Shedding light on ai in radiology: A systematic review and taxonomy of eye gaze-driven interpretability in deep learning.

X-ray imaging plays a crucial role in diagnostic medicine. Yet, a significant portion of the global population lacks access to this essential technology due to a shortage of trained radiologists. Eye-tracking data and deep learning models can enhance X-ray analysis by mapping expert focus areas, guiding automated anomaly detection, optimizing workflow efficiency, and bolstering training methods for novice radiologists. However, the literature shows contradictory results regarding the usefulness of eye-tracking data in deep-learning architectures for abnormality detection. We argue that these discrepancies between studies in the literature are due to (a) the way eye-tracking data is (or is not) processed, (b) the types of deep learning architectures chosen, and (c) the type of application that these architectures will have. We conducted a systematic literature review using PRISMA to address these contradicting results. We analyzed 60 studies that incorporated eye-tracking data in a deep-learning approach for different application goals in radiology. We performed a comparative analysis to understand if eye gaze data contains feature maps that can be useful under a deep learning approach and whether they can promote more interpretable predictions. To the best of our knowledge, this is the first survey in the area that performs a thorough investigation of eye gaze data processing techniques and their impacts in different deep learning architectures for applications such as error detection, classification, object detection, expertise level analysis, fatigue estimation and human attention prediction in medical imaging data. Our analysis resulted in two main contributions: (1) taxonomy that first divides the literature by task, enabling us to analyze the value eye movement can bring for each case and build guidelines regarding architectures and gaze processing techniques adequate for each application, and (2) an overall analysis of how eye gaze data can promote explainability in radiology.

BmooMPα-I, a Metalloproteinase Isolated from Bothrops moojeni Venom, Reduces Blood Pressure, Reverses Left Ventricular Remodeling and Improves Cardiac Electrical Conduction in Rats with Renovascular Hypertension.

BmooMPα-I has kininogenase activity, cleaving kininogen releasing bradykinin and can hydrolyze angiotensin I at post-proline and aspartic acid positions, generating an inactive peptide. We evaluated the antihypertensive activity of BmooMPα-I in a model of two-kidney, one-clip (2K1C). Wistar rats were divided into groups: Sham, who underwent sham surgery, and 2K1C, who suffered stenosis of the right renal artery. In the second week of hypertension, we started treatment (Vehicle, BmooMPα-I and Losartan) for two weeks. We performed an electrocardiogram and blood and heart collection in the fourth week of hypertension. The 2K1C BmooMPα-I showed a reduction in blood pressure (systolic pressure: 131 ± 2 mmHg; diastolic pressure: 84 ± 2 mmHg versus 174 ± 3 mmHg; 97 ± 4 mmHg, 2K1C Vehicle, p < 0.05), improvement in electrocardiographic parameters (Heart Rate: 297 ± 4 bpm; QRS: 42 ± 0.1 ms; QT: 92 ± 1 ms versus 332 ± 6 bpm; 48 ± 0.2 ms; 122 ± 1 ms, 2K1C Vehicle, p < 0.05), without changing the hematological profile (platelets: 758 ± 67; leukocytes: 3980 ± 326 versus 758 ± 75; 4400 ± 800, 2K1C Vehicle, p > 0.05), with reversal of hypertrophy (left ventricular area: 12.1 ± 0.3; left ventricle wall thickness: 2.5 ± 0.2; septum wall thickness: 2.3 ± 0.06 versus 10.5 ± 0.3; 2.7 ± 0.2; 2.5 ± 0.04, 2K1C Vehicle, p < 0.05) and fibrosis (3.9 ± 0.2 versus 7.4 ± 0.7, 2K1C Vehicle, p < 0.05). We concluded that BmooMPα-I improved blood pressure levels and cardiac remodeling, having a cardioprotective effect.

Clinical and Laboratory Profile of People Living with HIV/AIDS with Oral Kaposi Sarcoma.

