Manual Segmentation Research Articles

The Effect of Ultrasound Image Pre-Processing on Radiomics Feature Quality: A Study on Shoulder Ultrasound.

Radiomics, the extraction of quantitative features from medical images, has shown great promise in enhancing diagnostic and prognostic models, particularly in CT and MRI. However, its application in ultrasound (US) imaging, especially in musculoskeletal (MSK) imaging, remains underexplored. The inherent variability of ultrasound, influenced by operator dependency and various imaging settings, presents significant challenges to the reproducibility of radiomic features. This study aims to identify whether commonly used image pre-processing methods can increase the reproducibility of radiomics features, increasing the quality of analysis. This is performed with shoulder calcific tendinopathy as a case study. Ultrasound images from 84 patients with rotator cuff calcifications were retrospectively analysed. Three pre-processing techniques-Histogram Equalization, Standard CLAHE, and Advanced CLAHE-were applied to adjust image quality. Manual segmentation of calcifications was performed, followed by the extraction of 849 radiomic features. The reproducibility of these features was assessed using the intraclass correlation coefficient (ICC), comparing results across pre-processing methods within the dataset. The Advanced CLAHE pre-processing method consistently yielded the highest ICC values, indicating superior reproducibility of radiomic features compared to other methods. Wavelet-transformed features, particularly in the GLCM and GLRLM subgroups, demonstrated robust reproducibility across all pre-processing techniques. Shape features, however, continued to show lower reproducibility. Advanced CLAHE pre-processing significantly enhances the reproducibility of radiomic features in ultrasound imaging of calcifications. This study underscores the importance of pre-processing in achieving reliable radiomic analyses, particularly in operator-dependent imaging modalities like ultrasound.

Development and External Validation of [18F]FDG PET-CT-Derived Radiomic Models for Prediction of Abdominal Aortic Aneurysm Growth Rate

Objective (1): To develop and validate a machine learning (ML) model using radiomic features (RFs) extracted from [18F]FDG PET-CT to predict abdominal aortic aneurysm (AAA) growth rate. Methods (2): This retrospective study included 98 internal and 55 external AAA patients undergoing [18F]FDG PET-CT. RFs were extracted from manual segmentations of AAAs using PyRadiomics. Recursive feature elimination (RFE) reduced features for model optimisation. A multi-layer perceptron (MLP) was developed for AAA growth prediction and compared against Random Forest (RF), XGBoost, and Support Vector Machine (SVM). Accuracy was evaluated via cross-validation, with uncertainty quantified using dropout (MLP), standard deviation (RF), and 95% prediction intervals (XGBoost). External validation used independent data from two centres. Ground truth growth rates were calculated from serial ultrasound (US) measurements or CT volumes. Results (3): From 93 initial RFs, 29 remained after RFE. The MLP model achieved an MAE ± SEM of 1.35 ± 3.2e−4 mm/year with the full feature set and 1.35 ± 2.5e−4 mm/year with RFE. External validation yielded 1.8 ± 8.9e−8 mm/year. RF, XGBoost, and SVM models produced comparable accuracies internally (1.4–1.5 mm/year) but showed higher errors during external validation (1.9–1.97 mm/year). The MLP model demonstrated reduced uncertainty with the full feature set across all datasets. Conclusions (4): An MLP model leveraging [18F]FDG PET-CT radiomics accurately predicted AAA growth rates and generalised well to external data. In the future, more sophisticated stratification could guide individualised patient care, facilitating risk-tailored management of AAAs.

Automated Detection of Change of Direction in Basketball Players Using Xsens Motion Tracking

In sports science, accurate tracking of athletes’ movement patterns is essential for performance analysis and injury prevention. Changes of direction (COD), frequently executed during basketball games at cutting angles of around 135° (internal angle of 45°), are essential for agility and high-level performance. Moreover, mastering effective COD mechanics is associated with a lower risk of injuries and enhanced long-term athletic success. However, manual segmentation of data from wearable sensors is labor-intensive and time-consuming, often creating bottlenecks for sports practitioners. The aim of this study was to evaluate the feasibility and accuracy of an automated algorithm for detecting COD movements in basketball and to compare its performance with manual detection methods. Data were collected from 62 basketball players, each completing two tests (V-cut test and a modified V-cut test), totaling 248 trials. The system utilizes kinematic data from an Xsens full-body kit to analyze key variables that characterize direction changes. The proposed method detects COD events with a median error of one frame and an interquartile range of two frames. The system demonstrated nearly 80% accuracy in COD detection, as validated against manual video analysis. These findings indicate that automated COD detection can significantly reduce segmentation time for practitioners while providing actionable, data-driven insights to enhance kinematic assessment during sport-specific activities.

