Image Processing & Applications

Artificial Neural Network (ANN) has been used extensively and constantly developed. The combination of wavelet transform theory and the neural network has become an important branch to explore the optimization of neural network structure, and Wavelet Neural Network (WNN), a special network structure, was born. This paper reviews WNN's development and summarizes the system structure and algorithm implementation and presents derivative models and cutting-edge applications with obvious characteristics. The sorting and analysis of the above contents show that the combination of wavelet theory and neural network algorithm can make the network model have the advantages of fast convergence speed and high model accuracy, and has a rapid development trend in many fields such as audio signal and image processing. The work of this paper is intended to provide a reference for potential applications based on WNN and new network model design ideas.

Keywords: Wavelet Neural Network , Wavelet Transform

Background. Head and neck cancer (HNC) is the seventh most common neoplastic disorder at the global level. Contouring HNC lesions on [(Formula presented.) F] Fluorodeoxyglucose positron emission tomography/computed tomography (FDG PET/CT) scans plays a fundamental role for diagnosis, risk assessment, radiotherapy planning and post-treatment evaluation. However, manual contouring is a lengthy and tedious procedure which requires significant effort from the clinician. Methods. We evaluated the performance of six hand-crafted, training-free methods (four threshold-based, two algorithm-based) for the semi-automated delineation of HNC lesions on FDG PET/CT. This study was carried out on a single-centre population of (Formula presented.) subjects, and the standard of reference was manual segmentation generated by nuclear medicine specialists. Figures of merit were the Sørensen–Dice coefficient (DSC) and relative volume difference (RVD). Results. Median DSC ranged between 0.595 and 0.792, median (Formula presented.) between −22.0% and 87.4%. Click and draw and Nestle's methods achieved the best segmentation accuracy (median DSC, respectively, 0.792 ± 0.178 and 0.762 ± 0.107; median RVD, respectively, −21.6% ± 1270.8% and −32.7% ± 40.0%) and outperformed the other methods by a significant margin. Nestle's method also resulted in a lower dispersion of the data, hence showing stronger inter-patient stability. The accuracy of the two best methods was in agreement with the most recent state-of-the art results. Conclusions. Semi-automated PET delineation methods show potential to assist clinicians in the segmentation of HNC lesions on FDG PET/CT images, although manual refinement may sometimes be needed to obtain clinically acceptable ROIs.

Keywords: head and neck cancer , positron emission tomography , radiomics , region of interest , segmentation

The rapid development of production methods and the introduction of additive manufacturing (AM) technologies have increased the design freedom and availability of unlabeled data. Typically, manufacturing process selection requires a large amount of labeled data which is expensive and difficult to reach. This is more critical with the continuous improvement of AM systems, which can be increasingly used to substitute traditional manufacturing technologies. Implementing unsupervised learning into manufacturing process selection and classification is beneficial since unsupervised learning uses unlabeled data. Hence, this paper aims to differentiate parts to be fabricated favorably by means of additive or traditional manufacturing technologies using image processing and unsupervised learning. The input image dataset is constructed from freely accessible web databases in which forty CAD (computer-aided design) models are available. The corresponding images of CAD models are extracted using the SOLIDWORKS 2022 software, where 42.50% are AM-ed, and the remaining are chosen from traditionally manufactured parts. The hierarchical clustering algorithm reported 87.50% accuracy, showing promising potential for manufacturing process classification and image processing applications.

Keywords: hierarchical clustering , image processing , machine learning , manufacturing process selection , unsupervised learning

