Search Results (200)

Background/Objectives: Discrepancies in diabetic retinopathy (DR) grading are well-documented, with retinal non-perfusion (RNP) quantification posing greater challenges. This study assessed intergrader agreement in DR evaluation, focusing on qualitative severity grading and quantitative RNP measurement. We aimed to improve agreement through structured consensus meetings. Methods: A retrospective analysis of 100 comparisons from 50 eyes (36 patients) was conducted. Two paired medical retina fellows graded ultra-widefield color fundus photographs (CFP) and fundus fluorescein angiography (FFA) images. CFP assessments included DR severity using the International Clinical Diabetic Retinopathy (ICDR) grading system, DR Severity Scale (DRSS), and predominantly peripheral lesions (PPL). FFA-based RNP was defined as capillary loss with grayscale matching the foveal avascular zone. Weekly adjudication by a senior specialist resolved discrepancies. Intergrader agreement was evaluated using Cohen’s kappa (qualitative DRSS) and intraclass correlation coefficients (ICC) (quantitative RNP). Bland–Altman analysis assessed bias and variability. Results: After eight consensus meetings, CFP grading agreement improved to excellent: kappa = 91% (ICDR DR severity), 89% (DRSS), and 89% (PPL). FFA-based PPL agreement reached 100%. For RNP, the non-perfusion index (NPI) showed moderate overall ICC (0.49), with regional ICCs ranging from 0.40 to 0.57 (highest in the nasal region, ICC = 0.57). Bland–Altman analysis revealed a mean NPI difference of 0.12 (limits: −0.11 to 0.35), indicating acceptable variability despite outliers. Conclusions: Structured consensus training achieved excellent intergrader agreement for DR severity and PPL grading, supporting the clinical reliability of ultra-widefield imaging. However, RNP measurement variability underscores the need for standardized protocols and automated tools to enhance reproducibility. This process is critical for developing robust AI-based screening systems. Full article

Background: Limited availability of high-quality labeled biomedical image datasets presents a significant challenge for training deep learning models in medical diagnostics. This study proposes a novel image generation framework combining conditional generative adversarial networks (cGANs) with a Mixture-of-Experts (MoE) architecture and color histogram-aware loss functions to enhance synthetic blood cell image quality. Methods: RGB microscopic images from the BloodMNIST dataset (eight blood cell types, resolution 3 × 128 × 128) underwent preprocessing with k-means clustering to extract the dominant colors and UMAP for visualizing class similarity. Spearman correlation-based distance matrices were used to evaluate the discriminative power of each RGB channel. A MoE–cGAN architecture was developed with residual blocks and LeakyReLU activations. Expert generators were conditioned on cell type, and the generator’s loss was augmented with a Wasserstein distance-based term comparing red and green channel histograms, which were found most relevant for class separation. Results: The red and green channels contributed most to class discrimination; the blue channel had minimal impact. The proposed model achieved 0.97 classification accuracy on generated images (ResNet50), with 0.96 precision, 0.97 recall, and a 0.96 F1-score. The best Fréchet Inception Distance (FID) was 52.1. Misclassifications occurred mainly among visually similar cell types. Conclusions: Integrating histogram alignment into the MoE–cGAN training significantly improves the realism and class-specific variability of synthetic images, supporting robust model development under data scarcity in hematological imaging. Full article

See-through augmented reality glasses, such as HoloLens 2, are increasingly adopted in medical settings; however, their efficacy as medical display devices remains unclear, as current evaluation protocols are designed for traditional monitors. This study examined whether the established display-evaluation techniques apply to HoloLens 2 and whether it meets standards for primary and secondary medical displays. HoloLens 2 was assessed for overall image quality, luminance, grayscale consistency, and color uniformity. Five participants rated the TG18-OIQ pattern under ambient lighting conditions of 2.4 and 138.7 lx. Minimum and maximum luminance were measured using the TG18-LN12-03 and -18 patterns, targeting ≥ 300 cd/m² and a luminance ratio ≥ 250. Grayscale conformity to the standard grayscale display function allowed deviations of 10% for primary and 20% for secondary displays. Color uniformity was measured at five screen positions for red, green, and blue, with a chromaticity limit of 0.01 for primary displays. HoloLens 2 satisfied four of the ten primary and four of the seven secondary overall-quality criteria, achieving a maximum luminance of 2366 cd/m² and a luminance ratio of 1478.75. Grayscale uniformity was within tolerance for 10 of the 15 primary and 13 of the 15 secondary measurements, while 25 of the 30 color uniformity values exceeded the threshold. The adapted evaluation methods facilitate a systematic assessment of HoloLens 2 as a medical display. Owing to inadequate grayscale and color representation, the headset is unsuitable as a primary diagnostic display; for secondary use, requirements must be assessed based on specific application requirements. Full article