Segmentation of cortical bone, trabecular bone, and medullary pores from micro-CT images using 2D and 3D deep learning models.

Computed tomography (CT) enables rapid imaging of large-scale studies of bone, but those datasets typically require manual segmentation, which is time-consuming and prone to error. Convolutional neural networks (CNNs) offer an automated solution, achieving superior performance on image data. In this methodology-focused paper, we used CNNs to train segmentation models from scratch on 2D and 3D patches from micro-CT scans of otter long bones. These new models, collectively called BONe (Bone One-shot Network), aimed to be fast and accurate, and we expected enhanced results from 3D training due to better spatial context. Contrary to expectations, 2D models performed slightly better than 3D models in labeling details such as thin trabecular bone. Although lacking in some detail, 3D models appeared to generalize better and predict smoother internal surfaces than 2D models. However, the massive computational costs of 3D models limit their scalability and practicality, leading us to recommend 2D models for bone segmentation. BONe models showed potential for broader applications with variation in performance across species and scan quality. Notably, BONe models demonstrated promising results on skull segmentation, suggesting their potential utility beyond long bones with further refinement and fine-tuning.

Quantifying the tumour vasculature environment from CD-31 immunohistochemistry images of breast cancer using deep learning based semantic segmentation

BackgroundTumour vascular density assessed from CD-31 immunohistochemistry (IHC) images has previously been shown to have prognostic value in breast cancer. Current methods to measure vascular density, however, are time-consuming, suffer from high inter-observer variability and are limited in describing the complex tumour vasculature morphometry.MethodsWe propose a method for automatically measuring a range of vascular parameters from CD-31 IHC images, which together provide a detailed description of the vasculature morphology. We first used a U-Net based convolutional neural network, trained and validated using 36 partially annotated whole slide images from 27 patients, to segment vessel structures and tumour regions from which the measurements are taken. The model also segments the vascular smooth muscle, benign epithelium, adipose tissue, stroma, lymphocyte clusters, nerves and CD-31 positive leukocytes, and we applied it to an additional 21 images from 15 patients. Using these segmentations, we investigated the relationship between the various tissue types and the vasculature and studied the relationship of various vascular parameters with clinical parameters. We also performed a 3D histology analysis on a separate tumour sample as a proof of principle, providing a more comprehensive visualization of vasculature morphology compared to the standard 2D cross-section of a tissue sample.ResultsUsing two-way cross-validation, we show that vessels were accurately segmented, with Dice scores of 0.875 and 0.856, and were accurately identified, with F1 scores of 0.777 and 0.748. All vascular parameters exhibit strong (r>0.70.7$$\\end{document}]]>) and significant (p<0.001) correlations with measurements taken from the manual ground truth vessel segmentations. A significant relationship between the major/minor axis ratio, a measure of elongation, and the tumour grade was found.ConclusionOur proposed method shows promise as a tool for studying the tumour vasculature and its relationship with surrounding cells and tissue types. Furthermore, the correlation with tumour grade highlights the clinical relevance of our approach. These findings suggest that our method could have substantial implications for improving prognostic assessments and personalizing therapeutic strategies in breast cancer treatment.