Modern semiconductor manufacturing production lines include several technological processes such as diffusion, wafer cleaning and epitaxy and therefore they are subject to degradation phenomena that must be carefully monitored. The processed semiconductor wafer consequently requires a very accurate testing phase which allows to detect the presence of defects to be traced back to the production line and which can have an impact on the downstream yield. The electric wafer testing phase (EWS i.e. Electrical Wafer Sorting) is certainly one of the key stages in semiconductor wafer monitoring in both classical silicon-based technology and modern silicon carbide production lines. The EWS validation phase includes the execution of several electrical tests directly on the integrated circuits (dies) over the wafer. The EWS electrical assessment outcome can be reported in a 2D binary image-map enabling the formation of specific visual patterns in accordance with the defect statistics found in the testing phase. Specifically, from this intelligent EWS analysis it is possible to early-assess defects in the production lines, electrical defects in the devices and finally a proper wafer-yield forecasting estimation. The classification of the EWS image-map visual patterns is a very challenging and time-consuming task. In that context, ad-hoc deep learning solution has been developed in order to perform supervised/unsupervised EWS image-maps patterns recognition and tracking. The related pipeline is composed by ad-hoc convolutional deep architecture with an innovative augmentation system based on spatiotemporal preprocessing of input EWS image-maps. The implemented architecture is able to classify the input EWS binary maps into a well-known pattern-classes whose features at the production line level and the impact on yield are known. The designed deep network is able to identify new EWS patterns not included in the already known classes, so as to address the production-engineer accordingly. The proposed pipeline shows very promising results as it is able to classify the tested EWS with an average accuracy close to 97%.

Keywords: Clustering , CNN , EWS

Background: Addressing intraoperative bleeding remains a significant challenge in the field of robotic surgery. This research endeavors to pioneer a groundbreaking solution utilizing convolutional neural networks (CNNs). The objective is to establish a system capable of forecasting instances of intraoperative bleeding during robot-assisted radical prostatectomy (RARP) and promptly notify the surgeon about bleeding risks. Methods: To achieve this, a multi-task learning (MTL) CNN was introduced, leveraging a modified version of the U-Net architecture. The aim was to categorize video input as either “absence of blood accumulation” (0) or “presence of blood accumulation” (1). To facilitate seamless interaction with the neural networks, the Bleeding Artificial Intelligence-based Detector (BLAIR) software was created using the Python Keras API and built upon the PyQT framework. A subsequent clinical assessment of BLAIR's efficacy was performed, comparing its bleeding identification performance against that of a urologist. Various perioperative variables were also gathered. For optimal MTL-CNN training parameterization, a multi-task loss function was adopted to enhance the accuracy of event detection by taking advantage of surgical tools' semantic segmentation. Additionally, the Multiple Correspondence Analysis (MCA) approach was employed to assess software performance. Results: The MTL-CNN demonstrated a remarkable event recognition accuracy of 90.63%. When evaluating BLAIR's predictive ability and its capacity to pre-warn surgeons of potential bleeding incidents, the density plot highlighted a striking similarity between BLAIR and human assessments. In fact, BLAIR exhibited a faster response. Notably, the MCA analysis revealed no discernible distinction between the software and human performance in accurately identifying instances of bleeding. Conclusion: The BLAIR software proved its competence by achieving over 90% accuracy in predicting bleeding events during RARP. This accomplishment underscores the potential of AI to assist surgeons during interventions. This study exemplifies the positive impact AI applications can have on surgical procedures.

Keywords: artificial intelligence , complications , prostate cancer , robotics

This paper introduces an assistive system for diagnosing malignant pleural mesothelioma (MPM), a highly aggressive cancer caused by asbestos exposure. The system assists healthcare providers in accurately determining the tumor's volume and the appropriate dose of chemotherapy to administer. It employs a bi-level process that uses machine learning and deep learning techniques to classify CT images of lungs and then calculate the tumor's volume. The study addresses the challenges associated with deep nets, such as the need for large and diverse datasets, hyperparameter optimization, and potential data bias. Two CNN architectures, Inception-v3 and ResNet-50, were compared in terms of their features and performance, and three hyperparameters were optimized for each model to generate a broad range of training scenarios. To estimate the amount of cancer cells to target, CT images were used to calculate the tumor's volume. This process involved image pre-processing, cropping volume, registering and filtering images, filling holes, segmentation, and 3D reconstruction. The results demonstrate that the developed system utilizing CNN optimizations and reconstruction of 3D images from CT images can benefit the treatment of MPM patients. The system has the potential to improve the accuracy of MPM diagnosis and the effectiveness of chemotherapy, ultimately improving patient outcomes.