The aim of this article was to evaluate the clinical and laboratory profile of people with oral Kaposi's sarcoma (KS) associated with AIDS (KS-AIDS), followed-up at a public university hospital in Salvador, Bahia, Brazil, in the past 10 years. We identified patients diagnosed with KS-AIDS, presenting oral manifestation from January 2007 to December 2017. We searched, in the hospital information systems, the patient demographics, diagnostic data, treatment, image studies, and oral photographic records. Of the 39 cases of KS-AIDS identified at the institution, 14 (22.8%) presented oral lesions. There was a predominance of black men, with a mean age of 32.5 years. Most cases (85.1%) manifested signs of KS simultaneously with the diagnosis of HIV infection, with extremely low initial CD4 T cell counts (average of 52.6 cells/mm2) and visceral involvement (64.3%). The palate (32.1%) and gingiva (21.4%) were the most affected oral sites. Histologically, the tumors exhibited proliferation of spindle cells between vascular clefts and extravasated erythrocytes. Oral KS-AIDS was frequent in young black adult men, with severe immunosuppression and high viral load counting, mostly with lesions manifested in the same period of diagnosis of infection by the HIV.

Augmented situated visualization methods towards electromagnetic compatibility testing.

In electrical engineering, hardware experts often need to analyze electromagnetic radiation data to detect any external interference or anomaly. The field that studies this sort of assessment is called electromagnetic compatibility (EMC). As a way to support EMC analysis, we propose the use of Augmented Situated Visualization (ASV) to supply professionals with visual and interactive information that helps them to comprehend that data, however situating it where it is most relevant in its spatial context. Users are able to interact with the visualization by changing the attributes being displayed, comparing the overlaps of multiple fields, and extracting data, as a way to refine their search. The solutions being proposed in this work were tested against each other in comparable 2D and 3D interactive visualizations of the same data in a series of data-extraction assessments with users, as a means to validate the approaches. Results exposed a correctness-time trade-off between the interaction methods. The hand-based techniques (Hand Slider and Touch Lens) were the least error-prone, being near to half as error-inducing as the gaze-based method. Touch Lens also performed as the least time-consuming method, taking in average less than half of the average time required by the others. For the visualization methods tested, the 2D ray casts presented a higher usability score and lesser workload index than the 3D topology view, however exposing over two times the error ratio. Ultimately, this work exposes how AR can help users to have better performances in a decision-making context, particularly in EMC related tasks, while also furthering the research in the ASV field.

Altering the Stiffness, Friction, and Shape Perception of Tangible Objects in Virtual Reality Using Wearable Haptics.

Tangible objects are used in virtual reality (VR) and augmented reality (AR) to enhance haptic information on the general shape of virtual objects. However, they are often passive or unable to simulate rich varying mechanical properties. This article studies the effect of combining simple passive tangible objects and wearable haptics for improving the display of varying stiffness, friction, and shape sensations in these environments. By providing timely cutaneous stimuli through a wearable finger device, we can make an object feel softer or more slippery than it really is, and we can also create the illusion of encountering virtual bumps and holes. We evaluate the proposed approach carrying out three experiments with human subjects. Results confirm that we can increase the compliance of a tangible object by varying the pressure applied through a wearable device. We are also able to simulate the presence of bumps and holes by providing timely pressure and skin stretch sensations. Altering the friction of a tangible surface showed recognition rates above the chance level, albeit lower than those registered in the other experiments. Finally, we show the potential of our techniques in an immersive medical palpation use case in VR. These results pave the way for novel and promising haptic interactions in VR, better exploiting the multiple ways of providing simple, unobtrusive, and inexpensive haptic displays.

Diagnostic concordance between mobile interfaces and conventional workstations for emergency imaging assessment.

INTRODUCTION: Mobile devices and software are now available with sufficient computing power, speed and complexity to allow for real-time interpretation of radiology exams. In this paper, we perform a multivariable user study that investigates concordance of image-based diagnoses provided using mobile devices on the one hand and conventional workstations on the other hand. METHODS: We performed a between-subjects task-analysis using CT, MRI and radiography datasets. Moreover, we investigated the adequacy of the screen size, image quality, usability and the availability of the tools necessary for the analysis. Radiologists, members of several teams, participated in the experiment under real work conditions. A total of 64 studies with 93 main diagnoses were analyzed. RESULTS: Our results showed that 56 cases were classified with complete concordance (87.69%), 5 cases with almost complete concordance (7.69%) and 1 case (1.56%) with partial concordance. Only 2 studies presented discordance between the reports (3.07%). The main reason to explain the cause of those disagreements was the lack of multiplanar reconstruction tool in the mobile viewer. Screen size and image quality had no direct impact on the mobile diagnosis process. CONCLUSION: We concluded that for images from emergency modalities, a mobile interface provides accurate interpretation and swift response, which could benefit patients' healthcare.

Efficient Surgical Cutting with Position-Based Dynamics.