Vision transformers for automated detection of diabetic peripheral neuropathy in corneal confocal microscopy images

Early detection and management of diabetic peripheral neuropathy (DPN) are critical to reducing associated morbidity and mortality. Corneal Confocal Microscopy (CCM) facilitates the imaging of corneal nerves to detect early and progressive nerve damage in DPN. However, its wider adoption has been limited by the subjectivity and time-intensive nature of manual nerve fiber quantification. This study investigates the diagnostic utility of state-of-the-art Vision Transformer (ViT) models for the binary classification of CCM images to distinguish between healthy controls and individuals with DPN. The ViT model's performance was also compared to ResNet50, a convolutional neural network (CNN) previously applied for DPN detection using CCM images. Using a dataset of approximately 700 CCM images, the ViT model achieved an AUC of 0.99, a sensitivity of 98%, a specificity of 92%, and an F1-score of 95%, outperforming previously reported methods. These findings highlight the potential of the ViT model as a reliable tool for CCM-based DPN diagnosis, eliminating the need for time-consuming manual image segmentation. Moreover, the results reinforce CCM's value as a non-invasive and precise imaging modality for detecting nerve damage, particularly in neuropathy-related conditions such as DPN.

Deep learning and radiomics for gastric cancer serosal invasion: automated segmentation and multi-machine learning from two centers

ObjectiveThe objective of this study is to develop an automated method for segmenting spleen computed tomography (CT) images using a deep learning model. This approach is intended to address the limitations of manual segmentation, which is known to be susceptible to inter-observer variability. Subsequently, a prediction model of gastric cancer (GC) serosal invasion was constructed in conjunction with radiomics and deep learning features, and a nomogram was generated to explore the clinical guiding significance.MethodsThis study enrolled 311 patients from two centers with pathologically confirmed of GC. we employed a deep learning model, U-Mamba, to obtain fully automatic segmentation of the spleen CT images. Subsequently, radiomics features and deep learning features were extracted from the entire spleen CT images, and significant features were identified through dimensionality reduction. The clinical features, radiomic features, and deep learning features were organized and integrated, and five machine learning methods were employed to develop 15 predictive models. Ultimately, the model exhibiting superior performance was presented in the form of a nomogram.ResultsA total of 18 radiomics features, 30 deep learning features, and 1 clinical features were deemed valuable. The DLRA model demonstrated superior discriminative capacity relative to other models. A nomogram was constructed based on the logistic clinical model to facilitate the usage and verification of the clinical model.ConclusionRadiomics and deep learning features derived from automated spleen segmentation to construct a nomogram demonstrate efficacy in predicting serosal invasion in GC. Concurrently, fully automated segmentation provides a novel and reproducible approach for radiomics research.

Automated contouring for breast cancer radiotherapy in the isocentric lateral decubitus position: aneural network-based solution for enhanced precision and efficiency.

Adjuvant radiotherapy is essential for reducing local recurrence and improving survival in breast cancer patients, but it carries arisk of ischemic cardiac toxicity, which increases with heart exposure. The isocentric lateral decubitus position, where the breast rests flat on asupport, reduces heart exposure and leads to delivery of amore uniform dose. This position is particularly beneficial for patients with unique anatomies, such as those with pectus excavatum or larger breast sizes. While artificial intelligence (AI) algorithms for autocontouring have shown promise, they have not been tailored to this specific position. This study aimed to develop and evaluate aneural network-based autocontouring algorithm for patients treated in the isocentric lateral decubitus position. In this single-center study, 1189 breast cancer patients treated after breast-conserving surgery were included. Their simulation CT scans (1209 scans) were used to train and validate aneural network-based autocontouring algorithm (nnU-Net). Of these, 1087 scans were used for training, and 122 scans were reserved for validation. The algorithm's performance was assessed using the Dice similarity coefficient (DSC) to compare the automatically delineated volumes with manual contours. Aclinical evaluation of the algorithm was performed on 30additional patients, with contours rated by two expert radiation oncologists. The neural network-based algorithm achieved asegmentation time of approximately 4 min, compared to 20 min for manual segmentation. The DSC values for the validation cohort were 0.88 for the treated breast, 0.90 for the heart, 0.98 for the right lung, and 0.97 for the left lung. In the clinical evaluation, 90% of the automatically contoured breast volumes were rated as acceptable without corrections, while the remaining 10% required minor adjustments. All lung contours were accepted without corrections, and heart contours were rated as acceptable in 93.3% of cases, with minor corrections needed in 6.6% of cases. This neural network-based autocontouring algorithm offers apractical, time-saving solution for breast cancer radiotherapy planning in the isocentric lateral decubitus position. Its strong geometric performance, clinical acceptability, and significant time efficiency make it avaluable tool for modern radiotherapy practices, particularly in high-volume centers.

