Non-stop gated cone-beam computer tomography imaging

Abstract

Presented herein is an imaging technique named nonstop gated cone-beam computer tomograph (CBCT), for guiding free-breathing respiratory gating radiotherapy of thoracic and abdominal cancer patients (e.g., lung, liver, pancreas). The nonstop gated CBCT technique reduces the data acquisition time to 1 minute and decreases the imaging dose by >40% while simultaneously retaining high quality images was developed. This is done by allowing the gantry to rotate continuously and having the kV x-ray beam on when the patient breathing signal is within the respiratory gate and the x-ray beam off when it is out of the respiratory gate. A machine learning model can be used to reconstruct high-quality CBCT images from the acquired data.

Description

[0001]Atty. Dkt. No.: 115872-3198 NON-STOP GATED CONE-BEAM COMPUTER TOMOGRAPHY IMAGING CROSS REFERENCE TO RELATED APPLICATIONS This application claims priority to U.S. Provisional Patent Application No. 63 / 561,161, titled “Non-Stop Gated Cone-Beam Computer Tomography Imaging,” filed March 4, 2024, which is incorporated by reference in its entirety. BACKGROUND An imaging device can acquire image data of an object and provide the image data for additional processing to a computing device. SUMMARY Aspects of the present disclosure are directed to systems, methods, devices, and non-transitory computer readable media for reconstructing images using projection data from cone beam computed tomography (CBCT). One or more processors coupled with memory may receive a projection dataset of at least a portion of a volume of a subject, the projection dataset generated via a gantry of a CBCT scanner rotating along an arc at least partially about the volume during a time window. The volume of the subject may switch between a first state and a second state during the time window. The projection dataset may include: (i) a plurality of projections corresponding to a first plurality of segments of the arc, with each of the first plurality of segment corresponding to when the volume of the subject is in the first state, and (ii) a plurality of gaps corresponding to a second plurality of segments of the arc, with each of the second plurality of segment corresponding to when the volume of the subject is in the second state. The one or more processors may apply a machine learning (ML) architecture on the projection dataset. The ML architecture may be established using training data comprising a plurality of examples. Each of the plurality of examples may have: (i) a sample projection dataset comprising a sample plurality of projections corresponding to at least a portion of a sample arc about a sample volume of a sample subject over a sample time window and (ii) a sample CT image corresponding to the sample volume of the sample subject. The one or more processors may generate, based on -1- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 applying the ML architecture on the projection dataset, a CT image corresponding to the volume of the subject. The one or more processors may store, using one or more data structures, an association between the subject and the CT image. In some embodiments, the one or more processors may provide, for presentation via a user interface, an output based on the CT image corresponding to the volume of the subject. In some embodiments, the one or more processors may receive, via the user interface, an indication that a target within the volume corresponding to the CT image as matching a location identified in a therapy plan. The one or more processors may provide, responsive to the indication, an instruction to administer radiotherapy to the target in accordance with the therapy plan. In some embodiments, the target within the volume of the subject may be administered with the radiotherapy. The radiotherapy may include at least one of a stereotactic body radiation therapy (SBRT), an intensity-modulated radiation therapy (IMRT), a volumetric modulated arc therapy (VMAT), a conformal radiation therapy (CRT), or a proton beam therapy (PBT). In some embodiments, the one or more processors may receive, via the user interface, an indication that a target within the volume corresponding to the CT image as not matching a location identified in a therapy plan. The one or more processors may refrain, responsive to the indication, from providing an instruction to administer radiotherapy to the target. In some embodiments, the one or more processors may generate, using the projection dataset in accordance with an interpolator, a second projection dataset comprising (i) the plurality of projections corresponding to the first plurality of segments and (ii) a second plurality of projections corresponding