No evidence of disease was found on CT scan. An ultrasound examination of the contralateral testis and an FDG-PET/CT were performed in 15 (48%) and 7 patients (23%), respectively, also showing no evidence of disease. Serum AFP levels from family relatives were obtained for 14 patients (45%), with a family elevation identified for 10/14 patients (71%). After a median follow-up of 35 months (range: 7-253 months), AFP levels remained elevated in all 31 patients without documented disease relapse. Conclusions: Benign and often familial elevations of serum AFP levels are rare occurrences, which can be a chance finding following curative treatment for GCT. Ultrasound examination of the contralateral testis and AFP measurement in family members are recommended. Surveillance alone is appropriate.

The COVID-19 pandemic has caused massive devastation worldwide, and its effects still persist. Managing the early stages was difficult, but scientists worked tirelessly to control it. The emergence of variants continues to pose a threat, raising doubts about the capability of the healthcare system. Healthcare practitioners have faced immense strain under a massive patient load, while delays in testing have caused deaths due to untimely treatment. Moreover, relying only on RT-PCR testing is insufficient because of its diagnostic errors. To address these challenges, this study introduces a Smart Imaging Lab Framework for hospitals. The approach uses a convolutional neural network (CNN) model to carry out rapid X-ray and CT-scan assessments of emergency patients showing severe symptoms, following RT-PCR testing. In addition, blood tests help determine the severity of infection. Patients in critical condition are transferred to intensive care units, while those with milder cases remain in general wards. The framework uses a 16-layer CNN framework for X-ray and CT-scan imaging, achieving 99.02% and 98.49% accuracy, respectively. Severity assessment with Extra Randomized Trees reached 98.00% accuracy. These findings highlight the potential of the system to be adopted in hospitals, enabling regular health monitoring and timely intervention. In addition, explainable AI XAI tools like Grad-CAM increase transparency by highlighting the lung regions most relevant to the diagnosis. The study demonstrates the potential of artificial intelligence, internet of things, and cloud computing to address future pandemic-prone diseases.

Lung cancer is a highly perilous illness ranking as one of the primary causes of disease and death, particularly when diagnosed in its initial stages. It presents significant challenges, as it is often only discernible after it has already diffused. This study proposes a lung cancer prognostication framework that uses deep learning to enhance the accuracy of cancer forecasting and disease determination, thereby enabling personalized treatment approaches based on disease severity. It consists of various steps, including image preprocessing and segmentation of lung CT image features extracted from the segmented images. Three different models, namely a DCNN model, a DCDNN model, and an ANN model, were employed for image classification, and a deep convolutional neural network (DCNN) was employed to detect lung diagnosis based on the extracted feature evaluation results showing the best accuracy of 99.41% in accurately discerning the presence or absence of lung cancer. The GAN model generates realistic lung CT scan images by training a generator to produce authentic images, and a discriminator to distinguish between real and fake images. The outcome of the system depends on the quality of the data, and a well-trained DCNN through training, validation, and testing on diverse datasets is crucial to ensure the reliability and generalizability of the model.

Brachytherapy (BT) is a local radiation treatment method for solid tumors. A single 10 Gy high-dose-rate (HDR) BT acts as an "in situ" vaccination. Tumor microenvironment (TME)-dependent radio-resistance mechanisms, such as increasing immunosuppression and hypoxia, lead to tumor recurrence after radiotherapy. Our study aimed to determine whether adding imiquimod (IMQ) to anticancer therapy would overcome TME-mediated mechanisms of radiotherapy resistance. IMQ, a toll-like receptor 7 (TLR7) agonist, acts as an immunostimulant and a vascular normalizing agent. Mice with well-developed tumors were treated with IMQ at a vascular-normalized dose of 50 μg, followed 5 days later by a single 10 Gy HDR BT. The dose coverage was planned using Discovery RT computed tomography CT scans. Irradiation was performed with a high-dose-rate afterloader equipped with an iridium-192 radioactive source. In mice treated with a combination of IMQ and BT, we observed significant inhibition of melanoma tumor growth. We also noticed an effective therapeutic effect in mice with breast cancer, resulting in significantly prolonged survival and complete tumor regression in 20% of treated mice. In the blood of treated mice, we observed leukopenia with eosinophilia. In tumors, there was enhanced infiltration by cytotoxic CD8+ T lymphocytes. The depletion of CD8+ T cells completely abolished the effect of the combined therapy. The combination of IMQ with HDR brachytherapy induces a synergistic effect, improving the therapeutic antitumor effect of brachytherapy. Our data indicate that it is reasonable to use drugs that prevent changes in the TME in combination with radiotherapy.

