Automated review of patient position in DIBH breast hybrid IMRT with EPID images.

Deep Inspiration Breath Hold (DIBH) is a respiratory-gating technique adopted in radiation therapy to lower cardiac irradiation. When performing DIBH treatments, it is important to have a monitoring system to ensure the patient's breath hold level is stable and reproducible at each fraction. In this retrospective study, we developed a system capable of monitoring DIBH breast treatments by utilizing cine EPID images taken during treatment. Setup error and intrafraction motion were measured for all fractions of 20 left-sided breast patients. All patients were treated with a hybrid static-IMRT technique, with EPID images from the static fields analyzed. Ten patients had open static fields and the other ten patients had static fields partially blocked with the multileaf collimator (MLC). Three image-processing algorithms were evaluated on their ability to accurately measure the chest wall position (CWP) in EPID images. CWP measurements were recorded along a 61-pixel region of interest centered along the midline of the image. The median and standard deviation of the CWP were recorded for each image. The algorithm showing the highest agreement with manual measurements was then used to calculate intrafraction motion and setup error. To measure intrafraction motion, the median CWP of the first EPID frame was compared with that of the subsequent EPID images of the treatment. The maximum difference was recorded as the intrafraction motion. The setup error was calculated as the difference in median CWP between the MV DRR and the first EPID image of the lateral tangential field. The results showed that the most accurate image-processing algorithm can identify the chest wall within 1.2 mm on both EPID and MV DRR images, and measures intrafraction motion and setup errors within 1.4 mm.

Review and Analysis of Tumour Detection and Image Quality Analysis in Experimental Breast Microwave Sensing.

This review evaluates the methods used for image quality analysis and tumour detection in experimental breast microwave sensing (BMS), a developing technology being investigated for breast cancer detection. This article examines the methods used for image quality analysis and the estimated diagnostic performance of BMS for image-based and machine-learning tumour detection approaches. The majority of image analysis performed in BMS has been qualitative and existing quantitative image quality metrics aim to describe image contrast-other aspects of image quality have not been addressed. Image-based diagnostic sensitivities between 63 and 100% have been achieved in eleven trials, but only four articles have estimated the specificity of BMS. The estimates range from 20 to 65%, and do not demonstrate the clinical utility of the modality. Despite over two decades of research in BMS, significant challenges remain that limit the development of this modality as a clinical tool. The BMS community should utilize consistent image quality metric definitions and include image resolution, noise, and artifacts in their analyses. Future work should include more robust metrics, estimates of the diagnostic specificity of the modality, and machine-learning applications should be used with more diverse datasets and with robust methodologies to further enhance BMS as a viable clinical technique.

An Optimization-Based Approach to Radar Image Reconstruction in Breast Microwave Sensing.

Breast microwave sensing (BMS) has been studied as a potential technique for cancer detection due to the observed microwave properties of malignant and healthy breast tissues. This work presents a novel radar-based image reconstruction algorithm for use in BMS that reframes the radar image reconstruction process as an optimization problem. A gradient descent optimizer was used to create an optimization-based radar reconstruction (ORR) algorithm. Two hundred scans of MRI-derived breast phantoms were performed with a preclinical BMS system. These scans were reconstructed using the ORR, delay-and-sum (DAS), and delay-multiply-and-sum (DMAS) beamformers. The ORR was observed to improve both sensitivity and specificity compared to DAS and DMAS. The estimated sensitivity and specificity of the DAS beamformer were 19% and 44%, respectively, while for ORR, they were 27% and 56%, representing a relative increase of 42% and 27%. The DAS reconstructions also exhibited a hot-spot image artifact, where a localized region of high intensity that did not correspond to any physical phantom feature would be present in an image. This artifact appeared like a tumour response within the image and contributed to the lower specificity of the DAS beamformer. This artifact was not observed in the ORR reconstructions. This work demonstrates the potential of an optimization-based conceptualization of the radar image reconstruction problem in BMS. The ORR algorithm implemented in this work showed improved diagnostic performance and fewer image artifacts compared to the widely employed DAS algorithm.

The use of 0.5rcav as an effective point of measurement for cylindrical chambers may result in a systematic shift of electron percentage depth doses.

