RESUMEN

UAVs have been widely used in deformation monitoring because of their high availability and flexibility. However, the quality of UAV images is affected by changing attitude and surveying environments, resulting in a low monitoring accuracy. Cross-shaped markers are used to improve the accuracy of UAV monitoring due to their distinct center contrast and absence of eccentricity. However, existing methods cannot rapidly and precisely detect these markers in UAV images. To address these problems, this paper proposes an adaptive Radon-transform-based marker detection and localization method for UAV displacement measurements, focusing on two critical detection parameters, namely, the radius of marker information acquisition and the edge width of the cross-shaped scoring template. The experimental results show that the marker detection rate is 97.2% under different combinations of flight altitudes, radius ratios of marker information acquisition, and marker sizes. Furthermore, the root mean square error of detection and localization is 0.57 pixels, significantly surpassing the performance and accuracy of other methods. We also derive the critical detection radius and appropriate parameter combinations for different heights to further improve the practicality of the method.

RESUMEN

Barcodes are visual representations of data widely used in commerce and administration to compactly codify information about objects, services, and people. Specifically, a barcode is an image composed of parallel lines, with different widths, spacing and sizes. Generally, the lines are dark (usually black) on a bright background (usually white) or vice-versa. Thanks to this representation, barcodes can be detected and decoded in a way robust to changes of light and noise. However, using barcodes with several colours for the lines is quite intriguing because it enables boosting the barcode's data capacity. Colour barcodes still pose a challenge today, even though numerous studies on the topic were conducted between 1990 and 2022. The main issue that needs to be solved is the creation of an optical technology able to decode colour sequences regardless of the ambient light, the acquisition and printing or visualisation device, and the physical support on which the barcode is printed or displayed. To the best of our knowledge, the studies currently available in literature do not provide the experimental data on which they are based, nor are there online databases that can be used for further studies or for training data analysis procedures based on artificial intelligence techniques. To fill this gap and push further research in this technology, we built COCO-10, a public dataset of colour barcode images, that would like to become a testbench for the development and testing of colour barcode decoding algorithms, taking into account the colour variability due to the light, to the printer and camera gamuts and to the quality of the paper on which the barcode is printed. COCO-10 contains 5400 images of 150 colour barcodes, each of one printed on two white papers with different density and printers and acquired under six illuminations by three smartphones' cameras. For each colour barcode image, a mask identifying the region occupied by the barcode is released too. The 150 colour barcodes have been generated by colouring the lines of black & white barcodes with colours randomly selected from a palette of ten colours including both warm and colour hues. The name COCO-10 just refers to the fact that the dataset contains COlor BarCOdes with 10 possible colours for each line. We also provide a set of 300 images created as follows. The 150 COCO-10 colour barcodes were synthetically superimposed on 150 cluttered backgrounds, resulting in 150 images. The first 75 (group 1) were printed on thick paper, the others (group 2) on plain paper. Each group was further subdivided into subsets of 25 images, resulting in 3 subgroups, each of which was captured by 2 smartphones' cameras under one of the 6 illuminants mentioned above. We also provide masks for these images. These images would like to be a benchmark for testing the accuracy of barcode decoding algorithms, bearing in mind that the performance of these algorithms may be influenced by the accuracy of the previous detection of the barcodes themselves in the background. The total number of images in COCO-10 is 11700, including the 300 synthetic images of the colour barcodes displayed on white and cluttered background, the 5700 real-world images of the colour barcodes printed on white papers and with cluttered backgrounds and their corresponding 5700 masks. We finally highlight that COCO-10 can be also used for developing and testing algorithms for gamut and tone mapping, machine colour constancy, and colour correction.

RESUMEN

Fiducial marker detection in electron micrographs becomes an important and challenging task with the development of large-field electron microscopy. The fiducial marker detection plays an important role in several steps during the process of electron micrographs, such as the alignment and parameter calibrations. However, limited by the conditions of low signal-to-noise ratio (SNR) in the electron micrographs, the performance of fiducial marker detection is severely affected. In this work, we propose the MarkerDetector, a novel algorithm for detecting fiducial markers in electron micrographs. The proposed MarkerDetector is built upon the following contributions: Firstly, a wavelet-based template generation algorithm is devised in MarkerDetector. By adopting a shape-based criterion, a high-quality template can be obtained. Secondly, a robust marker determination strategy is devised by utilizing statistic-based filtering, which can guarantee the correctness of the detected fiducial markers. The average running time of our algorithm is 1.67 seconds with promising accuracy, indicating its practical feasibility for applications in electron micrographs.