An indigenous breast phantom was customized to optimize the imaging quality of the CT scan according to Taguchi's methodology. A 3D printer made the base gauge of the breast phantom, and the polysmoothTM filament was sprayed. The gauge can be categorized into three major parts vertical and horizontal line pairs, two V-shape slices, and two arrays of nodules with various sizes, then a spherical shell mad by paraffin casting and the sunflower oil was infused as filling material. The Taguchi approach led to the development of a customized measuring device that enabled quantitative assessment aimed at improving the spatial resolution of CT image quality. Five essential factors for operating the CT chest scan (kVp, mAs, FOV, Pitch, and gantry rotation time) were organized according to Taguchi L18(21 × 34) suggestion. Three well-trained radiologists ranked the imaging quality in three discrete time periods according to the imaging quality's sharpness, contrast, and spatial resolution. The derived average, standard deviation, and signal-to-noise ratio of specific factors were reorganized and analyzed from the multiple measurements to propose the optimal CT chest scan protocol recommendation. Accordingly, the optimal suggestion was A1(120 kVp), B3(200 mAs), C2(FOV 350 mm2), D1(pitch 0.516) and E2(rotation time 0.75 s) to fulfill the ALARA principle. In this research, a numerical metric termed the minimum detectable difference (MDD) was introduced to evaluate imaging performance, and its calculated value demonstrated a resolution capability of approximately 1.57 mm.

FVCM-Net: Interpretable privacy-preserved attention driven lung cancer detection from CT scan images with explainable HiRes-CAM attribution map and ensemble learning

Resource-limited settings continue to face challenges in the identification of COVID-19, bacterial pneumonia, viral pneumonia, and normal lung conditions because of the overlap of CT appearance and inter-observer variability. We justify a hybrid architecture of deep learning which combines hand-designed descriptors (Histogram of Oriented Gradients, Local Binary Patterns) and a 20-layer Convolutional Neural Network with dual self-attention. Handcrafted features were then trained with Support Vector Machines, and ensemble averaging was used to integrate the results with the CNN. The confidence level of 0.7 was used to mark suspicious cases to be reviewed manually. On a balanced dataset of 14,000 chest CT scans (3500 per class), the model was trained and cross-validated five-fold on a patient-wise basis. It had 97.43% test accuracy and a macro F1-score of 0.97, which was statistically significant compared to standalone CNN (92.0%), ResNet-50 (90.0%), multiscale CNN (94.5%), and ensemble CNN (96.0%). A further 2–3% enhancement was added by the self-attention module that targets the diagnostically salient lung regions. The predictions that were below the confidence limit amounted to only 5 percent, which indicated reliability and clinical usefulness. The framework provides an interpretable and scalable method of diagnosing multiclass lung disease, especially applicable to be deployed in healthcare settings with limited resources. The further development of the work will involve the multi-center validation, optimization of the model, and greater interpretability to be used in the real world.

Abstract In carbonate reservoir stimulation, achieving controlled acid reactivity is critical to prevent rapid acid spending and ensure deeper penetration into the formation. Acid retardation slows the reaction between acid and rock and helps extend the contact time, resulting in more uniform dissolution and enhancing the stimulation effect. Emulsified acid systems offer an effective way to achieve this by encapsulating the reactive phase within a non-reactive continuous phase, delaying the acid release and reducing premature reaction near the wellbore. This study investigates the performance of a Novel Emulsified Acid (NEA) system designed to improve acid penetration, control reaction kinetics, and minimize formation damage compared to conventional acid systems. The NEA formulation was evaluated through core flooding experiments, with structural alterations validated using Micro CT-Scan imaging. The system consists of hydrochloric acid as the reactive phase, stabilized within a Non-Aqueous Phase (NAP) as the continuous phase to delay reactivity and extend acid contact time. Unlike traditional emulsified acids, NEA employs an optimized NAP formulation for enhanced stability and retardation. Emulsion stability is further reinforced by an emulsifier, which helps stabilize the system, along with additives that serve as polymer stabilizers and metal-binding agents (chelating agents). Core flooding experiments demonstrated that NEA significantly improves reservoir characteristics by enhancing porosity and permeability, and by generating more efficient wormholes compared to using conventional acid systems. Micro CT-Scan imaging confirmed the formation of well-connected flow paths, validating the effectiveness of NEA in optimizing acid penetration and rock dissolution patterns. These findings demonstrate the potential of NEA as a superior stimulation fluid for unlocking untapped productivity in complex carbonate formations. Future field-scale implementation is planned as a continuation of this laboratory-scale experiment to validate performance under real reservoir conditions.

