Image Processing & Applications

Purpose: Planar cardiac scintigraphy with [123I]I-mIBG is widely used to image myocardial sympathetic activity. In multicentre studies, different γ-cameras can cause variability in quantitative parameters. This study aimed to harmonise multicentre [123I]I-mIBG data using a custom-designed phantom and to assess its feasibility and acceptability. Methods: A ‘tube and bottle' phantom was designed to standardise the heart-to-mediastinum (H/M) ratio across 15 centres. Each centre prepared three versions of the phantom (A, B, and C) with varying pertechnetate ([99mTc]Tc-O4-) activities, acquired static images using their own γ-camera, and uploaded DICOM data to a shared platform. In the phantom, the tube and bottle represent the heart and mediastinum, respectively, with the tube-to-bottle (T/B) ratio simulating the H/M ratio. The reference centre analysed the images and calculated the T/B ratios, applying linear regression for data harmonisation. A survey was conducted to assess the phantom's usability. Results: The harmonised T/B ratios for versions A and B were 20.46 ± 0.78 and 6.19 ± 0.39, respectively. The average slope and intercept of the regression line across the participating centres resulted in 1.07 ± 0.38 and − 0.82 ± 4.95. Survey feedback indicated high feasibility and acceptability, with all centres recommending the phantom for multicentre harmonisation.

Keywords: Artificial intelligence , Heart-to-mediastinum ratio , Multicentre studies , Radiomics , [¹²³I]I-mIBG , Γ-camera

The Legnaro National Laboratories (LNL) of the Italian Institute of Nuclear Physics (INFN) host the Selective Production of Exotic Species (SPES) facility, which will employ the Isotope Separation On-Line (ISOL) method to generate Radioactive Ion Beams (RIBs) for scientific research. At SPES, a proton beam interacts with a primary target, producing radioactive neutral isotopes subsequently ionized for experimentation. The FEBIAD type ion source, essential for the ionization process, operates at extreme working conditions (temperature above 2000 ℃ and high vacuum regime) and faces challenges due to conventional manufacturing limitations affecting its performance stability. The Laser Powder Bed Fusion (LPBF) technology emerges as a potential solution to produce the ion source parts overcoming such limits. In this paper, experimental methods for assessing LPBF components behavior at the effective working conditions are detailed. Specifically, an accurate measurement of deformation at high temperature caused by thermal expansion remains crucial, necessitating an updated experimental setup incorporating a vision system.

Keywords: experimental method , FEBIAD ion source , ISOL facility , Vision system

The development of additive manufacturing has made these technologies suitable for fabricating end products. This encourages companies to identify quickly parts in large databases for which switching from traditional manufacturing technologies to additive manufacturing is convenient. Typically, the manufacturing process selection is made by experts who weigh various parameters, but evidence suggests that intelligent systems could beneficially replace or aid this manual selection. One challenge in using manufacturing data for advanced analysis and machine learning is that it is usually unlabeled, and manual data labelling is expensive and time-consuming. This paper deals with the application of an enhanced unsupervised learning algorithm that automatically identifies parts suitable for additive manufacturing based on parts geometry as a preliminary step of process selection. One hundred randomly selected parts were evaluated by manufacturing experts through a survey and then clustered by the proposed algorithm. The comparison of the manual and algorithmic classifications, using unsupervised learning, regarding suitability for additive or traditional manufacturing is the main original contribution of this study. Overall, 78% convergence between most experts' designations and the unsupervised learning algorithm is achieved. For those parts where expert opinions are substantially aligned, the algorithm showed a 90% convergence rate with human choices.