Simulations of cuts on deformable bodies have been an active research subject for more than two decades. However, previous works based on finite element methods and mass spring meshes cannot scale to complex surgical scenarios. This article presents a novel method that uses position-based dynamics (PBD) for mesh-free cutting simulation. The proposed solutions include a method to efficiently render force feedback while cutting, an efficient heat diffusion model to simulate electrocautery, and a novel adaptive skinning scheme based on oriented particles.https://extras.computer.org/extra/mcg2017030024s1.mp4.

Designing a Vibrotactile Head-Mounted Display for Spatial Awareness in 3D Spaces.

Due to the perceptual characteristics of the head, vibrotactile Head-mounted Displays are built with low actuator density. Therefore, vibrotactile guidance is mostly assessed by pointing towards objects in the azimuthal plane. When it comes to multisensory interaction in 3D environments, it is also important to convey information about objects in the elevation plane. In this paper, we design and assess a haptic guidance technique for 3D environments. First, we explore the modulation of vibration frequency to indicate the position of objects in the elevation plane. Then, we assessed a vibrotactile HMD made to render the position of objects in a 3D space around the subject by varying both stimulus loci and vibration frequency. Results have shown that frequencies modulated with a quadratic growth function allowed a more accurate, precise, and faster target localization in an active head pointing task. The technique presented high usability and a strong learning effect for a haptic search across different scenarios in an immersive VR setup.

A Case-Based Study with Radiologists Performing Diagnosis Tasks in Virtual Reality.

In radiology diagnosis, medical images are most often visualized slice by slice. At the same time, the visualization based on 3D volumetric rendering of the data is considered useful and has increased its field of application. In this work, we present a case-based study with 16 medical specialists to assess the diagnostic effectiveness of a Virtual Reality interface in fracture identification over 3D volumetric reconstructions. We developed a VR volume viewer compatible with both the Oculus Rift and handheld-based head mounted displays (HMDs). We then performed user experiments to validate the approach in a diagnosis environment. In addition, we assessed the subjects' perception of the 3D reconstruction quality, ease of interaction and ergonomics, and also the users opinion on how VR applications can be useful in healthcare. Among other results, we have found a high level of effectiveness of the VR interface in identifying superficial fractures on head CTs.

Efficient medical image access in diagnostic environments with limited resources

Abstract Introduction A medical application running outside the workstation environment has to deal with several constraints, such as reduced available memory and low network bandwidth. The aim of this paper is to present an approach to optimize the data flow for fast image transfer and visualization on mobile devices and remote stationary devices. Methods We use a combination of client- and server-side procedures to reduce the amount of information transferred by the application. Our approach was implemented on top of a commercial PACS and evaluated through user experiments with specialists in typical diagnosis tasks. The quality of the system outcome was measured in relation to the accumulated amount of network data transference and the amount of memory used in the host device. Besides, the system's quality of use (usability) was measured through participants’ feedback. Results Contrarily to previous approaches, ours keeps the application within the memory constraints, minimizing data transferring whenever possible, allowing the application to run on a variety of devices. Moreover, it does that without sacrificing the user experience. Experimental data point that over 90% of the users did not notice any delays or degraded image quality, and when they did, they did not impact on the clinical decisions. Conclusion The combined activities and orchestration of our methods allow the image viewer to run on resource-constrained environments, such as those with low network bandwidth or little available memory. These results demonstrate the ability to explore the use of mobile devices as a support tool in the medical workflow.

Using Immersive Virtual Reality to Reduce Work Accidents in Developing Countries.

Thousands of people die or are injured in work accidents every year. Although the lack of safety equipment is one of the causes, especially in developing countries, behavioral issues caused by psychosocial factors are also to blame. This article introduces the use of immersive VR simulators to preventively reduce accidents in the workplace by detecting behavioral patterns that may lead to an increased predisposition to risk exposure. The system simulates day-to-day situations, analyzes user reactions, and classifies the behaviors according to four psychosocial groups. The results of a user study support the effectiveness of this approach.

Anatomic hepatectomy planning through mobile display visualization and interaction.

Hepatectomies are resections in which segments of the liver are extracted. While medical images are fundamental in the surgery planning procedure, the process of analysis of such images slice-by-slice is still tedious and inefficient. In this work we propose a strategy to efficiently and semi-automatically segment and classify patient-specific liver models in 3D through a mobile display device. The method is based on volume visualization of standard CT datasets and allows accurate estimation of functional remaining liver volume. Experiments showing effectiveness of the method are presented, and quantitative and qualitative results are discussed.