Asunto(s)

RESUMEN

Trace detection of argininosuccinate synthetase 1 (ASS1), a depression marker, in urine samples is difficult to achieve. In this work, a dual-epitope-peptides imprinted sensor for ASS1 detection in urine was constructed based on the high selectivity and sensitivity of the "epitope imprinting approach". First, two cysteine-modified epitope-peptides were immobilized onto gold nanoparticles (AuNPs) deposited on a flexible electrode (ITO-PET) by gold-sulfur bonds (Au-S), then a controlled electropolymerization of dopamine was carried out to imprint the epitope peptides. After removing epitope-peptides, the dual-epitope-peptides imprinted sensor (MIP/AuNPs/ITO-PET) which with multiple binding sites for ASS1 was obtained. Compared with single epitope-peptide, dual-epitope-peptides imprinted sensor had higher sensitivity, which presented a linear range from 0.15 to 6000 pg ml-1 with a low limit of detection (LOD = 0.106 pg mL-1, S/N = 3). It had good reproducibility (RSD = 1.74%), repeatability (RSD = 3.60%), stability (RSD = 2.98%), and good selectivity, and the sensor had good recovery (92.4%-99.0%) in urine samples. This is the first highly sensitive and selective electrochemical assay for the depression marker ASS1 in urine, which is expected to provide help for the non-invasive and objective diagnosis of depression.

Asunto(s)

RESUMEN

PURPOSE: To assess dynamic tumor tracking (DTT) target localization uncertainty for in-vivo marker-based stereotactic ablative radiotherapy (SABR) treatments of the liver using electronic-portal-imaging-device (EPID) images. The Planning Target Volume (PTV) margin contribution for DTT is estimated. METHODS: Phantom and patient EPID images were acquired during non-coplanar 3DCRT-DTT delivered on a Vero4DRT linac. A chain-code algorithm was applied to detect Multileaf Collimator (MLC)-defined radiation field edges. Gold-seed markers were detected using a connected neighbor algorithm. For each EPID image, the absolute differences between the measured center-of-mass (COM) of the markers relative to the aperture-center (Tracking Error, (ET )) was reported in pan, tilt, and 2D-vector directions at the isocenter-plane. PHANTOM STUDY: An acrylic cube phantom implanted with gold-seed markers was irradiated with non-coplanar 3DCRT-DTT beams and EPID images collected. Patient Study: Eight liver SABR patients were treated with non-coplanar 3DCRT-DTT beams. All patients had three to four implanted gold-markers. In-vivo EPID images were analyzed. RESULTS: Phantom Study: On the 125 EPID images collected, 100% of the markers were identified. The average ± SD of ET were 0.24 ± 0.21, 0.47 ± 0.38, and 0.58 ± 0.37 mm in pan, tilt and 2D directions, respectively. Patient Study: Of the 1430 EPID patient images acquired, 78% had detectable markers. Over all patients, the average ± SD of ET was 0.33 ± 0.41 mm in pan, 0.63 ± 0.75 mm in tilt and 0.77 ± 0.80 mm in 2D directions The random 2D-error, σ, for all patients was 0.79 mm and the systematic 2D-error, Σ, was 0.20 mm. Using the Van Herk margin formula 1.1 mm planning target margin can represent the marker based DTT uncertainty. CONCLUSIONS: Marker-based DTT uncertainty can be evaluated in-vivo on a field-by-field basis using EPID images. This information can contribute to PTV margin calculations for DTT.