ANALYSIS OF HENLE FIBER LAYER AND OUTER RETINAL LAYERS IN CONE DYSTROPHY.

To evaluate Henle fiber layer (HFL) thickness and volume parameters in patients with cone photoreceptor atrophy with directional optical coherence tomography. Macular 20°×20° standard and directional optical coherence tomography images were acquired from patients diagnosed with hereditary cone dystrophy with evident foveal ellipsoid zone defect in optical coherence tomography and age-matched healthy controls. Thickness and volume parameters of HFL, outer nuclear layer (ONL), and retinal layers between ellipsoid zone and Bruch membrane complex (EZ-BM) were calculated from manual segmentation through directional optical coherence tomography images, and comparative analysis is performed. Twelve eyes of six patients were compared with 12 eyes of six age-matched healthy controls (mean age: 29.5 ± 16.6 and 26.6 ± 3.9 years, respectively; P = 0.162). Patients had lower total HFL volume (0.45 ± 0.03 and 0.85 ± 0.15 mm 3 ; P < 0.001) and mean HFL thickness (16.1 ± 1.1 and 30.1 ± 5.3 µ m; P < 0.001) than healthy controls. Central subfield, parafoveal, and perifoveal ETDRS zone HFL parameters in patients were significantly lower than healthy controls. A centrifugal correlation was found between central outer nuclear layer and the corresponding parafoveal HFL (Spearman rho: 0.785; P < 0.001). Henle fiber layer assessment might be a useful optical coherence tomography biomarker in patients with cone photoreceptor atrophy. Henle fiber layer thinning is observed in foveal, parafoveal, and perifoveal areas of patients with cone photoreceptor atrophy, while volume reduction in outer nuclear layer and ellipsoid zone and Bruch membrane complex components were limited to central and parafoveal zones.

Development and validation of fully automated robust deep learning models for multi-organ segmentation from whole-body CT images.

This study aimed to develop a deep-learning framework to generate multi-organ masks from CT images in adult and pediatric patients. A dataset consisting of 4082 CT images and ground-truth manual segmentation from various databases, including 300 pediatric cases, were collected. In strategy#1, the manual segmentation masks provided by public databases were split into training (90%) and testing (10% of each database named subset #1) cohort. The training set was used to train multiple nnU-Net networks in five-fold cross-validation (CV) for 26 separate organs. In the next step, the trained models from strategy #1 were used to generate missing organs for the entire dataset. This generated data was then used to train a multi-organ nnU-Net segmentation model in a five-fold CV (strategy#2). Models' performance were evaluated in terms of Dice coefficient (DSC) and other well-established image segmentation metrics. The lowest CV DSC for strategy#1 was 0.804±0.094 for adrenal glands while average DSC > 0.90 were achieved for 17/26 organs. The lowest DSC for strategy#2 (0.833±0.177) was obtained for the pancreas, whereas DSC > 0.90 was achieved for 13/19 of the organs. For all mutual organs included in subset #1 and subset #2, our model outperformed the TotalSegmentator models in both strategies. In addition, our models outperformed the TotalSegmentator models on subset #3. Our model was trained on images with significant variability from different databases, producing acceptable results on both pediatric and adult cases, making it well-suited for implementation in clinical setting.

Assessment of deep learning technique for fully automated mandibular segmentation.

This study aimed to assess the precision of an open-source, clinician-trained, and user-friendly convolutional neural network-based model for automatically segmenting the mandible. A total of 55 cone-beam computed tomography scans that met the inclusion criteria were collected and divided into test and training groups. The MONAI (Medical Open Network for Artificial Intelligence) Label active learning tool extension was used to train the automatic model. To assess the model's performance, 15 cone-beam computed tomography scans from the test group were inputted into the model. The ground truth was obtained from manual segmentation data. Metrics including the Dice similarity coefficient, Hausdorff 95%, precision, recall, and segmentation times were calculated. In addition, surface deviations and volumetric differences between the automated and manual segmentation results were analyzed. The automated model showed a high level of similarity to the manual segmentation results, with a mean Dice similarity coefficient of 0.926 ± 0.014. The Hausdorff distance was 1.358 ± 0.466 mm, whereas the mean recall and precision values were 0.941 ± 0.028 and 0.941 ± 0.022, respectively. There were no statistically significant differences in the arithmetic mean of the surface deviation for the entire mandible and 11 different anatomic regions. In terms of volumetric comparisons, the difference between the 2 groups was 1.62 mm³, which was not statistically significant. The automated model was found to be suitable for clinical use, demonstrating a high degree of agreement with the reference manual method. Clinicians can use open-source software to develop custom automated segmentation models tailored to their specific needs.