Electron dosimetry can be performed using cylindrical chambers, plane-parallel chambers, and diode detectors. The finite volume of these detectors results in a displacement effect which is taken into account using an effective point of measurement (EPOM). Dosimetry protocols have recommended a shift of 0.5 rcav for cylindrical chambers; however, various studies have shown that the optimal shift may deviate from this recommended value. This study investigated the effect that the selection of EPOM shift for cylindrical chamber has on percentage depth dose (PDD) curves. Depth dose curves were measured in a water phantom for electron beams with energies ranging from 6 to 18 MeV. The detectors investigated were of three different types: diodes (Diode-E PTW 60017 and SFD IBA), cylindrical (Semiflex PTW 31010, PinPoint PTW 31015, and A12 Exradin), and parallel plate ionization chambers (Advanced Markus PTW 34045 and Markus PTW 23343). Depth dose curves measured with Diode-E and Advanced Markus agreed within 0.2 mm at R50 except for 18 MeV and extremely large field size. The PDDs measured with the Semiflex chamber and Exradin A12 were about 1.1 mm (with respect to the Advanced Markus chamber) shallower than those measured with the other detectors using a 0.5 rcav shift. The difference between the PDDs decreased when a Pinpoint chamber, with a smaller cavity radius, was used. Agreement improved at lower energies, with the use of previously published EPOM corrections (0.3 rcav ). Therefore, the use of 0.5 rcav as an EPOM may result in a systematic shift of the therapeutic portion of the PDD (distances < R90 ). Our results suggest that a 0.1 rcav shift is more appropriate for one chamber model (Semiflex PTW 31010).

A Comparison of the Efficiency of Using a Deep CNN Approach with Other Common Regression Methods for the Prediction of EGFR Expression in Glioblastoma Patients.

To estimate epithermal growth factor receptor (EGFR) expression level in glioblastoma (GBM) patients using radiogenomic analysis of magnetic resonance images (MRI). A comparative study using a deep convolutional neural network (CNN)-based regression, deep neural network, least absolute shrinkage and selection operator (LASSO) regression, elastic net regression, and linear regression with no regularization was carried out to estimate EGFR expression of 166 GBM patients. Except for the deep CNN case, overfitting was prevented by using feature selection, and loss values for each method were compared. The loss values in the training phase for deep CNN, deep neural network, Elastic net, LASSO, and the linear regression with no regularization were 2.90, 8.69, 7.13, 14.63, and 21.76, respectively, while in the test phase, the loss values were 5.94, 10.28, 13.61, 17.32, and 24.19 respectively. These results illustrate that the efficiency of the deep CNN approach is better than that of the other methods, including Lasso regression, which is a regression method known for its advantage in high-dimension cases. A comparison between deep CNN, deep neural network, and three other common regression methods was carried out, and the efficiency of the CNN deep learning approach, in comparison with other regression models, was demonstrated.

The impact of the number of projections on image quality in Compton scatter tomography.

The availability of high resolution, energy discriminating photon counting detectors should make it possible to use Compton scattered photons to improve the diagnostic capability of computed tomography (CT). With high, spatial and energy resolution detectors Compton scatter tomography (CST) images of adequate quality can be obtained with a single projection. In practice, the limitations of realistic detectors require multiple projections for good quality images. The relationship between the number of projections used for reconstruction and the reconstructed image quality obtained for conventional CT does not necessarily apply to multi-projection Compton scatter tomography (MPCST). The purpose of this work was to investigate the dependence of the reconstructed image quality on the number of projections for MPCST. Analytical simulations and reconstructions were used to evaluate the contrast and spatial resolution for images reconstructed with one to 720 projections. Contrast-to-noise ratios (CNR) and the modulation transfer functions (MTF) demonstrated that the contrast increases monotonically with the number of projections while spatial resolution was independent of the number of projections. The contrast initially increases rapidly with projection number, becoming more gradual as the number of projections increase, with the rate of change being a function of fluence. The number of projections required to asymptotically approach the maximum contrast decreases as the fluence increases, with no indication of an optimal value for the range of fluences and projections investigated. For the projections considered, an increase in the number of projections increases the CNR even though the number of photons per projection decreases.