Artificial intelligence (AI) is revolutionizing the field of medical imaging, offering unprecedented capabilities in data analysis, image interpretation, and decision support. Transcranial Doppler (TCD) and Transcranial Color-Coded Doppler (TCCD) are widely used, non-invasive modalities for evaluating cerebral hemodynamics in acute and chronic conditions. Yet, their reliance on operator expertise and subjective interpretation limits their full potential. AI, particularly machine learning and deep learning algorithms, has emerged as a transformative tool to address these challenges by automating image acquisition, optimizing signal quality, and enhancing diagnostic accuracy. Key applications reviewed include the automated identification of cerebrovascular abnormalities such as vasospasm and embolus detection in TCD, AI-guided workflow optimization, and real-time feedback in general ultrasound imaging. Despite promising advances, significant challenges remain, including data standardization, algorithm interpretability, and the integration of these tools into clinical practice. Developing robust, generalizable AI models and integrating multimodal imaging data promise to enhance diagnostic and prognostic capabilities in TCD and ultrasound. By bridging the gap between technological innovation and clinical utility, AI has the potential to reshape the landscape of neurovascular and diagnostic imaging, driving advancements in personalized medicine and improving patient outcomes. This review highlights the critical role of interdisciplinary collaboration in achieving these goals, exploring the current applications and future directions of AI in TCD and TCCD imaging. This review included 41 studies on the application of artificial intelligence (AI) in neurosonology in the diagnosis and monitoring of vascular and parenchymal brain pathologies. Machine learning, deep learning, and convolutional neural network algorithms have been effectively utilized in the analysis of TCD and TCCD data for several conditions. Conversely, the application of artificial intelligence techniques in transcranial sonography for the assessment of parenchymal brain disorders, such as dementia and space-occupying lesions, remains largely unexplored. Nonetheless, this area holds significant potential for future research and clinical innovation. Full article

Background/Objectives Recent advancements in large language models, such as ChatGPT-4o, have created new opportunities for analyzing complex multi-modal data, including medical images. This study aims to assess the potential of ChatGPT-4o in distinguishing between benign and malignant thyroid nodules via multi-modality ultrasound imaging: grayscale ultrasound, color Doppler ultrasound (CDUS), and shear wave elastography (SWE). Materials and Methods Patients who underwent thyroid nodule ultrasound examinations and had confirmed pathological diagnoses were included. ChatGPT-4o analyzed the multi-modality ultrasound data using two approaches: (1.) a dual-modality strategy which employed grayscale ultrasound and CDUS, and (2.) a triple-modality strategy which incorporated grayscale ultrasound, CDUS, and SWE. The diagnostic performance was compared against pathological findings utilizing receiver operating characteristic (ROC) curve analysis, while consistency was evaluated through Cohen’s Kappa analysis. Results A total of 106 thyroid nodules were evaluated; 65.1% were benign and 34.9% malignant. In the dual-modality approach, ChatGPT-4o achieved an area under the ROC curve (AUC) of 66.3%, moderate agreement with pathology results (Kappa = 0.298), a sensitivity of 70.3%, a specificity of 62.3%, and an accuracy of 65.1%. Conversely, the triple-modality approach exhibited higher specificity at 97.1% but lower sensitivity at 18.9%, with an accuracy of 69.8% and a reduced overall agreement (Kappa = 0.194), resulting in an AUC of 58.0%. Conclusions ChatGPT-4o exhibits potential, to some extent, in classifying thyroid nodules using multi-modality ultrasound imaging. However, the dual-modality approach unexpectedly outperforms the triple-modality approach. This indicates that ChatGPT-4o might encounter challenges in integrating and prioritizing different data modalities, particularly when conflicting information is present, which could impact diagnostic effectiveness. Full article