Keywords: Additive manufacturing , Computer-aided design , Hierarchical clustering , Image processing , Machine learning , Unsupervised learning

With approximately 431,000 cases, kidney cancer ranks 14th in terms of diagnostic frequency and accounts for 2.2% of all new cancer cases worldwide.12. For malignant clear cell renal cell carcinoma (ccRCC), which necessitates WHO/ISUP grading based on tissue microscopic analysis, early diagnosis is critical for targeted treatment. With the use of a convolutional neural network with three blocks, this study presents a machine learning-based system to grade kidney tumors using CT scans. The system divides tumours into low-grade (grades 1 and 2) and high-grade (grades 3 and 4), which corresponds to the tumour's aggressiveness and helps determine the prognosis. The challenge kits21 database's 237 preoperative CT scans were used to train the model, which yielded an accuracy of 93.67% and an f-score of 91.80%. For reliable performance validation, the approach consists of data preprocessing, augmentation, and a binary classifier with a majority voting ensemble model. The objective of this non-invasive method is to improve diagnostic accuracy in clinical settings by decreasing inter-observer variability. The dataset will be enlarged in future work to enable wider application. Although there is hope for the system's clinical support, more validation is necessary before generalization can occur.

Keywords: AI , ccRCC , Deep learning , grading , image processing , kidney cancer

This paper aims to provide an updated overview about the Human Pose Estimation (HPE), which is a typical Computer Vision-related problem. The review firstly analyzes the importance of the research topic, both in terms of number of publications and application fields. Thanks to the introduction of efficient machine learning techniques, nowadays HPE is feasible without the necessity of specialized and expensive instrumentations. The proposed review considers only the works based on the use of Machine Learning techniques, and it is divided in two parts: an analysis of methodologies presented for the 2D HPE and a similar one for the 3D HPE. Being this last generally approached as an extension of the 2D HPE, the Authors propose a section discussing the effective validity of the most diffused and open source 2D HPE methodologies.

Keywords: Computer Vision , Human Pose Estimation , Human Posture , Movement analysis

The shape and tuning of bells provide helpful information about where the instrument was manufactured or located and the casting and tuning processes used. For this purpose, both shapes catalogues and methodologies for physical modelling and simulations of musical instruments' dynamical behaviour are presented in the literature. In several areas of the world, medieval or Renaissance bells are rare because of the remelting of the vast majority of specimens following fractures or breaks caused by wear and tear or traumatic events, such as earthquakes. As a result, the original geometric shape can only be reconstructed by analysing the relevant melting pots excavated by archaeological campaigns. Traditionally, such an analysis is carried out manually by archaeologists. The results are blurred by the difficulty in determining the axis of symmetry and the profiles of small fragments; furthermore, the fragments can rarely be reassembled on the fracture surfaces. To address these drawbacks, this research suggests a computer-based method for virtually reconstructing the bell shape from mantle fragments.

Keywords: 3D archeology , Automatic feature recognition , Cultural heritage , Medieval Bells

Background: Anterior cruciate ligament (ACL) reconstruction is a widely performed procedure for ACL injury, but there are several factors which may lead to re-rupture or clinical failure. An intercondylar notch (or fossa) that is narrower may increase the likelihood of injury. Traditional two-dimensional assessments are limited, and three-dimensional (3D) volume analysis may offer more detailed insights. This study employs deep learning and statistical shape modeling (SSM) to enhance 3D modeling of the intercondylar notch, aiming to gain a deeper understanding of this complex 3D anatomical region. Methods: A methodology was developed to generate accurate 3D models of the intercondylar fossa within seconds. The variability of the intercondylar notch in ACL-injured samples was analyzed using SSM techniques, focusing on its principal components. Additionally, gender differences in notch volume were examined using t-tests. Results: The best deep learning method for automatic segmentation of the notch was SegResNet, which achieved a Dice similarity coefficient of over 0.88 and a Hausdorff distance of 0.73 mm. The small volume-related relative error (0.06) illustrates the goodness of the result. Three principal components accounted for 72.59% of the variation, including notch volume, shape, width, and height. Females had statistically significant smaller notch compared with males with ACL injury (P < 0.001).