The effect of detector size and energy resolution on image quality in multi-projection Compton scatter tomography.

The reconstructed electron density image quality is sensitive to the detector size and energy resolution, which contribute to the blurring and noise in the image. This work evaluates optimal values of the detector parameters for a realistic system through analytical simulations of the transverse slice of the dedicated breast CT system geometry. This study introduces a spectroscopic x-ray tomography technique which uses multiple projections to reconstruct electron density images by backprojecting scattered photons over isogonic curves. The reconstruction can be obtained using a single projection yet its quality degrades as the acquisition conditions i.e. detector size and energy resolution deviate from the ideal. The reconstruction quality becomes inconsistent throughout the image due to the data under sampling caused by the finite resolution of the detector. The extension to the multi-projection mode effectively fills-in the missing data space and improves the ability to reconstruct an object. This work demonstrates the possibility to obtain images in the presence of noise.

Therapeutic targeting of TGF-ß in lung cancer.

Transforming growth factor-ß (TGF-ß) plays a complex role in lung cancer pathophysiology, initially acting as a tumor suppressor by inhibiting early-stage tumor growth. However, its role evolves in the advanced stages of the disease, where it contributes to tumor progression not by directly promoting cell proliferation but by enhancing epithelial-mesenchymal transition (EMT) and creating a conducive tumor microenvironment. While EMT is typically associated with enhanced migratory and invasive capabilities rather than proliferation per se, TGF-ß's influence on this process facilitates the complex dynamics of tumor metastasis. Additionally, TGF-ß impacts the tumor microenvironment by interacting with immune cells, a process influenced by genetic and epigenetic changes within tumor cells. This interaction highlights its role in immune evasion and chemoresistance, further complicating lung cancer therapy. This review provides a critical overview of recent findings on TGF-ß's involvement in lung cancer, its contribution to chemoresistance, and its modulation of the immune response. Despite the considerable challenges encountered in clinical trials and the development of new treatments targeting the TGF-ß pathway, this review highlights the necessity for continued, in-depth investigation into the roles of TGF-ß. A deeper comprehension of these roles may lead to novel, targeted therapies for lung cancer. Despite the intricate behavior of TGF-ß signaling in tumors and previous challenges, further research could yield innovative treatment strategies.

Real time breast microwave radar image reconstruction using circular holography: a study of experimental feasibility.

PURPOSE: The purpose of this paper is to assess the experimental feasibility of a novel breast microwave radar reconstruction approach, circular holography, using realistic experimental datasets recorded using a preclinical experimental setup. The performance of this approach was quantitatively evaluated by calculating the signal to noise ratio, contrast to noise ratio, spatial accuracy, and reconstruction time. METHODS: Six datasets were recorded, three corresponding to fatty cases and three containing synthetic dense tissue structures. Five of these datasets contained an 8 mm inclusion that emulated a malignant lesion. The data were acquired from synthetic phantoms that mimic the dielectric properties of breast tissues in the 1-6 GHz range using a custom experimental breast microwave radar system. The spatial accuracy and signal to noise ratio of the reconstructed was calculated for all the reconstructed images. The contrast to noise ratio of the reconstructed images corresponding to the datasets containing fibroglandular tissue regions was determined. This was done to evaluate the ability of the circular holographic method to provide images in which the responses from tumors can be distinguished from adjacent dense tissue structures. The execution time required to form the images was also measured to evaluate the data throughput of the holographic approach. RESULTS: For all the reconstructed datasets, the location of the synthetic tumors in the experimental setup was consistent with its position in the reconstructed image. The average spatial error was 2.2 mm, which is less than half the spatial resolution of the data acquisition system. The average signal to noise ratio of the reconstructed images containing an artificial malignant lesion was 8.5 dB, while the average contrast to noise ratio was 6.7 dB. The reconstructed images presented no artifacts. The average execution time of the images formed using the proposed approach was 5 ms, which is six orders of magnitude faster than current state of the art breast microwave radar (BMR) reconstruction algorithms. CONCLUSIONS: The results show that circular holography is capable of forming accurate images with signal to noise levels higher than 8 dB in quasi real time. Compared to BMR reconstruction algorithms tested on datasets containing dense tissue structures, the holographic approach generated images of similar spatial accuracy with higher signal to noise ratios and an acceleration factor of one order of magnitude.