to one or more of the second plurality of segments. The one or more processors may apply the ML architecture to the second projection dataset to generate the CT image. In some embodiments, the ML architecture may include: a projection synthesizer configured to generate, using the projection dataset, a second projection dataset lacking one or more of the plurality of gaps; a domain converter configured to generate, from the second projection dataset, an initial CT image corresponding to the volume of the -2- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 subject; and an image reconstructor configured to generate, using the initial CT image, the CT image corresponding to the volume of the subject. In some embodiments, the ML architecture may include: an encoder configured to generate, using the projection dataset, a plurality of embeddings corresponding to features; and a decoder configured to generate, using the plurality of embeddings, the CT image corresponding to the volume of the subject. In some embodiments, the sample projection dataset in at least one of the plurality of examples may include at least one of: (i) a non-stop, gated sample projection dataset comprising: (a) the sample plurality of projections corresponding to a first sample plurality of segments of the sample arc, with each of the first sample plurality of segment corresponding to when the sample volume of the sample subject is in the first state, and (b) a sample plurality of gaps corresponding to a sample second plurality of segments of the sample arc, with each of the second plurality of segment corresponding to when the volume of the subject is in the second state, or (ii) a gated sample projection dataset comprising sample plurality of projections corresponding to an entirety of the sample arc during the sample time window. In some embodiments, the subject may be at risk of or diagnosed with cancer associated with an organ within the volume. The first state may correspond to the volume of the subject being within a gating window and the second state may correspond to the volume of the subject being outside the gating window. The cancer may include at least one of lung cancer, esophageal cancer, stomach cancer, colorectal cancer, liver cancer, pancreatic cancer, small intestine cancer, kidney cancer, prostate cancer, testicular cancer, ovarian cancer, uterine cancer, or cervical cancer. Aspects of the present disclosure are directed to systems, methods, devices, and non-transitory computer readable media for performing cone beam computed tomography (CBCT). One or more processors coupled with memory may cause a CBCT scanner to rotate a gantry along an arc at least partially about a volume of a subject during a time window, the subject arranged along a surface of an examination table. The one or more processors may identify, during the time window, a motion of the volume of the subject relative to the surface of the examination table. The one or more processors may determine, based on the motion of the volume, the volume of the subject as in a first state -3- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 instead of a second state. The one or more processors may cause, responsive to determining the volume as in the first state, a beam emitter of the CBCT scanner to be activated to produce a radiation beam while the gantry is rotating along a segment of the arc. The one or more processors may acquire, via a detector of the CBCT scanner, a plurality of projections corresponding to the radiation beam traversing through the volume of the subject via the segment. The one or more processors may generate a projection dataset comprising the plurality of projections corresponding to the segment of the arc about the volume of the subject. The one or more processors may store, using one or more data structures, an association between the subject and the projection dataset. In some embodiments, the one or more processors may determine, based on the motion of the volume, the volume of the subject as in the second state instead of the first state. The one or more processors may cause, responsive to determining the volume as in the second state, the beam emitter of the CBCT scanner to be deactivated to cease production of the radiation beam while the gantry is rotating along a second segment of the arc. The one or more processors may continue to cause the CBCT scanner to rotate the gantry along the second segment of the arc about the volume of a subject. The one or more processors may generate the projection dataset to include a gap corresponding to the second segment of the arc. In some embodiments, the one or more processors may determine, based on the motion of the volume, the volume of the subject as switching back to the