Surgical model-view-controller simulation software framework for local and collaborative applications.

PURPOSE: Surgical simulations require haptic interactions and collaboration in a shared virtual environment. A software framework for decoupled surgical simulation based on a multi-controller and multi-viewer model-view-controller (MVC) pattern was developed and tested. METHODS: A software framework for multimodal virtual environments was designed, supporting both visual interactions and haptic feedback while providing developers with an integration tool for heterogeneous architectures maintaining high performance, simplicity of implementation, and straightforward extension. The framework uses decoupled simulation with updates of over 1,000 Hz for haptics and accommodates networked simulation with delays of over 1,000 ms without performance penalty. RESULTS: The simulation software framework was implemented and was used to support the design of virtual reality-based surgery simulation systems. The framework supports the high level of complexity of such applications and the fast response required for interaction with haptics. The efficacy of the framework was tested by implementation of a minimally invasive surgery simulator. CONCLUSION: A decoupled simulation approach can be implemented as a framework to handle simultaneous processes of the system at the various frame rates each process requires. The framework was successfully used to develop collaborative virtual environments (VEs) involving geographically distributed users connected through a network, with the results comparable to VEs for local users.

Efficient liver surgery planning in 3D based on functional segment classification and volumetric information.

Anatomic hepatectomies are resections in which compromised segments or sectors of the liver are extracted according to the topological structure of its vascular elements. Such structure varies considerably among patients, which makes the current anatomy-based planning methods often inaccurate. In this work we propose a strategy to efficiently and semi-automatically segment and classify patient-specific liver models in 3D. The method is based on standard CT datasets and allows accurate estimation of functional remaining liver volume. Experiments showing effectiveness of the method are presented, and quantitative and qualitative results are discussed.

Using the PhysX engine for physics-based virtual surgery with force feedback.

BACKGROUND: The development of modern surgical simulators is highly challenging, as they must support complex simulation environments. The demand for higher realism in such simulators has driven researchers to adopt physics-based models, which are computationally very demanding. This poses a major problem, since real-time interactions must permit graphical updates of 30 Hz and a much higher rate of 1 kHz for force feedback (haptics). Recently several physics engines have been developed which offer multi-physics simulation capabilities, including rigid and deformable bodies, cloth and fluids. While such physics engines provide unique opportunities for the development of surgical simulators, their higher latencies, compared to what is necessary for real-time graphics and haptics, offer significant barriers to their use in interactive simulation environments. METHODS: In this work, we propose solutions to this problem and demonstrate how a multimodal surgical simulation environment may be developed based on NVIDIA's PhysX physics library. Hence, models that are undergoing relatively low-frequency updates in PhysX can exist in an environment that demands much higher frequency updates for haptics. We use a collision handling layer to interface between the physical response provided by PhysX and the haptic rendering device to provide both real-time tissue response and force feedback. RESULTS: Our simulator integrates a bimanual haptic interface for force feedback and per-pixel shaders for graphics realism in real time. To demonstrate the effectiveness of our approach, we present the simulation of the laparoscopic adjustable gastric banding (LAGB) procedure as a case study. CONCLUSIONS: To develop complex and realistic surgical trainers with realistic organ geometries and tissue properties demands stable physics-based deformation methods, which are not always compatible with the interaction level required for such trainers. We have shown that combining different modelling strategies for behaviour, collision and graphics is possible and desirable. Such multimodal environments enable suitable rates to simulate the major steps of the LAGB procedure.

Fratura de agulha - relato de caso / Broken needle – two cases report

O tratamento odontológico não está isento de riscos. Acidentes e complicações podem estar associados aos vários tipos de procedimentos realizados pelo cirurgião dentista. Dentre as especialidades, a cirurgia bucomaxilofacial é aquela em que o profissional está mais exposto a encontrar complicações. As agulhas anestésicas conduzem a solução anestésica do tubete aos tecidos, levando à interrupção temporária na condução dos impulsos dolorosos. Nas primeiras décadas do século 20, eram utilizadas, para anestesia local, agulhas rígidas e não descartáveis, o que proporcionava uma incidência muito grande de fraturas de agulha durante bloqueios anestésicos. Com a evolução tecnológica, foram desenvolvidas as agulhas descartáveis, fabricadas com aço inoxidável flexível, permitindo, assim, que esse tipo de complicação se tornasse rara. Atualmente, as razões para esse acidente são variadas e podem estar associadas a falhas na fabricação da agulha, a movimentação súbita do paciente durante a punção, ou a erros de técnica profissional, como a inserção de toda a haste da agulha no tecido-alvo, dobra da haste da agulha ou a sua reutilização excessiva, o que provoca a fadiga do metal. Este artigo se propõe a apresentar um relato de caso de fratura de agulha anestésica tratado no serviço de cirurgia e traumatologia bucomaxilofacial do Hospital Santo Antônio das Obras Sociais Irmã Dulce.