Keywords: ACL injury , Anatomical variation , Automatic segmentation , Deep learning , Principal component analysis , Statistical shape modelling

Purpose: Minimally invasive surgery (MIS) has emerged in clinical practice to minimize surgical trauma, providing patients with faster recovery, reduced pain and complications and enhanced aesthetic results compared to traditional surgery. However, this approach increase the risk of iatrogenic damage, i.e. the accidental injury to sensitive anatomical structures (eg. nerves and vascular strcuteres) not directly visible during a percutaneous access. Augmented reality (AR) can effectively mitigate these drawbacks by overlaying graphical information onto the surgical field and providing real-time feedback, offering support in training settings and clinical practice. Implementation challenges have limited the number of case studies in the scientific literature. This study presents a novel simulation paradigm for orthopedic surgery training, filling a gap in surgical skill development resources for trainees, and demonstrating the effectiveness of this approach. Methods: The proposed methodology provides a framework for building a cost-effective and easily reproducible surgical training simulation environment. To address the challenge of mental spatial navigation during MIS procedures, the framework's rationale is to address the challenge of mental spatial navigation during MIS procedures. A surgical gesture tracking system using a commercial depth camera for comfortable simulation was developed. The principles of the acquisition system, image processing, and spatial computation mechanics are detailed to illustrate the framework's applicability. Digital environments customization with game engines to simulate expensive medical instrumentation, such as the C-arm, is also demonstrated. The simulation platform comprises a Computer Vision (CV) module, an X-ray machine simulation module, and an AR module. Results: System validation involved analysis at different framework levels. From texture analysis of acquired images to application accuracy evaluation, the influence of various parameters on system performance is demonstrated. The simulation system is a valuable tool for learning minimally invasive procedures and for developers building AR systems for medical applications. The implementation is focused on the insertion surgical devices, including screws and K-wires. This is results in real application in minimizing the risk of iatrogenic injury to neural and vascular structures. To demonstrate the effectiveness of highly reproducible accuracy between real and virtual environment an analysis of errors and accuracies is illustrated at level of different subsystems. Measurement between comparative measurement between vernier caliper and simulation system methods shows a R>0.9 with a p<0.01. Application accuracy was evaluated using the following parameters. The relative point-to-point accuracy averaged 1.02mm with a standard deviation of 2.82mm.

Keywords: Augmented reality , Image guided surgery , Minimally invasive surgery , Orthopedic

The “Augmented Humans” term refers to the opportunity to improve human possibilities by using innovative technologies such as Artificial Intelligence (AI) and Extended Reality (XR). Digital therapies, particularly suitable for those treatments requiring multiple sessions, are increasingly being adopted for home-based treatment, enabling continuous monitoring and rehabilitation for patients, thus alleviating the burden on healthcare facilities by facilitating remote therapy sessions and follow-up visits. Among these, the Mirror Therapy (MT) for patients suffering from Phantom Limb Pain (PLP) could benefit greatly. This paper proposes a novel “Augmented Humans” framework for the treatment of PLP through home-based MT; the framework is designed to consider the activities carried on by the therapy center, the patient, and the system supporting the treatment. Moreover, an XR-based solution that integrates a Deep Learning (DL) approach has been developed to provide patients with a self-testing and self-assessment tool for conducting at-home rehabilitation sessions independently, even in the absence of physical medical staff. The DL algorithm enables real-time monitoring of rehabilitation exercises and automatic provision of personalized feedback on the gesture's performance, supporting the progressive improvement of the patient's movements and his ability to adhere to the treatment plan. The technical feasibility and usability of the proposed framework have been evaluated with 23 healthy subjects, highlighting an overall positive user experience. Remarkable results were obtained in terms of automatic gesture evaluation, with macro averaged accuracy and F1-score of 95%, paving the way for the adoption of the “Augmented Humans” approach in the healthcare domain.

Keywords: Augmented humans , Augmented therapy , Deep learning , Digital therapy , Extended reality , Home-based treatment , Phantom limb pain