Deep Learning Segmentation-Based Bone Removal from Computed Tomography of the Brain Improves Subdural Hematoma Detection

Timely identification of intracranial blood products is clinically impactful, however the detection of subdural hematoma (SDH) on non-contrast CT scans of the head (NCCTH) is challenging given interference from the adjacent calvarium. This work explores the utility of a NCCTH bone removal algorithm for improving SDH detection. A deep learning segmentation algorithm was designed/trained for bone removal using 100 NCCTH. Segmentation accuracy was evaluated on 15 NCCTH from the same institution and 22 NCCTH from an independent external dataset using quantitative overlap analysis between automated and expert manual segmentations. The impact of bone removal on detecting SDH by junior radiology trainees was evaluated with a reader study comparing detection performance between matched cases with and without bone removal applied. Average Dice overlap between automated and manual segmentations from the internal and external test datasets were 0.9999 and 0.9957, which was superior to other publicly available methods. Among trainee readers, SDH detection was statistically improved using NCCTH with and without bone removal applied compared to standard NCCTH alone (P value <0.001). Additionally, 12/14 (86 %) of participating trainees self-reported improved detection of extra axial blood products with bone removal, and 13/14 (93 %) indicated that they would like to have access to NCCTH bone removal in the on-call setting. Deep learning segmentation-based NCCTH bone removal is rapid, accurate, and improves detection of SDH among trainee radiologists when used in combination with standard NCCTH. This study highlights the potential of bone removal for improving confidence and accuracy of SDH detection.

Detection of Brain Tumor Employing Residual Network-based Optimized Deep Learning.

Diagnosis and treatment planning play a very vital role in improving the survival of oncological patients. However, there is high variability in the shape, size, and structure of the tumor, making automatic segmentation difficult. The automatic and accurate detection and segmentation methods for brain tumors are proposed in this paper. A modified ResNet50 model was used for tumor detection, and a ResUNetmodel-based convolutional neural network for segmentation is proposed in this paper. The detection and segmentation were performed on the same dataset consisting of pre-contrast, FLAIR, and postcontrast MRI images of 110 patients collected from the cancer imaging archive. Due to the use of residual networks, the authors observed improvement in evaluation parameters, such as accuracy for tumor detection and dice similarity coefficient for tumor segmentation. The accuracy of tumor detection and dice similarity coefficient achieved by the segmentation model were 96.77% and 0.893, respectively, for the TCIA dataset. The results were compared based on manual segmentation and existing segmentation techniques. The tumor mask was also individually compared to the ground truth using the SSIM value. The proposed detection and segmentation models were validated on BraTS2015 and BraTS2017 datasets, and the results were consensus. The use of residual networks in both the detection and the segmentation model resulted in improved accuracy and DSC score. DSC score was increased by 5.9% compared to the UNet model, and the accuracy of the model was increased from 92% to 96.77% for the test set.

Automated radiomics model for preoperative pancreatic neuroendocrine tumor grade prediction.