first state from the second state. The one or more processors may cause, responsive to determining the volume as switching back to the first state, the beam emitter of the CBCT scanner to be activated to produce the radiation beam while the gantry is rotating along a third segment of the arc. The one or more processors may acquire, via the detector of the CBCT scanner, a second plurality of projections corresponding to the radiation beam traversing through the volume of the subject via the third segment. The one or more processors may generate the projection dataset to include the first plurality of projections corresponding to the first segment, the gap corresponding to the second segment, and the third plurality of projections corresponding to the third segment. -4- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 In some embodiments, the one or more processors may provide the projection dataset to a machine learning (ML) architecture. The ML architecture may be established using training data comprising a plurality of examples. Each of the plurality of examples may have: (i) a sample projection dataset comprising a sample plurality of projections corresponding to at least a portion of a sample arc about a sample volume of a sample subject over a sample time window and (ii) a sample CT image corresponding to the sample volume of the sample subject. The one or more processors may obtain, based on applying the ML architecture, a CT image corresponding to the volume of the subject. In some embodiments, the subject may be administered with radiotherapy to a target within the volume in accordance with the therapy plan. The radiotherapy may include at least one of a stereotactic body radiation therapy (SBRT), an intensity-modulated radiation therapy (IMRT), a volumetric modulated arc therapy (VMAT), a conformal radiation therapy (CRT), or a proton beam therapy (PBT). In some embodiments, the subject may be at risk of or diagnosed with cancer associated with an organ within the volume. The first state may correspond to the volume of the subject being within a gating window and the second state may correspond to the volume of the subject being outside the gating window. The cancer may include at least one of lung cancer, esophageal cancer, stomach cancer, colorectal cancer, liver cancer, pancreatic cancer, small intestine cancer, kidney cancer, prostate cancer, testicular cancer, ovarian cancer, uterine cancer, or cervical cancer In some embodiments, the one or more processors may generate, using the motion of the volume of the subject prior to the time window, a threshold value for the motion to define the subject as switching between the first state and the second state. The one or more processors may determine the volume as in the first state instead of the second state based on a comparison of the motion with the threshold value. In some embodiments, the one or more processors may identify the motion of the volume relative to the surface of the examination table based on at least one of (i) image data of the subject on the examination table or (ii) a position of a structure situated on a side of the subject distal from the surface. -5- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 In some embodiments, the one or more processors may cause the beam emitter of the CBCT scanner to produce the radiation beam in accordance with one of a full- fan configuration or a half-fan configuration. The full-fan configuration may correspond to the radiation beam being fully collimated with the detector. The half-fan configuration may correspond to the radiation beam being partially collimated with the detector. In some embodiments, the one or more processors may cause the CBCT scanner to rotate the gantry further comprises causing the CBCT scanner to rotate the gantry along the arc without interruption during the time window, wherein the arc along which the gantry rotates ranges between 140° to 360°. BRIEF DESCRIPTION OF THE DRAWINGS The foregoing and other objects, aspects, features, and advantages of the disclosure will become more apparent and better understood by referring to the following description taken in conjunction with the accompanying drawings, in which: FIG.1: A diagram of the gated CBCT setup. The left part shows a patient lying on the treatment bed with the kV X-ray beam (green) turned on. The external breath monitor is placed on the patient’s chest. The breathing pattern and the gating window (yellow band) are displayed on the monitor. FIG.2: The free-breathing 3D CBCT (left), gated CBCT (middle), and the proposed nonstop gated CBCT (right) modes with acquisition time and the CT dose index. ngCBCT acquires non-uniform and under-sampled projections only in 1 minute and with reduced