In the integration of structured light systems (SLSs) into automotive manufacturing pipelines, achieving reliable 3D reconstruction under industrial conditions remains a critical challenge. Factors such as environmental variability, surface reflectivity, and optical configuration often compromise dimensional accuracy and point cloud quality, limiting the deployment of SLSs for inspection tasks. This paper presents a practical metric-based methodology for evaluating the dimensional accuracy and point cloud quality of SLSs targeted at automotive inspection applications. Unlike existing approaches focused primarily on theoretical or hardware-specific parameters, the proposed methodology considers all phases of the acquisition–reconstruction pipeline, including calibration, environmental variability, and image enhancement strategies, to practically guide engineers step-by-step in adapting SLSs to real-world operational constraints. The methodology was experimentally validated through a case study in an automotive production setting, where it enabled the detection of reconstruction biases caused by surface reflectivity and viewing angle. It also demonstrated the ability to quantify improvements obtained through image enhancement algorithms. These results confirm the methodology's capacity to expose critical performance trade-offs and guide optimization choices in practical inspection scenarios. By offering a repeatable and application-oriented evaluation framework, the methodology supports the robust integration of digital vision systems in industrial workflows, facilitating more informed decisions for system designers, process engineers, and quality control professionals.

Keywords: 3D reconstruction , accuracy evaluation , image enhancement , manufacturing inspection , point cloud analysis , structured light system

Digital Image Correlation (DIC) is a well-established technique that has recently gained interest in the field of vibration measurements. As vibration frequency increases, the displacement amplitude decreases, determining the need for extremely high subpixel measurement sensitivity. This work introduces a novel algorithm that exploits a spline-based approach to interpolate the integer-valued correlation map and enhance subpixel sensitivity. The approach allows for the calibration of the fitting procedure with respect to the local features of the speckle pattern, which is characterized by the reference image, improving the measurement's signal-to-noise ratio. The proposed procedure is compared with conventional approaches, which are based on polynomial fitting of the correlation map instead. Additionally, two different strategies are discussed to compute the integer-valued correlation map, i.e. pixel domain convolution and spatial-frequency domain convolution.

Keywords: Correlation map , Digital Image Correlation , Spline-based interpolation , Subpixel sensitivity , Vibration full-field measurements

Background: Despite its central role in pediatric pre-surgical evaluation of drug-resistant focal epilepsy, conventional analog 18F-fluorodeoxyglucose (18F-FDG) PET/CT (aPET) systems often yield modest epileptogenic zone (EZ) detection rates (~50–60%). Silicon photomultiplier–based digital PET/CT (dPET) promises enhanced image quality, but its performance in pediatric epilepsy remains untested. Methods: We retrospectively analyzed 22 children (mean age 11.5 ± 2.6 years) who underwent interictal brain 18F-FDG PET/CT: 11 on an analog system (Discovery ST, 2018–2019) and 11 on a digital system (Biograph Vision 450, 2020–2021). Three blinded nuclear medicine physicians independently scored EZ localization and image quality (4-point scale); post-surgical histology and ≥1-year clinical follow-up served as reference. Results: The EZ was correctly identified in 8/11 analog scans (72.7%) versus 10/11 digital scans (90.9%). Average image quality was significantly higher with dPET (3.0 ± 0.9 vs. 2.1 ± 0.9; p < 0.05), and inter-reader agreement improved from good (ICC = 0.63) to excellent (ICC = 0.91). Conclusions: Our preliminary findings suggest that dPET enhances image clarity and reader consistency, potentially improving localization accuracy in pediatric epilepsy presurgical workups.

Keywords: 18F-FDG , brain , epilepsy , neuroimaging , neurological disorders , PET/CT

Breast ultrasound imaging (BUS) plays a major role in the diagnosis of breast cancer. Consequently, computerised analysis of BUS images is being intensively investigated as a means to assist physicians in clinical decision making. The traditional approach involves the use of manually-designed (hand-crafted) morphology and texture features, although Deep Learning (DL) by convolutional neural networks (CNN) has received much interest in recent years. Advances in AI research has demonstrated the effectiveness of CNN models for classification tasks; however, this approach requires large datasets, extensive computational resources and solid knowledge of CNN architecture design.Deep feature extraction from pre-trained state-of-the art CNNs is an easy-to-implement approach that allows to exploit the information already learnt on large datasets (e.g., ImageNet) while saving substantial computational time and resources. In this work we compare the effectiveness of deep features from three pre-trained CNN with that of five types of hand-crafted features for differentiating benign vs. malignant lesions on BUS images. The analysis was carried out on 252 lesions (154 benign, 98 malignant) from the recently released BrEaST dataset. The effects of image preprocessing (resizing and denoising) and of the area from which the features are computed (region of interest - ROI or bounding box around the ROI) are also investigated. The results show that the deep features outperformed the handcrafted in terms of accuracy (best accuracy 81.4% vs. 78.6%) hence demonstrating the feasibility of the method.