A novel hybrid reconstruction algorithm for first generation incoherent scatter CT (ISCT) of large objects with potential medical imaging applications.

This work presents a first generation incoherent scatter CT (ISCT) hybrid (analytic-iterative) reconstruction algorithm for accurate ρ{e}imaging of objects with clinically relevant sizes. The algorithm reconstructs quantitative images of ρ{e} within a few iterations, avoiding the challenges of optimization based reconstruction algorithms while addressing the limitations of current analytical algorithms. A 4π detector is conceptualized in order to address the issue of directional dependency and is then replaced with a ring of detectors which detect a constant fraction of the scattered photons. The ISCT algorithm corrects for the attenuation of photons using a limited number of iterations and filtered back projection (FBP) for image reconstruction. This results in a hybrid reconstruction algorithm that was tested with sinograms generated by Monte Carlo (MC) and analytical (AN) simulations. Results show that the ISCT algorithm is weakly dependent on the ρ{e} initial estimate. Simulation results show that the proposed algorithm reconstruct ρ{e} images with a mean error of -1% ± 3% for the AN model and from -6% to -8% for the MC model. Finally, the algorithm is capable of reconstructing qualitatively good images even in the presence of multiple scatter. The proposed algorithm would be suitable for in-vivo medical imaging as long as practical limitations can be addressed.

A 1st generation scatter CT algorithm for electron density breast imaging which accounts for bound incoherent, coherent and multiple scatter: A Monte Carlo study.

Breast CT is an emerging modality that reconstructs 3D linear attenuation coefficient (µ) images of the breast. Its tomographic nature reduces the overlap of structures and may improve tissue visualization. Current prototype systems produce large levels of scatter that could be used to reconstruct electron density (ρ _{e}) images. This could potentially enhance diagnosis. We are developing a first generation bench top CT system to investigate the benefits of simultaneous imaging µ and ρ _{e} of the intact breast. The system uses an algorithm capable of reconstructing ρ _{e} images from single Klein-Nishina scatter. It has been suggested that this algorithm may be impractical since measurements include coherent, bound incoherent and multiple scatter. To investigate this, the EGSnrc Monte Carlo (MC) code was used to simulate scans using a first generation system. These simulations were used to quantify the dose per scan, to provide raw data for the ρ _{e} reconstructions and to investigate corrections for multiple and coherent scatter since these can not be directly related to ρ _{e}. MC simulations show that the dose coefficients are similar to those of cone beam breast CT. Coherent scatter is only ∼9% concentrated in scattering angles < 8°. Electron binding reduces the number of incoherently scattered photons but this reduction can be included in the quantification of scatter measured by the system. Multiple scatter was found to be the major source of errors and, if not corrected for, can result in an overestimation of ρ _{e} by more than a factor of two. Empirical corrections, based on breast thickness or radiological path, can be used to reconstruct images where the variance in ρ _{e} error is half of that found in images derived from primary photons only. Although some practical challenges remain in creating a laboratory system, this work has shown that it is possible to reconstruct scatter images of the breast with a 4 mGy dose and further experimental evaluation of this technique is warranted.

4D in vivo dose verification for real-time tumor tracking treatments using EPID dosimetry.