Keywords: 3D volume reconstruction , CNN , CT imaging , Malignant Pleural Mesothelioma

Segmentation of anatomical components is a critical step in creating accurate and realistic 3D models of the human body, which are employed in a wide range of clinical applications, particularly in orthopedics. Recently, many deep learning approaches have been proposed to solve the problem of manual segmentation. Among the available software for automatic segmentation, MONAI Label is a free open-source tool, which allows for the creation of annotated datasets and the development of AI-based annotation models for clinical assessment. In this context, the present study is designed to compare the performance of two well-known neural networks in segmenting knee bones. In spite of the fact that several studies have investigated the use of deep learning techniques for knee reconstruction, there is no consensus regarding the most effective method. In the present study, validation metrics are selected in order to assess the accuracy of the automated segmentation models in comparison with the ground truth data. Magnetic resonance images of 31 patients have been employed for the study. As result, U-Net shows better performance than the SegResNet in the automatic femur segmentation task.

Keywords: 3D Modelling Knee , Automated Segmentation , Deep Learning , SegResNet , U-Net

Autonomous vehicles companies are awakening of cutting-edge technology by offering self-driving cars, however, customers' acceptance and trust of this high-tech product has become a significant open challenge in the world. It can be expected that autonomous vehicles (AVs) ameliorate traffic flow, lessen accidents and injuries, and save the time on travel in addition to people's life. Despite the increment of human replacement by the artificial intelligence in various industries and artifacts, most customers have not convinced with this big revolutionize transportation yet. This paper presents an experiment that evaluated the impact of personality, affinity to technology and demographics on human's trust in autonomous vehicles. 19 engineering students did participate in the experiment by using an autonomous vehicle with level 2, that had automated speed and trajectory maintenance, but without automatic detection for obstacles, objects or events while driving. First, the AV programmer explained its function and information about the track to all participants. Then, they were asked to answer to 9 questions of Affinity for Technology Interaction (ATI) scale, 17 demographic questions, and 10 questions of Big Five Questionnaire (BFQ). Afterwards, all participants rode 10-minute path in a pre-defined rout with the accompaniment of the AV programmer to control probable dangerous situations. Finally, they completed Trust in Automation Scale questionnaire. Gathering data was analysed using IBM SPSS 26. The mean and standard deviation were calculated for descriptive data and the Spearman for analytical data. The results showed a behavioural split (50/50) among people having and not having complete confidence in the autonomous vehicle. Nervous people trust the AV less than others, like people with little or no experience in using an AV. As well, the more transparency we had in the AV's behaviours and presenting information about it, the more trust and security was perceived by participants. People with high level of technology affinity experienced more trust in the AV, as the effect of frequent exposures with an AV. It's worth mentioning that females are more conservative and prefer to trust an AV less than males. The results indicated that various variables could affect people's trust in AVs. Obviously, changing people's demographics and some aspects of personality seem to be impossible most of the times, while as approved by other recent studies, their affinity to technology, as a new type of personality, can change or be changed. As the results showed, it is recommended to AVs companies to design a good system in the vehicle, in order to present all information to passengers clearly. In addition, if some conditions are provided that people can test these vehicles and know more about their functions in detail, their trust will definitely increase significantly.

Keywords: Autonomous vehicle , Personality , Technology affinity , Trust , User experience

Humans drive in a holistic fashion which entails, in particular, understanding dynamic road events and their evolution. Injecting these capabilities in autonomous vehicles can thus take situational awareness and decision making closer to human-level performance. To this purpose, we introduce the ROad event Awareness Dataset (ROAD) for Autonomous Driving, to our knowledge the first of its kind. ROAD is designed to test an autonomous vehicles ability to detect road events, defined as triplets composed by an active agent, the action(s) it performs and the corresponding scene locations. ROAD comprises videos originally from the Oxford RobotCar Dataset annotated with bounding boxes showing the location in the image plane of each road event. We benchmark various detection tasks, proposing as a baseline a new incremental algorithm for online road event awareness termed 3D-RetinaNet. We also report the performance on the ROAD tasks of Slowfast and YOLOv5 detectors, as well as that of the winners of the ICCV2021 ROAD challenge, which highlight the challenges faced by situation awareness in autonomous driving. ROAD is designed to allow scholars to investigate exciting tasks such as complex (road) activity detection, future event anticipation and continual learning. The dataset is available at https://github.com/gurkirt/road-dataset; the baseline can be found at https://github.com/gurkirt/3D-RetinaNet.