imaging dose. *FB is short for free-breathing. The acquisition times and the CT dose index of each mode are also listed in the table. The proposed nonstop gated CBCT mode acquires non-uniform and under-sampled projections but with an acquisition time equal to the free-breathing CBCT (1 minute) and with only a fraction of the imaging dose (30-60%) to the patient. FIG.3: Gated CBCT (1stcolumn), nonstop gated CBCT with FDK (2nd), IR (3rd), and PIBR (4th) reconstructed images from the HF scan (projection distribution illustrated in the cartoon on the left) in transverse (top), coronal (middle), and sagittal (bottom) views. The blue arrows point to the lung tumors. The prior images used for PIBR -6- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 (5thcolumn) are included with the red arrows point anatomical differences. The PSNR and UQI for each reconstruction are labeled in the images. The scan time for nonstop gated CBCT is 1 minute compared to the 2.4 minutes for gated CBCT. In all scenarios, FDK reconstruction suffers from severe streak artifacts. The PIBR method demonstrates obvious superiority. FIG.4: Gated CBCT (1stcolumn), nonstop gated CBCT with FDK (2nd), IR (3rd), and PIBR (4th) reconstructed images from the FF scan in transverse (top), coronal (middle), and sagittal (bottom) views. The blue arrows point to the lung tumors. The prior images used for PIBR (5thcolumn) are included with the red arrows point anatomical differences. The PSNR and UQI for each reconstruction are labeled in the images. The PIBR method demonstrates significant improvement over FDK or IR. FIG.5: Clinical example: lung SBRT FIG.6: Nonstop Gated CBCT: projection data FIG.7: Nonstop vs. Sparse-view FIGs.8 and 9: Reconstruction of images from gated CBCT and nonstop Gated CBCT, Half-fan configuration. FIG.10: Nonstop Gated CBCT: half-fan vs full-fan FIG.11: Nonstop vs. Sparse-view FIGs.12 and 13: Reconstruction of images from gated CBCT and nonstop Gated CBCT, Full-fan configuration FIG.14: Nonstop Gated CBCT: Full-fan (lower duty cycle) FIG.15: Nonstop Gated CBCT: Full-fan and limited angle FIG.16: Nonstop gated CBCT projections (non-uniform and under- sampled) are down-sampled from clinical gated CBCT projections based on each patient’s actual breathing cycles. -7- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 FIG.17: Architecture of the proposed IResNet. It is based on U-Net with additional skip connections and ResNet blocks to improve the network’s performance. FIG.18: Comparisons between the gated CBCT and the reconstructed nonstop gated CBCT images by FDK, IR, FBPCONVNet, and IResNet. Both HF (top) and FF (bottom) scans are studied. The blue arrows indicate the tumors. Hallucinations occur frequently with FBPCONVNet and are pointed out by the red arrows. The proposed IResNet results in better image quality and detail preservation (green arrows) than traditional reconstruction methods and other deep convolutional networks. FIG.19: Comparisons between the gated CBCT, the various reconstructions, and IResNet. The sagittal (top) and coronal (bottom) views of the HF scans are obtained by stacking the transverse view slices. The blue arrows indicate the tumors. While the other reconstructions suffer from artifacts (streaks, fringes, and bands) and hallucinations (red arrows), the proposed IResNet results in better image quality and detail preservation (green arrows). FIG.20: Comparisons between the gated CBCT, the various reconstructions, and IResNet. The sagittal (top) and coronal (bottom) views of the FF scans are obtained by stacking the transverse view slices. The blue arrows indicate the tumors. Severe streak artifacts are seen in all the reconstructions (red arrows) except the proposed IResNet, which also results in better image quality and detail preservation (green arrows). However, the shape and intensity of the tumor changed in all methods; small-size tumors are currently a limitation factor. FIG.21: Conventional reconstruction methods. FDK and iterative reconstruction suffer from severe artifacts due to the non-uniform and under-sampled projections. FIG.22: IResNet result comparisons. Arrows in column (d): hallucinations generated by FBPCONVNet. Arrows in column (e): improved textures and bony details with IResNet. -8- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 FIG.23: IResNet result comparisons. Note: the network is training in 2D with transverse plane slices. Arrows under IR and FBPCONNet columns: fringe artifacts with IR, and hallucinations (also horizontal bands) with FBPCONVNet. Arrow on right: Fringe and band artifacts removed, improved textures and bony details. FIG.24: (a) Setup for weighted CBDI measurement. (b) CTDI body phantom and two pencil chambers for simultaneous central and