655 Background: Pancreatic neuroendocrine tumors (PNETs) have varying biologic behavior based on factors including tumor grade. Many prognostic factors can only be determined post-pancreatectomy; thus, are unable to guide clinical decision making preoperatively. This is particularly relevant in small, non-functional (NF)-PNETs. Imaging analysis with radiomics may be able to elucidate biologic data such as tumor grade, without the need for tissue sampling. Our study aims to create an automatic pipeline from segmentation to radiomics-signature building to preoperatively predict tumor grade. Methods: Patients that underwent resection between 2003-21 with adequate preoperative arterial phase CT scans were divided into training and validation cohorts. In the training cohort, the pancreas and tumor region were manually segmented and used to train an auto-segmentation model. The validation cohort then underwent automatic segmentation. A total of 255 radiomics features were extracted from the tumor region. Correlated features were removed, and the minimum redundancy maximum relevance (mRMR) method was used to select the final feature set. These features were used to train a Support Vector Machine (SVM)-based classifier through the training cohort to predict grade, which was dichotomized into grade I versus II/III. The radiomic based prediction model was then evaluated in the automatically segmented validation cohort. Results: A total of 186 patients were identified and divided into training (n = 140) and validation (n = 46) cohorts. Median age was 57 (27, 85) and 52% were male. 113 (62%) and 70 (38%) patients were grades I and II/III, respectively. Median tumor size was 24 mm (6, 200) and 43 (26%) patients had positive nodal disease. The auto-segmentation model was able to accurately segment the tumor region in 33 (72%) patients of the validation cohort. In the training cohort (n = 140), the radiomics-signature produced an area under the curve (AUC) of 0.87 (0.81, 0.93). Sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were 0.94 (0.9, 0.98), 0.74 (0.66, 0.81), 0.72 (0.65, 0.80) and 0.97 (0.94, 1.00) respectively. In the auto-segmented validation cohort (n = 33), the radiomics model produced an AUC of 0.75 (0.61, 0.90). Sensitivity, specificity, PPV and NPV were 0.88 (0.77, 0.99), 0.62 (0.46, 0.79), 0.71 (0.56, 0.87) and 0.83 (0.71, 0.96), respectively. Conclusions: The proposed auto-segmentation model was able to automatically identify tumor regions with a high degree of accuracy. Similarly, the subsequent radiomics-signature was also able to strongly discern PNET grade. This automatic pipeline bypasses manual segmentation and could help incorporate a radiomics-signature into preoperative clinical decision-making for PNETs.

Automated Dental Registration and TMJ Segmentation for Virtual Surgical Planning of Orthognathic Surgery via Three-Step Computer-Based Method

Objective: This study developed and evaluated a computer-based method for automating the registration of scanned dental models with 3D reconstructed skulls and segmentation of the temporomandibular joint (TMJ). Methods: A dataset comprising 1274 skull models and corresponding scanned dental models was collected. In total, 1066 cases were used for the development of the computer-based method, while 208 cases were used for validation. Performance was evaluated by comparing the automated results with manual registration and segmentation performed by clinicians, using accuracy and completeness metrics (e.g. intersection of union [IoU] and Dice similarity coefficient [DSC]). Results: The automated registration achieved a mean absolute error of 0.35 mm for the maxilla and 0.38 mm for the mandible, and a root mean squared error of 0.46 mm and 0.39 mm, respectively. The automatic TMJ segmentation exhibited an accuracy of 97.48%, a precision of 97.06%, a IoU of 95.72%, DSC of 97.3%, and a Hausdorff value of 1.87 mm, which were sufficient for clinical application. Conclusion: The proposed method significantly improved the efficiency of orthognathic surgical planning by automating the registration and segmentation processes. The accuracy and precision of the automated results were sufficient for clinical use, reducing the workload on clinicians and facilitating faster and more reliable surgical planning. Clinical significance: The computer-based method streamlines orthognathic surgical planning, enhancing precision and efficiency without compromising clinical accuracy, ultimately improving patient outcomes and reducing the workload of surgeons.

An Overview of Machine Learning Techniques for MRI-Based Knee Osteoarthritis Progression Prediction

One common degenerative joint condition that has a substantial negative influence on quality of life and is placing an increasing strain on healthcare systems throughout the globe is knee osteoarthritis (OA). Even while MRI is essential for OA diagnosis and monitoring, conventional techniques that depend on manual cartilage segmentation are labour-intensive and subject to error. In order to overcome these obstacles, this work uses the cutting-edge Cartilage Damage Index (CDI) to present a unique machine learning-based method for OA progression prediction using MRI data. More sensitive evaluations of cartilage changes over time are made possible by the CDI, which measures the thickness and structural integrity of cartilage at 36 distinct places inside the tibiofemoral cartilagecompartment. Principal Component Analysis (PCA) is used to improve prediction accuracy, increase computing efficiency, and reduce dimensionality while maintaining important information. This study evaluates four machine learning algorithms ANN, SVM, Naïve Bayes &amp; Random Forest for its efficacy in predicting the progression of osteoarthritis (OA) based on data obtained from CDI. Validated prediction findings utilise established clinical markers such as JSN in the medial and lateral compartments and abnormalities in KL grades. The findings indicate that the medial cartilage feature set had superior predictive capacity, with the highest overall accuracy achieved with the integration of medial and lateral features. This work provides a robust basis for the early diagnosis and monitoring of OA progression by integrating advanced data mining techniques wi machine learning, allowing more personalised and effective treatment strategies in clinical practice. Key Words: Cartilage Damage Index, Machine learning, Knee osteoarthritis, MRI-based prediction.