The use of sophisticated techniques such as gating and tracking treatments requires additional quality assurance to mitigate increased patient risks. To address this need, we have developed and validated an in vivo method of dose delivery verification for real-time aperture tracking techniques, using an electronic portal imaging device (EPID)-based, on-treatment patient dose reconstruction and a dynamic anthropomorphic phantom. Using 4DCT scan of the phantom, ten individual treatment plans were created, 1 for each of the 10 separate phases of the respiratory cycle. The 10 MLC apertures were combined into a single dynamic intensity-modulated radiation therapy (IMRT) plan that tracked the tumor motion. The tumor motion and linac delivery were synchronized using an RPM system (Varian Medical Systems) in gating mode with a custom breathing trace. On-treatment EPID frames were captured using a data-acquisition computer with a dedicated frame-grabber. Our in-house EPID-based in vivo dose reconstruction model was modified to reconstruct the 4D accumulated dose distribution for a dynamic MLC (DMLC) tracking plan using the 10-phase 4DCT dataset. Dose estimation accuracy was assessed for the DMLC tracking plan and a single-phase (50% phase) static tumor plan, represented a static field test to verify baseline accuracy. The 3%/3 mm chi-comparison between the EPID-based dose reconstruction for the static tumor delivery and the TPS dose calculation for the static plan resulted in 100% pass rate for planning target volume (PTV) voxels while the mean percentage dose difference was 0.6%. Comparing the EPID-based dose reconstruction for the DMLC tracking to the TPS calculation for the static plan gave a 3%/3 mm chi pass rate of 99.3% for PTV voxels and a mean percentage dose difference of 1.1%. While further work is required to assess the accuracy of this approach in more clinically relevant situations, we have established clinical feasibility and baseline accuracy of using the transmission EPID-based, in vivo patient dose verification for MLC-tracking treatments.

The Application of an Iterative Structure to the Delay-and-Sum and the Delay-Multiply-and-Sum Beamformers in Breast Microwave Imaging.

Breast microwave imaging (BMI) is a potential breast cancer screening method. This manuscript presents a novel iterative delay-and-sum (DAS) based reconstruction algorithm for BMI. This iterative-DAS (itDAS) algorithm uses a forward radar model to iteratively update an image estimate. A variation of the itDAS reconstruction algorithm that uses the delay-multiply-and-sum (DMAS) beamformer was also implemented (the itDMAS algorithm). Both algorithms were used to reconstruct images from experimental scans of an array of 3D-printed MRI-based breast phantoms performed with a clinical BMI system. The signal-to-clutter ratio (SCR) and signal-to-mean ratio (SMR) were used to compare the performance of the itDAS and itDMAS methods to the DAS and DMAS beamformers. While no significant difference between the itDAS and itDMAS methods was observed in most images, the itDAS algorithm produced reconstructions that had significantly higher SMR than the non-iterative methods, increasing contrast by as much as 19 dB over DAS and 13 dB over DMAS. The itDAS algorithm also increased the SCR of reconstructions by up to 5 dB over DAS and 4 dB over DMAS, indicating that both high-intensity and background clutter are reduced in images reconstructed by the itDAS algorithm.

Classification of breast tumor models with a prototype microwave imaging system.

PURPOSE: The assessment of the size and shape of breast tumors is of utter importance to the correct diagnosis and staging of breast cancer. In this paper, we classify breast tumor models of varying sizes and shapes using signals collected with a monostatic ultra-wideband radar microwave imaging prototype system with machine learning algorithms specifically tailored to the collected data. METHODS: A database comprising 13 benign and 13 malignant tumor models with sizes between 13 and 40 mm was created using dielectrically representative tissue mimicking materials. These tumor models were placed inside two breast phantoms: a homogeneous breast phantom and a breast phantom with clusters of fibroglandular mimicking tissue, accounting for breast heterogeneity. The breast phantoms with tumors were imaged with a monostatic microwave imaging prototype system, over a 1-6 GHz frequency range. The classification of benign and malignant tumors embedded in the two breast phantoms was completed, and tumor classification was evaluated with Principal Component Analysis as a feature extraction method, and tuned Naïve Bayes (NB), decision trees (DT), and k-nearest neighbours (kNN) as classifiers. We further study which antenna positions are better placed to classify tumors, discuss the feature extraction method and optimize classification algorithms, by tuning their hyperparameters, to improve sensitivity, specificity and the receiver operating characteristic curve, while ensuring maximum generalization and avoiding overfitting and data contamination. We also added a realistic synthetic skin response to the collected signals and examined its global effect on classification of benign vs malignant tumors. RESULTS: In terms of global classification performance, kNN outperformed DT and NB machine learning classifiers, achieving a classification accuracy of 96.2% when classifying between benign and malignant tumor phantoms in a homogeneous breast phantom (both when the skin artifact is and is not considered). CONCLUSIONS: We experimentally classified tumor models as benign or malignant with a microwave imaging system, and we showed a methodology that can potentially assess the shape of breast tumors, which will give further insight into the correct diagnosis and staging of breast cancer.