Abstract Kidney stones, formed by the accumulation of minerals and salts in the kidneys, can lead to severe pain and significant complications. Timely and accurate detection of these stones, particularly in medical imaging like CT scans, is crucial for effective treatment. This study introduces a novel method for detecting and classifying kidney stones in CT images, addressing the challenges posed by the variety of stone shapes and sizes, image noise, and the difficulty of identifying small or atypical stones. The proposed method utilizes a graph attention network to enhance detection accuracy and minimize identification errors. Initially, keypoints are extracted from the images using the ORB algorithm. These keypoints are considered as nodes in a graph, and using an appropriate threshold, the nodes are connected to each other to form the graph. The relationships between them are then defined through an adjacency matrix. Subsequently, for each node, a feature vector is obtained by using the pixel values within a 5 × 5 window centered on the corresponding node, and a feature matrix is constructed for the entire graph. The graph is then transformed into an embedded space, where features are updated through the graph attention network. This attention mechanism allows the model to discern the significance of each node and its connections, effectively extracting structural information. The updated embedded data is subsequently fed into a deep neural network for training and classification. Evaluation of the model’s performance is conducted using a dataset of coronal CT scan images. The results indicate that the proposed model achieves an impressive accuracy of 99.14% in detecting kidney stones, significantly outperforming existing methods, particularly in identifying small and atypical stones. This study highlights the effectiveness of integrating graphical features with attention mechanisms for enhanced medical image analysis, aiding in the early and accurate diagnosis of kidney stones.

Although previous research has examined the first lumbar (L1) vertebra for sex estimation, studies focusing on its role in age estimation have been limited. This study expands on earlier work by investigating the morphological variations in shape and size of the L1 vertebra across different age groups within the Malaysian population. A sample of 440 abdominal CT images was collected from the Radiology Department at Universiti Kebangsaan Malaysia Medical Centre. These images included adults aged 18 to 80, who visited the department in 2019. Twenty-seven 3D landmarks were marked on each L1 vertebra using digitized 3D CT scan images. Statistical analyses were performed using a geometric morphometric approach to evaluate age-related variations in the shape and size of the L1 vertebra. Principal Component Analysis identified 74 shape variables describing the shape of the L1 vertebra, with the first five principal components explaining 38.27% of the variance. The Canonical Variate Analysis scatter plot showed slight separation among the confidence ellipses for the three age groups, with significant p-values (p < 0.001). Procrustes ANOVA revealed significant differences in both the size and shape of the L1 vertebra across all age groups. Additionally, multivariate regression of shape on continuous age revealed a significant, biologically meaningful pattern (R² = 0.022, p = 0.001). This study shows that the size and shape of the L1 vertebra differ across various age groups. In elderly individuals, the L1 vertebra is characterized by longer spinous processes and shorter, flatter vertebral bodies. Conversely, younger individuals tend to have L1 vertebrae with shorter transverse spinous processes and taller vertebral bodies.

Background:The cancellous tissue forming the inner layer of long bones is highly porous at the center, with porosity decreasing toward the outer layer, leading to gradual variations in mechanical properties. Hence, cancellous tissue can be regarded as a functionally graded material (FGM). This study investigates the mechanical properties of graded cancellous bone.Methods:CT scan images combined with image processing techniques were used to extract gradients in mechanical properties of the femoral neck in bovine samples. Several unit cells were employed to model the microstructure of cancellous bone. The graded properties were validated through both numerical and experimental approaches. Cylindrical models are used for finite element analysis and complementary experimental tests were carried out on the femoral neck region.Results:Analytical relationships for mechanical properties of femur spongy bone have been presented. The Cubic and BCC unit cell structures, with and have maximum and minimum flexural stiffness values, respectively. Also, discrepancies between experimental, analytical, and numerical results were discussed.Conclusions:The tesseract unit cell showed the most similarity with the cancellous bone properties, with only 0.11% difference in flexural stiffness, whereas the cubic unit cell, with an 8.48% difference, was the least suitable for modeling spongy bone.

Chronic subdural hematoma (CSDH) commonly affects older adults and remains associated with recurrence despite surgical evacuation. Coagulation of middle meningeal artery (MMA) during burr-hole drainage may be useful where MMA embolization is unavailable, however reliable external landmarks for MMA localization are poorly defined. This study aimed to identify the MMA confluence point on 3D cranial CT scans of elderly Filipino patients and define its relationship to external cranial landmarks to support pre-operative planning. A retrospective cross-sectional morphometric study was performed using plain cranial CT scans of patients aged ≥65 years from 2019 to 2023. Scans with intact calvarial anatomy and adequate visualization of MMA groove were included. Threedimensional reconstructions were generated, and bilateral distances from lateral canthus, external auditory canal, and canthomeatal line to MMA confluence point were measured. Analyses were performed by sex and laterality (p<0.05). A total of 221 patients were included (mean age 76 years; 63% female). The external auditory canal and lateral canthus showed low variability. Male patients had greater external auditory canal and canthomeatal line distances (p<0.05), while confluence measurements did not differ by sex. The confluence point was located farther from the canthomeatal line on the left (p=0.0125). These findings provide a practical framework for MMA localization during pre-operative planning. In settings without MMA embolization, landmark-based localization offers a low-cost method to guide burr-hole placement and potential MMA coagulation, supporting resource-adapted strategy to reduce CSDH recurrence.