peripheral dose measurements. (c) Dose readers for two pencil chambers. (d) CIRS surrogate motion platform accompanied by Varian RGSC gating block (with infrared markers). FIG.25: Representative surrogate motion patterns and amplitude-based gating windows (upper / lower thresholds are labeled). (a) CIRS motion control software generated ^^^^^^^^^^^^4 motion trace. (b) One clinical patient breathing trace. FIG.26: Scan time and weighted CBDI vs. duty cycle for ^^^^^^^^^^^^4 motion patterns (four different cycle periods). FIG.27: Scan time and weighted CBDI vs. duty cycle for three patient breathing traces. For each patient, one Gcbct and two ngCBCT scans (all using the same gating window) were acquired. It’s not applicable to achieve the same gating duty cycles due to irregular periodicity in patient breathing traces. FIG.28: Dose rate profiles for (a) the central and (b) peripheral positions of an ngCBCT scan using a patient breathing trace. For the ngCBCT scan, the x-ray is activated only when the breathing signal falls within the defined gating window, resulting in gaps in the profiles above. The dose rate profile for the peripheral position shows significant variation due to the rotating x-ray source positions and limited field size of half- fan imaging geometry. FIG.29: The ngCBCT projections of patient A are emulated from the clinical gCBCT projections. The ngCBCT projections were down sampled based on patient specific breathing cycles in Figure 1(a). -9- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 FIG.30: The kV source distributions for the half-fan (a) and full-fan (b) are illustrated with green dots. The detector (orange) was offset and rotated 360º around the isocenter (pink hexagram) with the HF mode. The detector was not offset and rotated 200º for the FF mode. The projections in both scans are under-sampled and non-uniform. FIG.31: For patient A, the clinical gCBCT (1stcolumn) in transverse (top row), coronal (middle), and sagittal (bottom) views are compared with the much faster ngCBCT acquisition reconstructed with the FDK (2ndcolumn), IR (3rdcolumn), and the PIBR (4thcolumn) methods. The PIBR method results in much improved image quality compared to the FDK and IR methods. The PSNR and UQI for the FDK, IR, and PIBR methods are listed in Table 2. The tumor is pointed out by arrows in the gCBCT images. The patient specific prior images used for PIBR (5thcolumn) show some anatomical difference to the clinical gCBCT images. These differences have minimal impact on the PIBR reconstruction. FIG.32: For patient B, the clinical gCBCT (1stcolumn) in transverse (top row), coronal (middle), and sagittal (bottom) views are compared with the much faster ngCBCT acquisition reconstructed with the FDK (2ndcolumn), IR (3rdcolumn), and the PIBR (4thcolumn) methods. The PIBR method results in much improved image quality compared to the FDK and IR methods. The PSNR and UQI for the FDK, IR, and PIBR methods are listed in Table 2. The tumor is pointed out by arrows in the gCBCT images. The patient specific prior images used for PIBR (5thcolumn) show some anatomical difference to the clinical gCBCT images. These differences have minimal impact on the PIBR reconstruction. FIG.33: The half fan ngCBCT scan emulated with a lower duty cycle than used for Figure 3 (from 53% to 32%). The FDK and IR image quality are worse due to the lower duty cycle. The PIBR method can improve image quality. The tumor locations are pointed out by arrows in the PIBR images. FIG.34: The full fan ngCBCT scan emulated with a lower duty cycle than used for Figure 4 (from 49% to 34%). The FDK and IR image quality are worse due to the -10- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 lower duty cycle. The PIBR method can improve image quality. The tumor locations are pointed out by arrows in the PIBR images. FIG.35: Sparse view CBCT simulations for both HF and FF scans. The FDK reconstruction with roughly 50% projections (a and d), 25% projections (b and e), and 12.5% projections (c and f) are shown. For sparse view CBCT reconstructions, even at only 12.5% projections, the image quality is considerably better than ngCBCT reconstruction (the under sampled but non uniform projections) with the same FDK method. FIG.36: The acquisition time and CTDI of the free-breathing CBCT methods: 3D, gated, and the proposed nonstop gated. The nonstop gated method is faster than the gated method and has lower CTDI. The scan geometry and projection data (green and red dots) are shown – nonstop gated projections are under-sampled and non-uniform. FIG.37: A dual-domain reconstruction approach using two CNNs. The projection domain CNN completes the projection data. The image domain CNN improves reconstruction image quality. The FDK algorithm