An Air-Operated Bistatic System for Breast Microwave Radar Imaging: Pre-Clinical Validation.

This study presents a breast microwave radar imaging system designed for bistatic operation in air. The system operates in the range of 1-8 GHz and employs a double-ridged horn antenna array with four degrees of freedom. A hemispherical breast phantom with a tumor inclusion of 1.5 cm diameter was used for validation. Reconstructed datasets resulted in artifact-free images where the tumor presence was detected. Signal to clutter ratios greater than 30 dB and tumor to fibroglandular ratios of 3 dB were measured. The encouraging images obtained with the system validate its potential as a clinical tool for breast cancer detection.

Evaluating a potential technique with local optical flow vectors for automatic organ-at-risk (OAR) intrusion detection and avoidance during radiotherapy.

Various techniques of deep inspiration breath hold (DIBH) have been used to mitigate the likelihood and risk of exposing the heart, an organ-at-risk (OAR) for unintended radiation during left breast radiotherapy. However, issues of reproducibility of these techniques warrant further investigation into the feasibility of detecting the intrusion of an OAR into the treatment field during intra-fractional treatment delivery. The increase of high-dose, low-fraction radiotherapy treatments makes it important to immediately adapt treatment once an OAR is detected in the treatment field. This proof-of-concept implementation includes an algorithm that detects and tracks the motion at the edges of a treatment field and a control algorithm that adapts the treatment aperture according to the motion detected. In accordance to the AAPM Task-Group (TG-132) report, image registration techniques should be verified with virtual and physical phantoms prior to clinical application. Since most OARs move as a result of respiration-induced motion, we have used a lung phantom to generate images of a generic OAR intruding into a treatment field with known velocity. The phantom was programmed to move with sinusoidal and lung patient tumor motion patterns and the accuracy of intrusion tracking and MLC adaptation were benchmarked with the ground truth-programmed motion of the OAR. The motions were recorded with an electronic portal imaging device (EPID). An optimal cluster size of 9 × 9 motion vectors was found to provide the smallest average absolute position error of 0.3 mm. A strong linear correlation between the adapted MLC leaves and the actual OAR position was observed. The algorithm had a mean position tracking error of -0.4 ± 0.3 mm and a precision of 1.1 mm. It is possible to adapt MLC leaves based on the motion detected at the edges of the irradiated field, and it would be feasible to shield an unplanned intrusion of an OAR into the treatment field.

Reducing the tracking drift of an uncontoured tumor for a portal-image-based dynamically adapted conformal radiotherapy treatment.

Accurate tracking of organ motion during treatment is needed to improve the efficacy of radiation therapy. This work investigates the feasibility of tracking an uncontoured target using the motion detected within a moving treatment aperture. Tracking was achieved with a weighted optical flow algorithm, and three different techniques for updating the reference image were evaluated. The accuracy and susceptibility of each approach to the accumulation of position errors were verified using a 3D-printed tumor (mounted on an actuator) and a virtual treatment aperture. Tumor motion up to 15.8 mm (peak-to-peak) taken from the breathing patterns of seven lung cancer patients was acquired using an amorphous silicon portal imager at ~ 7.5 frames/s. The first approach (INI) used the initial image detected, as a fixed reference, to determine the target motion for each new incoming image, and performed the best with the smallest errors. This method was also the most robust against the accumulation of position errors. Mean absolute errors of 0.16, 0.32, and 0.38 mm were obtained for the three methods, respectively. Although the errors are comparable to other tracking methods, the proposed method does not require prior knowledge of the tumor shape and does not need a tumor template or contour for tracking. Graphical abstract.

A method for 3D electron density imaging using single scattered x-rays with application to mammographic screening.