Glaucoma detection: Binocular approach and clinical data in machine learning.

In this work, we present a multi-modal machine learning method to automate early glaucoma diagnosis. The proposed methodology introduces two novel aspects for automated diagnosis not previously explored in the literature: simultaneous use of ocular fundus images from both eyes and integration with the patient's additional clinical data. We begin by establishing a baseline, termed monocular mode, which adheres to the traditional approach of considering the data from each eye as a separate instance. We then explore the binocular mode, investigating how combining information from both eyes of the same patient can enhance glaucoma diagnosis accuracy. This exploration employs the PAPILA dataset, comprising information from both eyes, clinical data, ocular fundus images, and expert segmentation of these images. Additionally, we compare two image-derived data modalities: direct ocular fundus images and morphological data from manual expert segmentation. Our method integrates Gradient-Boosted Decision Trees (GBDT) and Convolutional Neural Networks (CNN), specifically focusing on the MobileNet, VGG16, ResNet-50, and Inception models. SHAP values are used to interpret GBDT models, while the Deep Explainer method is applied in conjunction with SHAP to analyze the outputs of convolutional-based models. Our findings show the viability of considering both eyes, which improves the model performance. The binocular approach, incorporating information from morphological and clinical data yielded an AUC of 0.796 (±0.003 at a 95% confidence interval), while the CNN, using the same approach (both eyes), achieved an AUC of 0.764 (±0.005 at a 95% confidence interval).

A novel AI model for detecting periapical lesion on CBCT: CBCT-SAM.

Periapical lesions are not always evident on radiographic scans. Sometimes, asymptomatic or initial periapical lesions on cone-beam computed tomography (CBCT) could be missed by inexperienced dentists, especially when the scan has a large field of view and is not for endodontic treatment purposes. Previously, numerous algorithms have been introduced to assist radiographic assessment and diagnosis in the field of endodontics. This study aims to investigate the efficacy of CBCT-SAM, a new artificial intelligence (AI) model, in identifying periapical lesions on CBCT. Model training and validation in this study was performed using 185 CBCT scans with confirmed periapical lesions. Manual segmentation labels were prepared by a trained operator and validated by a maxillofacial radiologist. The diagnostic and segmentation performances of four AI models were evaluated and compared: CBCT-SAM, CBCT-SAM without progressive Prediction Refinement Module (PPR), and two previously developed models: Modified U-Net and PAL-Net. Accuracy was used to evaluated the diagnostic performance of the models, and accuracy, sensitivity, specificity, precision and Dice Similarity Coefficient (DSC) were used to evaluate the models' segmentation performance. CBCT-SAM achieved an average diagnostic accuracy of 98.92% ± 010.37% and an average segmentation accuracy of 99.65% ± 0.66%. The average sensitivity, specificity, precision and DSC were 72.36 ± 21.61%, 99.87% ± 0.11%, 0.73 ± 0.21 and 0.70 ± 0.19. CBCT-SAM and PAL-Net performed significantly better than Modified U-Net in segmentation accuracy (p = 0.023, p = 0.041), sensitivity (p = 0.000, p = 0.002), and DSC (p= 0.001, p= 0.004). There is no significant difference between CBCT-SAM, CBCT-SAM without PPR and PAL-Net. However, with PPR incorporated into the model, CBCT-SAM slightly surpassed PAL-Net in the diagnostic and segmentation tasks. CBCT-SAM is capable of providing expert-level assistance in the identification of periapical lesions on CBCT. The application of artificial intelligence could increase dentists' chairside diagnostic accuracy and efficiency. By assisting radiographic assessment, such as periapical lesions on CBCT, it helps reduce the chance of missed diagnosis by human errors and facilitates early detection and treatment of dental pathologies at the early stage.