Linfoma não-Hodgkin de células B: relato de caso / Non-Hodgkin B-cells lymphoma: a case report

O linfoma não-Hodgkin é a segunda neoplasia maligna mais prevalente na região da cabeça e do pescoço. A etiologia dos linfomas é desconhecida, porém distúrbios da função imune podem conferir risco aumentado para essa desordem. Apesar de as lesões orais serem freqüentemente uma manifestação da doença amplamente disseminada, às vezes, o linfoma origina-se nos tecidos bucais, sem ter se espalhado para outros locais. Paciente P.J.Q., 54 anos, masculino, compareceu ao ambulatório de cirurgia bucomaxilofacial do hospital São Félix, queixando-se de aumento de volume na região zigomática e na paranasal esquerda, após trauma severo por martelo, há um mês. Foi realizado punção aspiratória sem ter sido detectado qualquer sinal de infecção ou líquido cístico, no exame radiográfico não foi detectado fratura óssea. Foi colhido material para análise histopatológica, tendo como diagnóstico: linfoma não-Hodgkin de células B. Um ano após o diagnóstico, finalizado o tratamento quimioterápico, o paciente retornou para revisão quando foram realizadas extrações dentárias, para que pudesse ser submetido à radioterapia sem complicações. A proposta deste trabalho é apresentar um caso clínico de linfoma não-Hodgkin de células B, enfatizando a sua importância clínica e a necessidade do diagnóstico precoce na tentativa de melhorar a qualidade de vida dos pacientes.

Non-Hodgkin lymphoma is the second most prevalent malignant cancer of the head and neck region. The etiology of the lymphomas is unknown, but disorders of the immune function may result in an increased risk for this disease. Despite the fact that oral lesions are frequently a manifestation of a systemic pathology, the lymphoma sometimes arises in the oral tissues, without spreading to other organs. Patient P.J.Q, a fifty-four-year-old, male, presented at the clinic of oral and maxillofacial surgery of São Felix hospital, complaining of a swelling located in the left zygomatic paranasal region resulting from severe trauma caused by a hammer a month previously. Aspiratory punction was performed, resulting in no signs of infection or cystic fluid, and a radiograph revealed no bone fracture. Material from the lesion was collected sent for a histopathology examination, which led to a diagnosis of non-Hodgkin B-cell lymphoma. One year after the diagnosis, having completed a course of chemotherapy, the patient returned for follow-up, at which time some teeth were extracted so that the patient could undergo radiotherapy without complications. The aim of this paper is to present a case report of non-Hodgkin B-cell lymphoma, emphasizing the clinical importance of the disease and the need for an early diagnosis with a view to achieving a better life quality for patients.

An efficient dynamic point algorithm for line-based collision detection in real time surgery simulation involving haptics.

In this paper, we introduce a novel "dynamic point" algorithm for computing the interaction of a line-shaped haptic cursor and polygonal surface models which has a near constant complexity. The algorithm is applied in laparoscopic surgery simulation for interaction of surgical instruments with physics-based deformable organ models.

Physics-based real time laparoscopic electrosurgery simulation.

While physics-based modeling of electrosurgical procedures is essential for most laparoscopic simulation systems, we present such a system for the first time in this paper. We have implemented a physics-based model of electrosurgery to control the temperature distribution on the tissue as a function of time. Then, we evaluate the algorithm within a complete graphics-haptics-physics-based system.

Towards a virtual basic laparoscopic skill trainer (VBLaST).

Surgical skill training is a long and tedious process of acquiring fine motor skills. To overcome the drawbacks of the existing toolbox trainer systems, we develop, for the first time, a virtual basic laparoscopic skill trainer (VBLaST) whereby tasks, such as the ones available in the FLS toolbox system, may be performed on the computer.