Asunto(s)

RESUMEN

The outbreak of COVID-19 has positively impacted the NGS market recently. Targeted sequencing (TS) has become an important routine technique in both clinical and research settings, with advantages including high confidence and accuracy, a reasonable turnaround time, relatively low cost, and fewer data burdens with the level of bioinformatics or computational demand. Since there are no clear consensus guidelines on the wide range of next-generation sequencing (NGS) platforms and techniques, there is a vital need for researchers and clinicians to develop efficient approaches, especially for the molecular diagnosis of diseases in the emergency of the disease and the global pandemic outbreak of COVID-19. In this review, we aim to summarize different methods of TS, demonstrate parameters for TS assay designs, illustrate different TS panels, discuss their limitations, and present the challenges of TS concerning their clinical application for the molecular diagnosis of human diseases.

Asunto(s)

RESUMEN

The automatic positioning of machines in a large number of application areas is an important aspect of automation. Today, this is often done using classic geodetic sensors such as Global Navigation Satellite Systems (GNSS) and robotic total stations. In this work, a stereo camera system was developed that localizes a machine at high frequency and serves as an alternative to the previously mentioned sensors. For this purpose, algorithms were developed that detect active markers on the machine in a stereo image pair, find stereo point correspondences, and estimate the pose of the machine from these. Theoretical influences and accuracies for different systems were estimated with a Monte Carlo simulation, on the basis of which the stereo camera system was designed. Field measurements were used to evaluate the actual achievable accuracies and the robustness of the prototype system. The comparison is present with reference measurements with a laser tracker. The estimated object pose achieved accuracies higher than 16 mm with the translation components and accuracies higher than 3 mrad with the rotation components. As a result, 3D point accuracies higher than 16 mm were achieved by the machine. For the first time, a prototype could be developed that represents an alternative, powerful image-based localization method for machines to the classical geodetic sensors.

RESUMEN

Vehicles carrying hazardous material (hazmat) are severe threats to the safety of highway transportation, and a model that can automatically recognize hazmat markers installed or attached on vehicles is essential for intelligent management systems. However, there is still no public dataset for benchmarking the task of hazmat marker detection. To this end, this paper releases a large-scale vehicle hazmat marker dataset named VisInt-VHM, which includes 10,000 images with a total of 20,023 hazmat markers captured under different environmental conditions from a real-world highway. Meanwhile, we provide an compact hazmat marker detection network named HMD-Net, which utilizes a revised lightweight backbone and is further compressed by channel pruning. As a consequence, the trained-model can be efficiently deployed on a resource-restricted edge device. Experimental results demonstrate that compared with some established methods such as YOLOv3, YOLOv4, their lightweight versions and popular lightweight models, HMD-Net can achieve a better trade-off between the detection accuracy and the inference speed.

RESUMEN

PURPOSE: Fiducial markers are commonly used in navigation-assisted minimally invasive spine surgery and they help transfer image coordinates into real-world coordinates. In practice, these markers might be located outside the field-of-view (FOV) of C-arm cone-beam computed tomography (CBCT) systems used in intraoperative surgeries, due to the limited detector sizes. As a consequence, reconstructed markers in CBCT volumes suffer from artifacts and have distorted shapes, which sets an obstacle for navigation. METHODS: In this work, we propose two fiducial marker detection methods: direct detection from distorted markers (direct method) and detection after marker recovery (recovery method). For direct detection from distorted markers in reconstructed volumes, an efficient automatic marker detection method using two neural networks and a conventional circle detection algorithm is proposed. For marker recovery, a task-specific data preparation strategy is proposed to recover markers from severely truncated data. Afterwards, a conventional marker detection algorithm is applied for position detection. The networks in both methods are trained based on simulated data. For the direct method, 6800 images and 10 000 images are generated, respectively, to train the U-Net and ResNet50. For the recovery method, the training set includes 1360 images for FBPConvNet and Pix2pixGAN. The simulated data set with 166 markers and four cadaver cases with real fiducials are used for evaluation. RESULTS: The two methods are evaluated on simulated data and real cadaver data. The direct method achieves 100% detection rates within 1 mm detection error on simulated data with normal truncation and simulated data with heavier noise, but only detect 94.6% markers in extremely severe truncation case. The recovery method detects all the markers successfully in three test data sets and around 95% markers are detected within 0.5 mm error. For real cadaver data, both methods achieve 100% marker detection rates with mean registration error below 0.2 mm. CONCLUSIONS: Our experiments demonstrate that the direct method is capable of detecting distorted markers accurately and the recovery method with the task-specific data preparation strategy has high robustness and generalizability on various data sets. The task-specific data preparation is able to reconstruct structures of interest outside the FOV from severely truncated data better than conventional data preparation.