Keywords: action detection , Autonomous driving , Autonomous vehicles , Benchmark testing , decision making , Decision making , road agents , Roads , situation awareness , Task analysis , Vehicle dynamics , Videos

Background and Objective:In biomedical fields, image analysis is often necessary for an accurate diagnosis. In order to obtain all the information needed to form an in-depth clinical picture, it may be useful to combine the contents of images taken under different diagnostic modes. Multimodal medical image fusion techniques enable complementary information acquired by different imaging devices to be automatically combined into a unique image. Methods:In this paper, multimodal medical images fusion method based on multiresolution analysis (MRA) is proposed, with the aim to combine the high geometric content of magnetic resonance imaging (MRI) and the elasticity information of magnetic resonance elastography (MRE), simultaneously acquired on the same organs of a patient. First, the slices of MRE are volumetrically interpolated to exactly overlap, each with a slice of MRI. Then, the spatial details of MRI are extracted by means of MRA and injected into the corresponding slices of MRE. Due to the intrinsic dissimilarity between corresponding slices of MRE and MRI, the spatial details of MRI are modulated by local or global matching functions. Results:The performance of the proposed method is quantitatively assessed considering radiometric and geometric consistency of the fused images with respect to their originals, in a comparison with two popular methods from the literature. For a qualitative evaluation, a visual inspection is carried out. Conclusions:The results show that the proposed method enables an effective MRI-MRE fusion that allows the elasticity information and geometric details of the examined organs to be evaluated in a single image.

Keywords: Biomedical images , Image processing , Magnetic resonance elastography , Multimodal fusion , Multiresolution analysis , Tomographic sequences

In this paper an interactive computational methodology was developed assuming that shape and size optimization of flexible components can significantly improve energy absorption or storage ability in assembled systems with flexible components (AS-FC). A radial basis functions mesh morphing formulation in non-linear numerical finite element analysis, including contact problems and flow interaction, was adopted as optimal design method to optimize shape and size design parameters in AS-FC. Flexible components were assembled in finite element environment according to functional ISO-ASME tolerances specification; non-linear structural analysis with flow interaction analysis was performed. The results of the study showed that the proposed method allows to optimize the shape and size of the flexible components in AS-FC maximizing the system's ability to absorb or store energy. The potentiality of the method and its forecasting capability were discussed for the case study of an automotive crash shock in which the specific energy absorption was increased by over 40%. The case studied refers to a simple flexible component geometry, but the method could be extended to systems with more complex geometries.

Keywords: Crash shock absorber , ISO-ASME tolerances specification , Radial basis functions , Shape and size optimization , Specific energy absorption

In the Cultural Heritage field, the choice of materials and exhibit structures is essential to properly house and support artifacts without causing damage or deterioration. This problem is even more evident in the case of finds made of stone for which, due to their weight, a proper selection and dimensioning of the relative supports is required. In fact, without adequate support, this can result in stress concentrations that could compromise the artifact's state of conservation. As a consequence, more often such exhibition supports are customized items, that are designed and manufactured to meet specific functional and artistic setup needs. In this context, the paper presents a design approach that combines topology optimization and additive manufacturing techniques to develop customized support structures which undertake the twofold purpose of preserving the artifact and making it available for the exhibition in the museum. The proposed approach has been assessed through the case study of a sandstone Ionic capital hosted in the Brettii & Enotri Museum in Cosenza (Italy). The proposed approach is therefore meant as a guideline for the design of customized exhibit supports especially in the case of sandstone artifacts with a complex shape or a conservation condition that requires specific attention.