Keywords: Breast cancer , Computer-assisted diagnosis , Deep learning , Hand-crafted features , Ultrasound images

Breast ultrasound images play a pivotal role in assessing the nature of suspicious breast lesions, particularly in patients with dense tissue. Computerized analysis of breast ultrasound images has the potential to assist the physician in the clinical decision-making and improve subjective interpretation. We assess the performance of conventional features, deep learning features and ensemble schemes for discriminating benign versus malignant breast lesions on ultrasound images. A total of 19 individual feature sets (1 morphological, 2 first-order, 10 texture-based, and 6 CNN-based) were included in the analysis. Furthermore, four combined feature sets (Best by class; Top 3, 5, and 7) and four fusion schemes (feature concatenation, majority voting, sum and product rule) were considered to generate ensemble models. The experiments were carried out on three independent open-access datasets respectively containing 252 (154 benign, 98 malignant), 232 (109 benign, 123 malignant), and 281 (187 benign, 94 malignant) lesions. CNN-based features outperformed the other individual descriptors achieving levels of accuracy between 77.4% and 83.6%, followed by morphological features (71.6%–80.8%) and histograms of oriented gradients (71.4%–77.6%). Ensemble models further improved the accuracy to 80.2% to 87.5%. Fusion schemes based on product and sum rule were generally superior to feature concatenation and majority voting.

Keywords: breast cancer , breast ultrasound images , convolutional neural networks , radiomics , texture

Purpose: To compare PET-derived metrics between digital and analogue PET/CT in hyperparathyroidism, and to assess whether machine learning (ML) applied to quantitative PET parameters can distinguish parathyroid adenoma (PA) from hyperplasia (PH). Methods: From an initial multi-centre cohort of 179 patients, 86 were included, comprising 89 PET-positive lesions confirmed histologically (74 PA, 15 PH). Quantitative PET parameters—maximum standardised uptake value (SUVmax), metabolic tumour volume (MTV), target-to-background ratio (TBR), and maximum diameter—along with serum PTH and calcium levels, were compared between digital and analogue PET scanners using the Mann–Whitney U test. Receiver operating characteristic (ROC) analysis identified optimal threshold values. ML models (LASSO, decision tree, Gaussian naïve Bayes) were trained on harmonised quantitative features to distinguish PA from PH. Results: Digital PET detected significantly smaller lesions than analogue PET, in both metabolic volume (1.32 ± 1.39 vs. 2.36 ± 2.01 cc; p < 0.001) and maximum diameter (8.35 ± 4.32 vs. 11.87 ± 5.29 mm; p < 0.001). PA lesions showed significantly higher SUVmax and TBR compared to PH (SUVmax: 8.58 ± 3.70 vs. 5.27 ± 2.34; TBR: 14.67 ± 6.99 vs. 8.82 ± 5.90; both p < 0.001). The optimal thresholds for identifying PA were SUVmax > 5.89 and TBR > 11.5. The best ML model (LASSO) achieved an AUC of 0.811, with 79.7% accuracy and balanced sensitivity and specificity. Conclusions: Digital PET outperforms analogue system in detecting small parathyroid lesions. Additionally, ML analysis of PET-derived metrics and PTH may support non-invasive distinction between adenoma and hyperplasia.