Screening mammography is the current standard in detecting breast cancer. However, its fundamental disadvantage is that it projects a 3D object into a 2D image. Small lesions are difficult to detect when superimposed over layers of normal, heterogeneous tissue. In this work, we examine the potential of single scattered photon electron density imaging in a mammographic environment. Simulating a low-energy (<20 keV) scanning pencil beam, we have developed an algorithm capable of producing 3D electron density images from a single projection. We have tested the algorithm by imaging parts of a simulated mammographic accreditation phantom containing lesions of various sizes. The results indicate that the group of imaged lesions differ significantly from background breast tissue (p<0.005), confirming that electron density imaging may be a useful diagnostic test for the presence of breast cancer.

A wavefront reconstruction method for 3-D cylindrical subsurface radar imaging.

In recent years, the use of radar technology has been proposed in a wide range of subsurface imaging applications. Traditionally, linear scan trajectories are used to acquire data in most subsurface radar applications. However, novel applications, such as breast microwave imaging and wood inspection, require the use of nonlinear scan trajectories in order to adjust to the geometry of the scanned area. This paper proposes a novel reconstruction algorithm for subsurface radar data acquired along cylindrical scan trajectories. The spectrum of the collected data is processed in order to locate the spatial origin of the target reflections and remove the spreading of the target reflections which results from the different signal travel times along the scan trajectory. The proposed algorithm was successfully tested using experimental data collected from phantoms that mimic high contrast subsurface radar scenarios, yielding promising results. Practical considerations such as spatial resolution and sampling constraints are discussed and illustrated as well.

Feasibility of predicting tumor motion using online data acquired during treatment and a generalized neural network optimized with offline patient tumor trajectories.

PURPOSE: The accurate prediction of intrafraction lung tumor motion is required to compensate for system latency in image-guided adaptive radiotherapy systems. The goal of this study was to identify an optimal prediction model that has a short learning period so that prediction and adaptation can commence soon after treatment begins, and requires minimal reoptimization for individual patients. Specifically, the feasibility of predicting tumor position using a combination of a generalized (i.e., averaged) neural network, optimized using historical patient data (i.e., tumor trajectories) obtained offline, coupled with the use of real-time online tumor positions (obtained during treatment delivery) was examined. METHODS: A 3-layer perceptron neural network was implemented to predict tumor motion for a prediction horizon of 650 ms. A backpropagation algorithm and batch gradient descent approach were used to train the model. Twenty-seven 1-min lung tumor motion samples (selected from a CyberKnife patient dataset) were sampled at a rate of 7.5 Hz (0.133 s) to emulate the frame rate of an electronic portal imaging device (EPID). A sliding temporal window was used to sample the data for learning. The sliding window length was set to be equivalent to the first breathing cycle detected from each trajectory. Performing a parametric sweep, an averaged error surface of mean square errors (MSE) was obtained from the prediction responses of seven trajectories used for the training of the model (Group 1). An optimal input data size and number of hidden neurons were selected to represent the generalized model. To evaluate the prediction performance of the generalized model on unseen data, twenty tumor traces (Group 2) that were not involved in the training of the model were used for the leave-one-out cross-validation purposes. RESULTS: An input data size of 35 samples (4.6 s) and 20 hidden neurons were selected for the generalized neural network. An average sliding window length of 28 data samples was used. The average initial learning period prior to the availability of the first predicted tumor position was 8.53 ± 1.03 s. Average mean absolute error (MAE) of 0.59 ± 0.13 mm and 0.56 ± 0.18 mm were obtained from Groups 1 and 2, respectively, giving an overall MAE of 0.57 ± 0.17 mm. Average root-mean-square-error (RMSE) of 0.67 ± 0.36 for all the traces (0.76 ± 0.34 mm, Group 1 and 0.63 ± 0.36 mm, Group 2), is comparable to previously published results. Prediction errors are mainly due to the irregular periodicities between cycles. Since the errors from Groups 1 and 2 are within the same range, it demonstrates that this model can generalize and predict on unseen data. CONCLUSIONS: This is a first attempt to use an averaged MSE error surface (obtained from the prediction of different patients' tumor trajectories) to determine the parameters of a generalized neural network. This network could be deployed as a plug-and-play predictor for tumor trajectory during treatment delivery, eliminating the need for optimizing individual networks with pretreatment patient data.