Keywords: Additive manufacturing , Cultural heritage , Design methods , Exhibit supports , Photogrammetry , Topology optimization

Although computer vision techniques are widely and effectively used for terrestrial applications, the adoption of these techniques in underwater scenarios is often a challenging task. This is due to the water medium that degrades the quality of the acquired images requiring the use of specific methodologies for their correction and improvement. In this context, this paper presents a low-cost system for underwater 3D reconstruction and 3D measurements based on an off-the-shelf depth camera. The system consists of a Stereolabs ZED stereo camera and its waterproof housing designed to be fabricated through additive manufacturing technologies. A dedicated software has been developed to manage all the steps of the acquisition process, starting from the calibration phase and continuing for rectification, depth map computation, 3D scene reconstruction, and 3D measurement. A preliminary laboratory test has been carried out in a controlled environment to quantify system performance in accuracy of measurement.

Keywords: Calibration Techniques , Stereo Camera , Stereolabs ZED , Underwater 3D Reconstruction , Underwater Inspection

Indeterminate lung nodules detected on CT scans are common findings in clinical practice. Their correct assessment is critical, as early diagnosis of malignancy is crucial to maximise the treatment outcome. In this work, we evaluated the role of form factors as imaging biomarkers to differentiate benign vs. malignant lung lesions on CT scans. We tested a total of three conventional imaging features, six form factors, and two shape features for significant differences between benign and malignant lung lesions on CT scans. The study population consisted of 192 lung nodules from two independent datasets, containing 109 (38 benign, 71 malignant) and 83 (42 benign, 41 malignant) lung lesions, respectively. The standard of reference was either histological evaluation or stability on radiological followup. The statistical significance was determined via the Mann–Whitney U nonparametric test, and the ability of the form factors to discriminate a benign vs. a malignant lesion was assessed through multivariate prediction models based on Support Vector Machines. The univariate analysis returned four form factors (Angelidakis compactness and flatness, Kong flatness, and maximum projection sphericity) that were significantly different between the benign and malignant group in both datasets. In particular, we found that the benign lesions were on average flatter than the malignant ones; conversely, the malignant ones were on average more compact (isotropic) than the benign ones. The multivariate prediction models showed that adding form factors to conventional imaging features improved the prediction accuracy by up to 14.5 pp. We conclude that form factors evaluated on lung nodules on CT scans can improve the differential diagnosis between benign and malignant lesions.

Keywords: computed tomography , form factors , lung cancer , radiomics

Novel therapeutic approaches, consisting of immune check-point inhibitors (ICIs) and molecularly targeted therapy, have thoroughly changed the clinical management of malignant melanoma (MM), the most frequent and deadly skin cancer. Since only 30–40% of MM patients respond to ICIs, imaging biomarkers suitable for the pre-therapeutic stratification and response assessment are warmly welcome. In this scenario, positron emission computed tomography (PET/CT) with18 F-fluorodeoxyglucose (18 F-FDG) has been successfully utilized for advanced MM staging and therapy response evaluation. Furthermore, several PET-derived parameters (SUVmax, MTV, TLG) were particularly impactful for the prognostic evaluation of patients submitted to targeted and immunotherapy. In this review, we performed a web-based and desktop research on the clinical applications of18 F-FDG PET/CT in MM, with a particular emphasis on the various metabolic criteria developed for interpreting PET/CT scan in patients undergoing immunotherapy or targeted therapy or a combination of both. Furthermore, the emerging role of radiomics, a quantitative approach to medical imaging applying analysis methodology derived by the field of artificial intelligence, was examined in the peculiar context, putting a particular emphasis on the potential of this discipline to support clinicians in the delicate process of building patient-tailored pathways of care.