Background and Objectives: Traditionally, we evaluate the size of a wound by using Opsite Flexigrid transparent film dressing, placing it over the wound, tracing the edges of the wound, and then calculating the area. However, this method is both time-consuming and subjective, often leading to varying results depending on the individual performing the assessment. In this study, our goal is to provide an objective method to calculate the wound size and solve variations in photo-taking distance caused by different medical practitioners or at different times, as these can lead to inaccurate wound size assessments. To evaluate this, we employed K-means clustering and used a QR code as a reference to analyze images of the same wound captured at varying distances, objectively quantifying the areas of 40 wounds. This study aims to develop an objective method for calculating the wound size, addressing variations in photo-taking distance that occur across different medical personnel or time points—factors that can compromise measurement accuracy. By improving consistency and reducing the manual workload, this approach also seeks to enhance the efficiency of healthcare providers. We applied K-means clustering for wound segmentation and used a QR code as a spatial reference. Images of the same wounds taken at varying distances were analyzed, and the wound areas of 40 cases were objectively quantified. Materials and Methods: We employed K-means clustering and used a QR code as a reference to analyze wound photos taken by different medical practitioners in the outpatient consulting room. K-means clustering is a machine learning algorithm that segments the wound region by grouping pixels in an image according to their color similarity. It organizes data points into clusters based on shared features. Based on this algorithm, we can use it to identify the wound region and determine its pixel area. We also used a QR code as a reference because of its unique graphical pattern. We used the printed QR code on the patient’s identification sticker as a reference for length. By calculating the ratio of the number of pixels within the square area of the QR code to its actual area, we applied this ratio to the detected wound pixel area, enabling us to calculate the wound’s actual size. The printed patient identification stickers were all uniform in size and format, allowing us to apply this method consistently to every patient. Results: The results support the accuracy of our algorithm when tested on a standard one-cent coin. The paired t-test comparing the first and second photos shot yielded a p-value of 0.370, indicating no significant difference between the two. Similarly, the t-test comparing the first and third photos shot produced a p-value of 0.179, also showing no significant difference. The comparison between the second and third photos shot resulted in a p-value of 0.547, again indicating no significant difference. Since all p-values are greater than 0.05, none of the test pairs show statistically significant differences. These findings suggest that the three randomly taken photo shots produce consistent results and can be considered equivalent. Conclusions: Our algorithm for wound area assessment is highly reliable, interchangeable, and consistently produces accurate results. This objective and practical method can aid clinical decision-making by tracking wound progression over time. Full article

This study improves existing protection strategies for image processing models by embedding invisible watermarks into model outputs to verify the sources of images. Most current methods rely on CNN-based architectures, which are limited by their local perception capabilities and struggle to effectively capture global information. To address this, we introduce the Swin-UNet, originally designed for medical image segmentation tasks, into the watermark embedding process. The Swin Transformer’s ability to capture global information enhances the visual quality of the embedded image compared to CNN-based approaches. To defend against surrogate attacks, data augmentation techniques are incorporated into the training process, enhancing the watermark extractor’s robustness specifically against surrogate attacks. Experimental results show that the proposed watermarking approach reduces the impact of watermark embedding on visual quality. On a deraining task with color images, the average PSNR reaches 45.85 dB, while on a denoising task with grayscale images, the average PSNR reaches 56.60 dB. Additionally, watermarks extracted from surrogate attacks closely match those from the original framework, with an accuracy of 99% to 100%. These results confirm the Swin Transformer’s effectiveness in preserving visual quality. Full article

Pupil segmentation in visible-light (RGB) images presents unique challenges due to variable lighting conditions, diverse eye colors, and poor contrast between iris and pupil, particularly in individuals with dark irises. While near-infrared (NIR) imaging has been the traditional solution for eye-tracking systems, the accessibility and practicality of RGB-based solutions make them attractive for widespread adoption in consumer devices. This paper presents a baseline for RGB pupil segmentation by adapting the Segment Anything Model (SAM). We introduce a multi-stage fine-tuning approach that leverages SAM’s exceptional generalization capabilities, further enhancing its elemental capacity for accurate pupil segmentation. The staged approach consists of SAM-BaseIris for enhanced iris detection, SAM-RefinedIris for improving iris segmentation with automated bounding box prompts, and SAM-RefinedPupil for precise pupil segmentation. Our method was evaluated on three standard visible-light datasets: UBIRIS.v2, I-Social DB, and MICHE-I. The results demonstrate robust performance across diverse lighting conditions and eye colors. Our method achieves near SOTA results for iris segmentation and attains mean mIOU and DICE scores of 79.37 and 87.79, respectively, for pupil segmentation across the evaluated datasets. This work establishes a strong foundation for RGB-based eye-tracking systems and demonstrates the potential of adapting foundation models for specialized medical imaging tasks. Full article