Cone-Beam CT Segmentation for Intraoperative Electron Radiotherapy Based on U-Net Variants with Transformer and Extended LSTM Approaches

AI applications are increasingly prevalent in radiotherapy, including commercial software solutions for automatic segmentation of anatomical structures for 3D CT. However, their use in intraoperative electron radiotherapy (IOERT) remains limited. In particular, no AI solution is available for contouring cone beam CT (CBCT) images acquired with a mobile CBCT device. The U-Net convolutional neural network architecture has gained huge success for medical image segmentation but still has difficulties capturing the global context. To increase the accuracy in CBCT segmentation for IOERT, three different AI architectures were trained and evaluated. The features of the natural language processing models Transformer and xLSTM were added to the popular U-Net architecture and compared with the standard U-Net and manual segmentation performance. These networks were trained and tested using 55 CBCT scans obtained from breast cancer patients undergoing IOERT in the department of radiotherapy and radiation oncology in Salzburg, and each architecture’s segmentation performance was assessed using the dice coefficient (DSC) as a similarity measure. The average DSC values were 0.83 for the standard U-Net, 0.88 for the U-Net with transformer features, and 0.66 for the U-Net with xLSTM. The hybrid U-Net architecture, including Transformer features, achieved the best segmentation accuracy, demonstrating an improvement of 5% on average over the standard U-Net, while the U-Net with xLSTM showed inferior performance compared to the standard U-Net. With the help of automatic contouring, synthetic CT images can be generated, and IOERT challenges related to the time-consuming nature of 3D image-based treatment planning can be addressed.

Transformers for Neuroimage Segmentation: Scoping Review.

Neuroimaging segmentation is increasingly important for diagnosing and planning treatments for neurological diseases. Manual segmentation is time-consuming, apart from being prone to human error and variability. Transformers are a promising deep learning approach for automated medical image segmentation. This scoping review will synthesize current literature and assess the use of various transformer models for neuroimaging segmentation. A systematic search in major databases, including Scopus, IEEE Xplore, PubMed, and ACM Digital Library, was carried out for studies applying transformers to neuroimaging segmentation problems from 2019 through 2023. The inclusion criteria allow only for peer-reviewed journal papers and conference papers focused on transformer-based segmentation of human brain imaging data. Excluded are the studies dealing with nonneuroimaging data or raw brain signals and electroencephalogram data. Data extraction was performed to identify key study details, including image modalities, datasets, neurological conditions, transformer models, and evaluation metrics. Results were synthesized using a narrative approach. Of the 1246 publications identified, 67 (5.38%) met the inclusion criteria. Half of all included studies were published in 2022, and more than two-thirds used transformers for segmenting brain tumors. The most common imaging modality was magnetic resonance imaging (n=59, 88.06%), while the most frequently used dataset was brain tumor segmentation dataset (n=39, 58.21%). 3D transformer models (n=42, 62.69%) were more prevalent than their 2D counterparts. The most developed were those of hybrid convolutional neural network-transformer architectures (n=57, 85.07%), where the vision transformer is the most frequently used type of transformer (n=37, 55.22%). The most frequent evaluation metric was the Dice score (n=63, 94.03%). Studies generally reported increased segmentation accuracy and the ability to model both local and global features in brain images. This review represents the recent increase in the adoption of transformers for neuroimaging segmentation, particularly for brain tumor detection. Currently, hybrid convolutional neural network-transformer architectures achieve state-of-the-art performances on benchmark datasets over standalone models. Nevertheless, their applicability remains highly limited by high computational costs and potential overfitting on small datasets. The heavy reliance of the field on the brain tumor segmentation dataset hints at the use of a more diverse set of datasets to validate the performances of models on a variety of neurological diseases. Further research is needed to define the optimal transformer architectures and training methods for clinical applications. Continuing development may make transformers the state-of-the-art for fast, accurate, and reliable brain magnetic resonance imaging segmentation, which could lead to improved clinical tools for diagnosing and evaluating neurological disorders.