Asunto(s)

RESUMEN

PURPOSE: For the visualization of pulmonary ventilation with Electrical Impedance Tomography (EIT) most devices use standard reconstruction models, featuring common thorax dimensions and predetermined electrode locations. Any discrepancies between the available model and the patient in terms of body shape and electrode position lead to incorrectly displayed impedance distributions. This work addresses that problem by presenting and evaluating a method for 3D model generation of the thorax and any affixed electrodes based on handheld video-footage. METHODS: Therefore, a process was developed, providing users with the ability to capture a patient's chest and the attached electrodes via smartphone. Once data is collected, extracted images are used to generate a 3D model with a structure from motion approach and locate electrodes with ArUco markers. For the evaluation of the developed method, multiple tests were performed in laboratory environments, which were compared with manually created reference models and differences quantified based on mean distance, standard deviation, and maximum distance. RESULTS: The implemented workflow allows for automated model reconstruction based on videos or selected images captured with a handheld device. It generates sparse point clouds from which a surface mesh is reconstructed and returns relative coordinates of any identified ArUco marker. The average value for the mean distance error of two model generations was 5.4 mm while the mean standard deviation was 6.0 mm. The average runtime of twelve reconstructions was 5:17 min, with a minimal runtime of 3:22 min and a maximal runtime of 7:29 min. CONCLUSION: The presented methods and results show that model reconstruction of a patient's thorax and applied electrodes at an emergency site is feasible with already available devices. This is a first step toward the automated generation of patient-specific reconstruction models for Electrical Impedance Tomography based on images recorded with handheld devices.

Asunto(s)

RESUMEN

PURPOSE: Gold fiducial markers are used to guide liver stereotactic body radiation therapy (SBRT) and are hard to detect by magnetic resonance imaging (MRI). In this study, the parameters of the three-dimensional T1-weighted turbo gradient-echo (3D T1W-GRE) sequence were optimized for gold marker detection without degrading tumor delineation. METHODS: Custom-made phantoms mimicking tumor and normal liver parenchyma were prepared and embedded with a gold marker. The 3D T1W-GRE was scanned by varying echo time (TE), bandwidth (BW), flip angle (FA), and base matrix size. The signal-to-noise ratio (SNR), contrast ratio (CR), and relative standard deviation (RSD) of the signal intensity in the area including the gold marker were evaluated, and the parameters were optimized accordingly. The modified 3D T1W-GRE (called HYBRID) was compared with the conventional T1W-GRE- and T2*-sequences in both phantom and clinical studies. In the clinical study of six patients with primary liver tumors, two observers visually assessed marker detection, tumor delineation, and overall image quality on a four-point scale. RESULTS: In the phantom study, HYBRID showed significantly higher SNR and RSD than those of conventional T1W-GRE (P < 0.001). In the clinical study, HYBRID yielded significantly higher scores than conventional T1W-GRE did in terms of marker detection (P < 0.001). The scores of both sequences were not statistically different in terms of tumor delineation and overall image quality (P = 0.56 and P = 0.32). CONCLUSIONS: The proposed HYBRID sequence improved gold fiducial marker detection without degrading tumor delineation in MRI for SBRT of primary liver tumor.

Asunto(s)

RESUMEN

Almost from the time of its discovery, the prostate specific antigen (PSA) has been one of the most accurate and most extensively studied indicators of prostate cancer (PC). Because of advancements in biosensing systems and technology, PSA analysis methods have been substantially updated and enhanced as compared to their first instances. With the development of techniques in biosensor technology, the number of PSA biosensors that can be used in the biomedical sector is increasing year by year. Many different recognition elements and transducers have been used in the development of biosensor systems that exhibit high sensitivity, selectivity, and specificity. Here in this review, we provide a current overview of the different approaches to PSA detection.