connects the two domains during training. FIG.38: Reconstruction comparisons between gated CBCT (1stcolumn), and the nonstop gated CBCT projections reconstructed with FDK (2ndcolumn), the IResNet was developed previously (3rdcolumn), and the dual-domain reconstruction network (4thcolumn). Half-fan (top row) and full-fan (bottom row) scan modes are studied. The blue arrows point out the tumors. Artifacts that were hard for the IResNet to handle, even artifacts in the gated CBCT (pointed out by red arrows), can now be better handled by the dual-domain approach (green arrows). FIG.39: The dual-domain CNN handles small-size tumors better than the IResNet. For one half-fan scan in the sagittal view, the tumor in gated CBCT (a) is labeled with a blue arrow. The tumor is faint in the IResNet result (c). The tumor shape and intensity with the DDCNN (d) are similar to the gated CBCT. FIG.40: The dual-domain CNN (d) handles streak artifacts caused by high- density structures better than the IResNet (c). -11- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 FIG.41: As shown, dual-domain CNN is efficient in removing streak artifacts. It improves some artifacts in the ground truth gated CBCT. FIG.42: Nonstop Gated CBCT: cos4 motion. Periods of 3s,4s,5s,6s; gating duty cycles of 30%, 40%, 50%, 60% FIG.43: Nonstop Gated CBCT: patient breathing traces. For each patient, one gated CBCT and two nonstop CBCT scans (all using the same gating window) were acquired. It’s not applicable to achieve the same gating duty cycles due to irregular periodicity in patient breathing traces. FIGs.44–46: Nonstop Gated CBCT: reconstructions of various scans. FIG.47 depicts a block diagram of a system for processing projection data acquired via non-stop gated CBCT, in accordance with an illustrative embodiment. FIG.48 depicts a block diagram of a process for acquiring projection data acquired via non-stop gated CBCT, in accordance with an illustrative embodiment. FIG.49 depicts a block diagram of a process for training a machine learning (ML) architectures for reconstructing CT images, in accordance with an illustrative embodiment. FIG.50 depicts a block diagram of a process for reconstructing CT images using ML architectures, in accordance with an illustrative embodiment. FIG.51 depicts a block diagram of a process for providing outputs based on the reconstructed CT images, in accordance with an illustrative embodiment. FIG.52 depicts a flow diagram of a method of performing CBCT to acquire projection data, in accordance with an illustrative embodiment. FIG.53 depicts a flow diagram of a method of reconstructing images using projection data from CBCT, in accordance with an illustrative embodiment. -12- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 FIG.54 is a block diagram of a computing environment according to an example implementation of the present disclosure. DETAILED DESCRIPTION Following below are more detailed descriptions of various concepts related to, and embodiments of, systems and methods for non-stop gated cone-beam computer tomography (CBCT) imaging. It should be appreciated that various concepts introduced above and discussed in greater detail below may be implemented in any of numerous ways, as the disclosed concepts are not limited to any particular manner of implementation. Examples of specific implementations and applications are provided primarily for illustrative purposes. Section A describes next-generation nonstop gated CBCT imaging technique. Section B describes deep-learning empowered fast-gated CBCT imaging for respiratory gating radiotherapy. Section C describes scan efficiency and imaging dose analysis of next- generation nonstop gated CBCT for respiratory gating lung radiotherapy. Section D describes nonstop gated CBCT for respiratory gating lung stereotactic body radiation therapy (SBRT) with image reconstruction using prior-image- based iterative reconstruction Section E describes application in pretreatment setup verification of respiratory gating lung SBRT. Section F describes systems and methods for performing nonstop gated cone beam computed tomography (CBCT) imaging and reconstructing images using projection data from nonstop gated CBCT imaging. -13- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 Section G describes a network environment and computing environment which may be useful for practicing various computing related embodiments described herein. A. Next-Generation Nonstop Gated CBCT Imaging Technique Presented herein is an imaging technique named “nonstop gated cone-beam computer tomograph (CBCT)”, for guiding free-breathing respiratory gating radiotherapy of thoracic and abdominal cancer patients (e.g., lung, liver, pancreas). Free-breathing gated CBCT depicts an