The successful application of artificial intelligence (AI) techniques for the quantitative analysis of hybrid medical imaging data such as PET/CT is challenged by the differences in the type of information and image quality between the two modalities. The purpose of this work was to develop color-based, pre-processing methodologies for PET/CT data that could yield a better starting point for subsequent diagnosis and image processing and analysis. Two methods are proposed that are based on the encoding of Hounsfield Units (HU) and Standardized Uptake Values (SUVs) in separate transformed .png files as reversible color information in combination with .png basic information metadata based on DICOM attributes. Linux Ubuntu using Python was used for the implementation and pilot testing of the proposed methodologies on brain ¹⁸F-FDG PET/CT scans acquired with different PET/CT systems. The range of HUs and SUVs was mapped using novel weighted color distribution functions that allowed for a balanced representation of the data and an improved visualization of anatomic and metabolic differences. The pilot application of the proposed mapping codes yielded CT and PET images where it was easier to pinpoint variations in anatomy and metabolic activity and offered a potentially better starting point for the subsequent fully automated quantitative analysis of specific regions of interest or observer evaluation. It should be noted that the output .png files contained all the raw values and may be treated as raw DICOM input data. Full article

Material Jetting (MJT) 3D printing (3DP) is a specific technology that deposits photocurable droplets of material and colored inks to fabricate objects layer-by-layer. The high resolution and full color capability render MJT 3DP an ideal technology for 3DP in medicine as evidenced by the 3DP literature. The technology has been adopted globally across the Americas, Europe, Asia, and Australia. While MJT 3D printers can be expensive, their ability to fabricate highly accurate and multi-color parts provides a lucrative opportunity in the creation of advanced prototypes and medical models. The literature on MJT 3DP has expanded greatly as of late, in part aided by the lowering costs of the technology, and this report is the first review to document the applications of MJT in medicine. Additionally, this report portrays the technological information behind MJT 3DP, cases involving fabricated MJT 3DP models from the University of Cincinnati 3DP lab, as well as the challenges of MJT in a clinical setting, including cost, expertise in managing the machines, and scalability issues. It is expected that MJT 3DP, as imaging and segmentation technologies undergo future improvement, will be best poised with representing the voxel-level-variations captured by radiologic-image-sets due to its capacity for voxel-level-control. Full article

Due to a significant shortage of healthcare staff, medical facilities are increasingly challenged by the need to deploy current staff more intensively, which can lead to significant complications for patients and staff. Digital surgical assistance systems that track all instruments used in procedures can make a significant contribution to relieving the load on staff, increasing efficiency, avoiding errors and improving hygiene. Due to data safety concerns, laborious data annotation and the complexity of the scenes, as well as to increase prediction accuracy, the provision of synthetic data is key to enabling the wide use of artificial intelligence for object recognition and tracking in OR settings. In this study, a synthetic data generation pipeline is introduced for the detection of eight surgical instruments during open surgery. Using 3D models of the instruments, synthetic datasets consisting of color images and annotations were created. These datasets were used to train common object detection networks (YOLOv8) and compared against networks solely trained on real data. The comparison, conducted on two real image datasets with varying complexity, revealed that networks trained on synthetic data demonstrated better generalization capabilities. A sensitivity analysis showed that synthetic data-trained networks could detect surgical instruments even at higher occlusion levels than real data-trained networks. Additionally, 1920 datasets were generated using different parameter combinations to evaluate the impact of various settings on detection performance. Key findings include the importance of object visibility, occlusion, and the inclusion of occlusion objects in improving detection accuracy. The results highlight the potential of synthetic datasets to simulate real-world conditions, enhance network generalization, and address data shortages in specialized domains like surgical instrument detection. Full article

Integrating deep learning into microbiological and cell analysis from microscopic image samples has gained significant attention in recent years, driven by the rise of novel medical technologies and pressing global health challenges. Numerous methods for segmentation and classification in microscopic images have emerged in the literature. However, key challenges persist due to the limited development of specialized deep learning models to accurately detect and quantify microorganisms and cells from microscopic samples. In response to this gap, this paper introduces MBnet, an Extreme-Lightweight Neural Network for Microbiological and Cell Analysis. MBnet is a binary segmentation method based on a Fully Convolutional Network designed to detect and quantify microorganisms and cells, featuring a low computational cost architecture with only 575 parameters. Its innovative design includes a foreground module and an encoder–decoder structure composed of traditional, depthwise, and separable convolution layers. These layers integrate color, orientation, and morphological features to generate an understanding of different contexts in microscopic sample images for binary segmentation. Experiments were conducted using datasets containing bacteria, yeast, and blood cells. The results suggest that MBnet outperforms other popular networks in the literature in counting, detecting, and segmenting cells and unicellular microorganisms. These findings underscore the potential of MBnet as a highly efficient solution for real-world applications in health monitoring and bioinformatics. Full article