Asunto(s)

RESUMEN

OBJECTIVE: Fluoroscopic guidance is a critical step for the puncture procedure in percutaneous endoscopic transforaminal discectomy (PETD). However, two-dimensional observations of the three-dimensional anatomic structure suffer from the effects of projective simplification. To accurately assess the spatial relations between the patient vertebra tissues and puncture needle, a considerable number of fluoroscopic images from different orientations need to be acquired by the surgeons. This process significantly increases the radiation risk for both the patient and surgeons. METHODS: In this paper, we propose an augmented reality (AR) surgical navigation system for PETD based on multi-modality information, which contains fluoroscopy, optical tracking, and depth camera. To register the fluoroscopic image with the intraoperative video, we design a lightweight non-invasive fiducial with markers and detect the markers based on the deep learning method. It can display the intraoperative video fused with the registered fluoroscopic images. We also present a self-adaptive calibration and transformation method between a 6-DOF optical tracking device and a depth camera, which are in different coordinate systems. RESULTS: With the substantially reduced frequency of fluoroscopy imaging, the system can accurately track and superimpose the virtual puncture needle on fluoroscopy images in real-time. From operating theatre in vivo animal experiments, the results illustrate that the system average positioning accuracy can reach 1.98mm and the orientation accuracy can reach 1.19∘. From the clinical validation results, the system significantly lower the frequency of fluoroscopy imaging (42.7%) and reduce the radiation risk for both the patient and surgeons. CONCLUSION: Coupled with the user study, both the quantitative and qualitative results indicate that our navigation system has the potential to be highly useful in clinical practice. Compared with the existing navigation systems, which are usually equipped with a variety of large and high-cost medical equipments, such as O-arm, cone-beam CT, and robots, our navigation system does not need special equipment and can be implemented with common equipment in the operating room, such as C-arm, desktop, etc., even in small hospitals.

Asunto(s)

RESUMEN

Landing an unmanned aerial vehicle (UAV) autonomously and safely is a challenging task. Although the existing approaches have resolved the problem of precise landing by identifying a specific landing marker using the UAV's onboard vision system, the vast majority of these works are conducted in either daytime or well-illuminated laboratory environments. In contrast, very few researchers have investigated the possibility of landing in low-illumination conditions by employing various active light sources to lighten the markers. In this paper, a novel vision system design is proposed to tackle UAV landing in outdoor extreme low-illumination environments without the need to apply an active light source to the marker. We use a model-based enhancement scheme to improve the quality and brightness of the onboard captured images, then present a hierarchical-based method consisting of a decision tree with an associated light-weight convolutional neural network (CNN) for coarse-to-fine landing marker localization, where the key information of the marker is extracted and reserved for post-processing, such as pose estimation and landing control. Extensive evaluations have been conducted to demonstrate the robustness, accuracy, and real-time performance of the proposed vision system. Field experiments across a variety of outdoor nighttime scenarios with an average luminance of 5 lx at the marker locations have proven the feasibility and practicability of the system.

RESUMEN

HIGHLIGHTS: A zero-reflection-induced phase singularity is achieved through precisely controlling the resonance characteristics using two-dimensional nanomaterials. An atomically thin nano-layer having a high absorption coefficient is exploited to enhance the zero-reflection dip, which has led to the subsequent phase singularity and thus a giant lateral position shift. We have improved the detection limit of low molecular weight molecules by more than three orders of magnitude compared to current state-of-art nanomaterial-enhanced plasmonic sensors. Detection of small cancer biomarkers with low molecular weight and a low concentration range has always been challenging yet urgent in many clinical applications such as diagnosing early-stage cancer, monitoring treatment and detecting relapse. Here, a highly enhanced plasmonic biosensor that can overcome this challenge is developed using atomically thin two-dimensional phase change nanomaterial. By precisely engineering the configuration with atomically thin materials, the phase singularity has been successfully achieved with a significantly enhanced lateral position shift effect. Based on our knowledge, it is the first experimental demonstration of a lateral position signal change > 340 µm at a sensing interface from all optical techniques. With this enhanced plasmonic effect, the detection limit has been experimentally demonstrated to be 10-15 mol L-1 for TNF-α cancer marker, which has been found in various human diseases including inflammatory diseases and different kinds of cancer. The as-reported novel integration of atomically thin Ge2Sb2Te5 with plasmonic substrate, which results in a phase singularity and thus a giant lateral position shift, enables the detection of cancer markers with low molecular weight at femtomolar level. These results will definitely hold promising potential in biomedical application and clinical diagnostics.