anatomy as it may appear in the subsequent gated treatment, thus permitting more precise tumor localization and reduced target margins. But the clinical gated CBCT scan takes 2–7 mins because the gantry is interrupted and resumed by the respiratory gate multiple times over the scan. The long data acquisition time extends the overall treatment time and leads to increased patient discomfort and likelihood of patient movement and breathing pattern drift. Additionally, in clinical gated CBCT acquisition, a substantial fraction of the kV projections is acquired while the gantry is accelerating, resulting in higher imaging dose. Thus, the “nonstop gated CBCT” technique that reduces the data acquisition time to 1 minute and decreases the imaging dose by >40% while simultaneously retaining high quality images was developed. This is done by allowing the gantry to rotate continuously and having the kV x-ray beam on when the patient breathing signal is within the respiratory gate and the x-ray beam off when it is out of the respiratory gate. A prior-image-based iterative reconstruction framework is utilized to reconstruct high-quality CBCT images from the acquired data. Respiratory motion management is critical for thoracic and abdominal cancer radiotherapy patients. Managing tumor motion reduces margin around target and supports dose escalation, thereby improving local tumor control and decreasing the normal tissue complication probability. Compared to deep inspiration breath hold (DIBH), free-breathing respiratory gating is less demanding and thus more applicable for patients. In respiratory gating radiotherapy, a device external to the patient monitors breathing and allows delivery of radiation only during preset gating window of the breathing cycle. The same motion management strategy is applied for treatment and pre-treatment CBCT imaging. -14- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 In the current implementation of free-breathing gated CBCT, the gantry movement is interrupted and resumed by the respiratory gate and thus takes 2-7 minutes to acquire the projection data, depending on the duty cycle and patient breathing period. The long data acquisition time extends the overall on-table time and leads to increased patient discomfort and likelihood of patient movement and breathing pattern drift. Thus, it affects not only patient experience but also treatment outcome. Additionally, in gated CBCT acquisition, a substantial fraction of the kV projections is acquired while the gantry is accelerating, resulting in higher imaging dose than standard CBCT. Generally, the imaging dose increases with the number of required gantry accelerations (equal to the number of beam-on windows) during the scan. Thus, more widespread use of gated CBCT is hampered by these limitations. The “nonstop gated CBCT” technique solves the major drawbacks of the current clinical implementation of gated CBCT by substantially reducing the scan time and imaging dose. In the “nonstop gated CBCT”, the gantry rotates continuously at the maximum permitted speed (e.g., 6º / second or 1 revolution / minute) irrespective of the patient’s breathing signal or preset gating window. Also, the kV X-ray is only turned on when the patient’s breathing signal is within the respiratory gate (FIG.1), thus substantially decreasing the imaging dose to the patient (FIG.2). The prior-image-based iterative reconstruction (PIBR) framework can reconstruct high-quality images from the non-uniform and under-sampled projections (FIG.2) resulting from this data acquisition strategy. In summary, this technique has great potential to improve both patient experience and treatment outcome by reducing overall patient on-table time and alleviating patient discomfort throughout the treatment. This technique may expand the patient population that can benefit from the use of high-quality gated CBCT imaging combined with gating radiotherapy. Introduction Purpose: To introduce a next-generation imaging paradigm named “nonstop gated cone-beam CT (ngCBCT)” that improves upon current clinical free-breathing gated CBCT (gCBCT) on C-arm linear accelerators by substantially reducing the scan time and imaging dose while retaining high-quality images. -15- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 Methods: ngCBCT is achieved by allowing the gantry to rotate continuously and having the kV X-ray beam turned on only when the breathing signal is within the specified gating window of the respiratory cycle. Clinical gCBCT projections of 10 respiratory gating lung SBRT patients were retrospectively retrieved and intentionally sampled based on each patient’s actual respiratory cycles to emulate the ngCBCT acquisitions. The breathing cycle information was used to calculate