Keywords: 18 F-FDG , artificial intelligence , BRAF mutation , malignant melanoma , PET/CT , precision medicine , radiomics

Hepatic diseases are serious condition worldwide, and several times doctors analyse the situation and elaborates a preoperative planning based exclusively on the medical images, which are a drawback since they only provide a 2D vision and the location of the damaged tissues in the three-dimensional space cannot be easily determined by surgeons. Nowadays, with the advancement of Computer Aided Design (CAD) technologies and image segmentation, a digital liver model can be obtained to help understand the particular medical case; even with the geometric model, a virtual simulation can be elaborated. This work is divided into two phases; the first phase involves a workflow to create a liver geometrical model from medical images. Whereas the second phase provides a methodology to achieve liver prototype, using the technique of fused deposition modelling (FDM). The two stages determine and evaluate the most influencing parameters to make this design repeatable in different hepatic diseases. The reported case study provides a valuable method for optimizing preoperative plans for liver disease. In addition, the prototype built with additive manufacturing will allow the new doctors to speed up their learning curve, since they can manipulate the real geometry of the patient's liver with their hands.

Keywords: 3D printing , Convolutional neural network , FDM , Image segmentation , Liver disease

There is a large body of research devoted to identifying the complexity of structures in networks. In the context of network theory, a complex network is a graph with nontrivial topological features—features that do not occur in simple networks, such as lattices or random graphs, but often occur in graphs modeling real systems. The study of complex networks is a young and active area of scientific research inspired largely by the empirical study of real-world networks, such as computer networks and logistic transport networks. Transport is of great importance for the economic and cultural cooperation of any country with other countries, the strengthening and development of the economic management system, and in solving social and economic problems. Provision of the territory with a well-developed transport system is one of the factors for attracting population and production, serving as an important advantage for locating productive forces and providing an integration effect. In this paper, we introduce a new method for quantifying the complexity of a network based on presenting the nodes of the network in Cartesian coordinates, converting to polar coordinates, and calculating the fractal dimension using the ReScaled ranged (R/S) method. Our results suggest that this approach can be used to determine complexity for any type of network that has fixed nodes, and it presents an application of this method in the public transport system.

Keywords: complexity , fractal , Hurst exponent H , network , public transport

The compression behaviour in quasi-static and dynamic conditions of cellular materials is crucial for their applications both for ensuring structural strength and high energy absorption capability. Despite the recent progress made in understanding the experimental observations, analytical and numerical modelling still requires improvements in the Representative Volume Element (RVE) identification that can be uncertain due to the limited dimensions of the investigated specimens and to the cell inhomogeneity. The objective of this paper is to implement a material model able to consider the statistical distribution so that its effect can be quantitatively highlighted, mitigating uncertainty of the RVE identification. The applied methodology started with morphological and topological analyses on samples extracted from an ingot of AA7075T6 foam, which was manufactured by compact powder technology. Quasi-static and dynamic experimental compression tests have been carried out and compared with 3D mesoscale numerical simulations in order to correlate the mechanical behaviour of the foam to the cell characteristics. Finally, an equivalent material model, which is a function of the statistical distributions of cells morphology and topology, has been proposed and analytically verified.

Keywords: Aluminium foam , Cellular material , High strain rate , Homogeneous formulation , RVE , SHPB

In hip arthroplasty, preoperative planning is fundamental to reaching a successful surgery. Nowadays, several software tools for computed tomography (CT) image processing are available. However, research studies comparing segmentation tools for hip surgery planning for patients affected by osteoarthritic diseases or osteoporotic fractures are still lacking. The present work compares three different software from the geometric, dimensional, and usability perspectives to identify the best three-dimensional (3D) modelling tool for the reconstruction of pathological femoral heads. Syngo.via Frontier (by Siemens Healthcare) is a medical image reading and post-processing software that allows low-skilled operators to produce prototypes. Materialise (by Mimics) is a commercial medical modelling software. 3D Slicer (by slicer.org) is an open-source development platform used in medical and biomedical fields. The 3D models reconstructed starting from the in vivo CT images of the pathological femoral head are compared with the geometries obtained from the laser scan of the in vitro bony specimens. The results show that Mimics and 3D Slicer are better for dimensional and geometric accuracy in the 3D reconstruction, while syngo.via Frontier is the easiest to use in the hospital setting.

Keywords: bio-imaging , CT image segmentation , hip surgery , orthopaedics , reverse engineering , software comparison , surgical planning