RESUMEN

Foodborne illness is a major public health issue that results in millions of global infections annually. The burden of such illness sits mostly with developing countries, as access to advanced laboratory equipment and skilled lab technicians, as well as consistent power sources, is limited and expensive. Current gold standards in foodborne pathogen screening involve labor-intensive sample enrichment steps, pathogen isolation and purification, and costly readout machinery. Overall, time to detection can take multiple days, excluding the time it takes to ship samples to off-site laboratories. Efforts have been made to simplify the workflow of such tests by integrating multiple steps of foodborne pathogen screening procedures into a singular device, as well as implementing more point-of-need readout methods. In this review, we explore recent advancements in developing point-of-need devices for foodborne pathogen screening. We discuss the detection of surface markers, nucleic acids, and metabolic products using both paper-based and microfluidic devices, focusing primarily on developments that have been made between 2015 and mid-2020.

Asunto(s)

RESUMEN

PURPOSE: This study aimed to design a fully automated framework to evaluate intrafraction motion using orthogonal x-ray images from CyberKnife. METHODS: The proposed framework includes three modules: (a) automated fiducial marker detection, (b) three-dimensional (3D) position reconstruction, and (c) intrafraction motion evaluation. A total of 5927 images from real patients treated with CyberKnife fiducial tracking were collected. The ground truth was established by labeling coarse bounding boxes manually, and binary mask images were then obtained by applying a binary threshold and filter. These images and labels were used to train a detection model using a fully convolutional network (fCN). The output of the detection model can be used to reconstruct the 3D positions of the fiducial markers and then evaluate the intrafraction motion via a rigid transformation. For a patient test, the motion amplitudes, rotations, and fiducial cohort deformations were calculated used the developed framework for 13 patients with a total of 52 fractions. RESULTS: The precision and recall of the fiducial marker detection model were 98.6% and 95.6%, respectively, showing high model performance. The mean (±SD) centroid error between the predicted fiducial markers and the ground truth was 0.25 ± 0.47 pixels on the test data. For intrafraction motion evaluation, the mean (±SD) translations in the superior-posterior (SI), left-right (LR), and anterior-posterior (AP) directions were 13.1 ± 2.2 mm, 2.0 ± 0.4 mm, and 5.2 ± 1.4 mm, respectively, and the mean (±SD) rotations in the roll, pitch and yaw directions were 2.9 ± 1.5°, 2.5 ± 1.5°, and 3.1 ± 2.2°. Seventy-one percent of the fractions had rotations larger than the system limitations. With rotation correction during rigid registration, only 2 of the 52 fractions had residual errors larger than 2 mm in any direction, while without rotation correction, the probability of large residual errors increased to 46.2%. CONCLUSION: We developed a framework with high performance and accuracy for automatic fiducial marker detection, which can be used to evaluate intrafraction motion using orthogonal x-ray images from CyberKnife. For liver patients, most fractions have fiducial cohort rotations larger than the system limitations; however, the fiducial cohort deformation is small, especially for the scenario with rotation correction.

Asunto(s)

RESUMEN

Deep learning-based marker detection for autonomous drone landing is widely studied, due to its superior detection performance. However, no study was reported to address non-uniform motion-blurred input images, and most of the previous handcrafted and deep learning-based methods failed to operate with these challenging inputs. To solve this problem, we propose a deep learning-based marker detection method for autonomous drone landing, by (1) introducing a two-phase framework of deblurring and object detection, by adopting a slimmed version of deblur generative adversarial network (DeblurGAN) model and a You only look once version 2 (YOLOv2) detector, respectively, and (2) considering the balance between the processing time and accuracy of the system. To this end, we propose a channel-pruning framework for slimming the DeblurGAN model called SlimDeblurGAN, without significant accuracy degradation. The experimental results on the two datasets showed that our proposed method exhibited higher performance and greater robustness than the previous methods, in both deburring and marker detection.