Background/Objectives: Glaucomatous neuropathy, a progressive deterioration of retinal ganglion cells, is the leading cause of irreversible blindness worldwide. While elevated intraocular pressure (IOP) is a well-established modifiable risk factor, increasing attention is being directed towards IOP-independent factors, such as vascular alterations. Color Doppler imaging (CDI) is a prominent technique for investigating blood flow parameters in extraocular vessels. This prospective, nonrandomized clinical trial aimed to assess the impact of ab externo canaloplasty on ocular blood flow parameters in patients with primary open-angle glaucoma (POAG) at a three-month follow-up. Methods: Twenty-five eyes of twenty-five patients with early or moderate POAG underwent canaloplasty with simultaneous cataract removal. CDI was used to measure peak systolic velocity (PSV), end-diastolic velocity (EDV), and resistive index (RI) in the ophthalmic artery (OA), central retinal artery (CRA), and short posterior ciliary arteries (SPCAs) before and after surgery. Results: The results showed a significant reduction in IOP and improvement in mean deviation at three months post-surgery. Best corrected visual acuity and retinal nerve fiber layer thickness significantly increased at each postoperative control visit. However, no significant changes were observed in PSV, EDV, and RI in the studied vessels. Conclusions: In conclusion, while canaloplasty effectively reduced IOP and medication burden, it did not significantly improve blood flow parameters in vessels supplying the optic nerve at three months post-surgery. Careful patient selection considering glaucoma severity and vascular risk factors is crucial when choosing between canaloplasty and more invasive procedures like trabeculectomy. Further larger studies are needed to comprehensively analyze this issue. Full article

The early and accurate detection of kidney stones is crucial for effective treatment and improved patient outcomes. This paper proposes a novel modification of the YOLOv5 model, specifically tailored for detecting kidney stones in CT images. Our approach integrates the squeeze-and-excitation (SE) block within the C3 block of the YOLOv5m architecture, thereby enhancing the ability of the model to recalibrate channel-wise dependencies and capture intricate feature relationships. This modification leads to significant improvements in the detection accuracy and reliability. Extensive experiments were conducted to evaluate the performance of the proposed model against standard YOLOv5 variants (nano-sized, small, and medium-sized). The results demonstrate that our model achieves superior performance metrics, including higher precision, recall, and mean average precision (mAP), while maintaining a balanced inference speed and model size suitable for real-time applications. The proposed methodology incorporates advanced noise reduction and data augmentation techniques to ensure the preservation of critical features and enhance the robustness of the training dataset. Additionally, a novel color-coding scheme for bounding boxes improves the clarity and differentiation of the detected stones, facilitating better analysis and understanding of the detection results. Our comprehensive evaluation using essential metrics, such as precision, recall, mAP, and intersection over union (IoU), underscores the efficacy of the proposed model for detecting kidney stones. The modified YOLOv5 model offers a robust, accurate, and efficient solution for medical imaging applications and represents a significant advancement in computer-aided diagnosis and kidney stone detection. Full article

This study proposes the Pairwise Attention Enhancement (PAE) model to address the limitations of the Vision Transformer (ViT). While the ViT effectively models global relationships between image patches, it encounters challenges in medical image analysis where fine-grained local features are crucial. Although the ViT excels at capturing global interactions within the entire image, it may potentially underperform due to its inadequate representation of local features such as color, texture, and edges. The proposed PAE model enhances local features by calculating cosine similarity between the attention maps of training and reference images and integrating attention maps in regions with high similarity. This approach complements the ViT’s global capture capability, allowing for a more accurate reflection of subtle visual differences. Experiments using Clock Drawing Test data demonstrated that the PAE model achieved a precision of 0.9383, recall of 0.8916, F1-Score of 0.9133, and accuracy of 92.69%, showing a 12% improvement over API-Net and a 1% improvement over the ViT. This study suggests that the PAE model can enhance performance in computer vision fields where local features are crucial by overcoming the limitations of the ViT. Full article