Keywords: 18F-choline , Artificial intelligence , Automated classification , Digital PET/CT , Hyperparathyroidism , Metabolic imaging

Intraosseous access is a critical emergency procedure for introducing fluids and blood components into the systemic circulation, necessitating thorough training for healthcare personnel. In collaboration with Meyer Children's University Hospital IRCCS, the Department of Industrial Engineering at the University of Florence has developed a high-fidelity, low-cost pediatric simulator specifically for intraosseous access in the proximal tibia. Using Materialise Mimics 26® software, anonymized tomographic images of a 5-year-old patient were processed to obtain the three-dimensional geometry of relevant anatomical structures: tibia, fibula, distal femur, muscles, and skin. These models were further refined with Geomagic DesignX® for the final simulator design. The simulator features silicone replicas for soft tissues, with rigid structures and silicone casting molds produced through Fused Deposition Modeling. Preliminary tests were conducted to select suitable materials, which were then evaluated by specialized medical personnel using the EZ-IO® device to ensure accurate haptic feedback. The resulting simulator provides a realistic platform for intraosseous access training, enabling healthcare practitioners to enhance their procedural skills and ensure patient safety.

Keywords: additive manufacturing , intraosseous access , pediatric mannequin

Unmanned aerial vehicles are increasingly used in civil Search and Rescue operations due to their rapid deployment and wide-area coverage capabilities. However, detecting missing persons from aerial imagery remains challenging due to small object sizes, cluttered backgrounds, and limited onboard computational resources, especially when managed by civil agencies. In this work, we present a comprehensive methodology for optimizing YOLO-based object detection models for real-time Search and Rescue scenarios. A two-stage transfer learning strategy was employed using VisDrone for general aerial object detection and Heridal for Search and Rescue-specific fine-tuning. We explored various architectural modifications, including enhanced feature fusion (FPN, BiFPN, PB-FPN), additional detection heads (P2), and modules such as CBAM, Transformers, and deconvolution, analyzing their impact on performance and computational efficiency. The best-performing configuration (YOLOv5s-PBfpn-Deconv) achieved a mAP@50 of 0.802 on the Heridal dataset while maintaining real-time inference on embedded hardware (Jetson Nano). Further tests at different flight altitudes and explainability analyses using EigenCAM confirmed the robustness and interpretability of the model in real-world conditions. The proposed solution offers a viable framework for deploying lightweight, interpretable AI systems for UAV-based Search and Rescue operations managed by civil protection authorities. Limitations and future directions include the integration of multimodal sensors and adaptation to broader environmental conditions.

Keywords: model optimization , real-time inference , search and rescue , small-object detection , unmanned aerial vehicles , YOLO

Cerebral aneurysm is a complex vascular pathology, and the assessment of its rupture risk is often based on qualitative criteria and clinical expertise. The present study aims to develop an objective parameter to support risk evaluation by integrating morphological and hemodynamic analyses. A novel morphological parameter has been introduced, combining key geometric features extracted from 3D vessel reconstructions of middle cerebral artery bifurcations. Computational fluid dynamics (CFD) simulations were conducted to analyze hemodynamic factors such as wall shear stress and intra-aneurysmal pressure, both of which are associated with aneurysm growth and rupture risk. Particular attention was given to defining physiologically accurate boundary conditions, essential for obtaining reliable results. The new morphological parameter alignment with clinical evaluations suggesting its potential to enhance the assessment of aneurysm operability and improve surgical decision-making.

Keywords: Cerebral aneurysm , CFD , Hemodynamics , Morphology

Assessing the pliability of burn scars represents a key clinical challenge for monitoring the healing process and optimising rehabilitation treatments. Currently, most of the methods used to assess scar quality are based on subjective evaluations, limiting the accuracy and reproducibility of the results. This study presents the development of an innovative scanner based on the Digital Image Correlation (DIC) technique to objectively measure changes in scarred skin deformation. The designed system integrates a hardware configuration with an image processing algorithm able to detect with high accuracy the deformations of the skin subjected to a known stress, allowing the calculation of the pliability value. By comparing two values obtained in two successive sessions, it is possible to verify the improvement in the health of the scar. Preliminary tests have shown that the system is able to provide reliable and reproducible data, confirming the technology's potential for scar quality assessment. The implementation of this technology could help change the management of burn scars, improving patients' quality of life through more accurate diagnoses and more effective treatments.

Keywords: 3D printing , Burn scars , CAD modelling , Digital image correlation , Pliability