RESUMEN

BACKGROUND: Accurate autonomous marker detection and measurement is essential for high precision anatomical registration. The measurement should be in real-time, accurate, and robust to the varied conditions of the operation theatre. METHODS: The purpose is to design and implement a robust real-time algorithm to measure the coordinates of the point on the marker for robot-based autonomous registration and surgery. The algorithm is built in two parts based on the recursive Taguchi method. The first part deals with the detection of markers. In the second part, the center of the marker is located, and the coordinates are measured by fitting the concentric ellipse. RESULTS: Three case studies are presented where the algorithm is tested for extreme conditions of uneven lighting, distorted color, surface distortions, and significant random orientation of the marker. The robustness of the algorithm in successfully detecting and measuring in real-time is presented. CONCLUSION: The algorithm is successfully implemented for real-time detection and coordinate measurement of the markers.

Asunto(s)

RESUMEN

PURPOSE: Real-time image-guided adaptive radiation therapy (IGART) requires accurate marker segmentation to resolve three-dimensional (3D) motion based on two-dimensional (2D) fluoroscopic images. Most common marker segmentation methods require prior knowledge of marker properties to construct a template. If marker properties are not known, an additional learning period is required to build the template which exposes the patient to an additional imaging dose. This work investigates a deep learning-based fiducial marker classifier for use in real-time IGART that requires no prior patient-specific data or additional learning periods. The proposed tracking system uses convolutional neural network (CNN) models to segment cylindrical and arbitrarily shaped fiducial markers. METHODS: The tracking system uses a tracking window approach to perform sliding window classification of each implanted marker. Three cylindrical marker training datasets were generated from phantom kilovoltage (kV) and patient intrafraction images with increasing levels of megavoltage (MV) scatter. The cylindrical shaped marker CNNs were validated on unseen kV fluoroscopic images from 12 fractions of 10 prostate cancer patients with implanted gold fiducials. For the training and validation of the arbitrarily shaped marker CNNs, cone beam computed tomography (CBCT) projection images from ten fractions of seven lung cancer patients with implanted coiled markers were used. The arbitrarily shaped marker CNNs were trained using three patients and the other four unseen patients were used for validation. The effects of full training using a compact CNN (four layers with learnable weights) and transfer learning using a pretrained CNN (AlexNet, eight layers with learnable weights) were analyzed. Each CNN was evaluated using a Precision-Recall curve (PRC), the area under the PRC plot (AUC), and by the calculation of sensitivity and specificity. The tracking system was assessed using the validation data and the accuracy was quantified by calculating the mean error, root-mean-square error (RMSE) and the 1st and 99th percentiles of the error. RESULTS: The fully trained CNN on the dataset with moderate noise levels had a sensitivity of 99.00% and specificity of 98.92%. Transfer learning of AlexNet resulted in a sensitivity and specificity of 99.42% and 98.13%, respectively, for the same datasets. For the arbitrarily shaped marker CNNs, the sensitivity was 98.58% and specificity was 98.97% for the fully trained CNN. The transfer learning CNN had a sensitivity and specificity of 98.49% and 99.56%, respectively. The CNNs were successfully incorporated into a multiple object tracking system for both cylindrical and arbitrarily shaped markers. The cylindrical shaped marker tracking had a mean RMSE of 1.6 ± 0.2 pixels and 1.3 ± 0.4 pixels in the x- and y-directions, respectively. The arbitrarily shaped marker tracking had a mean RMSE of 3.0 ± 0.5 pixels and 2.2 ± 0.4 pixels in the x- and y-directions, respectively. CONCLUSION: With deep learning CNNs, high classification performances on unseen patient images were achieved for both cylindrical and arbitrarily shaped markers. Furthermore, the application of CNN models to intrafraction monitoring was demonstrated using a simple tracking system. The results demonstrate that CNN models can be used to track markers without prior knowledge of the marker properties or an additional learning period.

Asunto(s)