and compare the acquisition time and CT dose index (CTDI) of the current clinical gCBCT against the proposed ngCBCT acquisitions. Three reconstruction algorithms (FDK, compressed sensing-based iterative reconstruction (IR), and prior image-based iterative reconstruction (PIBR)) were each applied to these ngCBCT emulations to examine their performance on the non-uniform and under-sampled projections resulting from this acquisition strategy. Results: Compared with current clinical gCBCT, the proposed ngCBCT can reduce data acquisition time from 2-7 mins to 1 min for half-fan mode (offset detector, 360- degree rotation) or from 1.1-3.9 mins to 0.56 min for full-fan mode (no detector offset, 200- degree rotation). Meanwhile, the CTDI is also reduced by 45%-78% for ngCBCT acquisitions. Comparing the three reconstruction algorithms for ngCBCT, FDK resulted in images with insufficient quality for clinical use, while the PIBR method consistently delivers better visual and quantitative results than IR. Conclusions: An imaging paradigm that addresses the major challenges of current clinical gCBCT was proposed and validated. Using the PIBR method for ngCBCT yields adequate image quality for tumor localization and patient alignment tasks. This work lays the foundation to realize ngCBCT in clinical use to improve both patient experience and clinical workflow. Impact The novel nonstop gated CBCT (ngCBCT) technique solves the major drawback of the current clinical implementation of free-breathing gated CBCT (gCBCT) on a C-arm linear accelerator (Varian’s TrueBeam®), which is the long data acquisition time (2-7 minutes for a typical duty cycle of 30%-60% and patient breathing period of 3-6 seconds). In the existing gCBCT implementation, the gantry movement is interrupted and -16- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 resumed by the respiratory gating signal multiple times over the scan (the gantry stops whenever the breathing signal is out of the respiratory gate). In the proposed ngCBCT implementation, the gantry rotates continuously at the maximum permitted speed (e.g., 6º / second or 1 revolution / minute on TrueBeam®) irrespective of the patient’s breathing signal. Also, the kV X-ray is only turned on when the patient’s breathing signal is within the respiratory gate (FIG.1), thus substantially decreasing the imaging dose to the patient (FIG.2). The prior-image-based iterative reconstruction (PIBR) framework can reconstruct high-quality images from the non-uniform and under-sampled projections (FIG.2) resulting from this data acquisition strategy. This accelerated scan reduces overall patient on-table time, alleviates patient discomfort, and decreases the likelihood of patient movement and breathing pattern drift during subsequent gating radiotherapy, thus improving both patient experience and treatment outcome. Additionally, in clinical gCBCT, a significant fraction of the kV projections is acquired while the gantry is accelerating, resulting in higher imaging dose than standard 3D CBCT; in the proposed ngCBCT, the imaging dose is much lower than that of 3D CBCT because the kV beam is only on within gating window (FIG. 2). To benefit more patients with the advantages of gated SBRT treatments (e.g., minimize the organs-at-risk toxicity), innovations, such as the proposed ngCBCT, are desired to fulfill the unmet clinical needs. Methods Clinical gCBCT projections of 10 gating lung SBRT patients were retrospectively retrieved and intentionally sampled based on actual patient respiratory cycles to emulate the ngCBCT acquisitions (because the ngCBCT acquisition mode is not yet available on TrueBeam®). These gCBCT scans were acquired with either Half-Fan (HF, offset detector, and 360º rotation, FIG.3) or Full-Fan (FF, no detector offset, 200º rotation, FIG.4) mode. Each kV projection has 1024x768 pixels with a resolution of 0.388x0.388 mm2. Before image reconstruction, the projections were processed with Varian iTools for scatter and beam hardening corrections. Three reconstruction algorithms: FDK, compressed sensing-based iterative reconstruction (IR), and prior image-based iterative reconstruction (PIBR) have been investigated. Specifically, the PIBR method is written as: -17- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 ^̂^^^ = arg where two regularization terms follow the likelihood-based data fidelity term with current projection data (^^^^), current anatomy to be estimated (^^^^), and a prior image volume (^^^^^^^^) with registration ^^^^ applied (rigid registration in this work). The reconstructed image smoothness is controlled via ^^^^^^^^and set to be 10 for all studies. The amount of prior information integrated into a reconstruction is controlled via ^^^^^^^^and set to be 10 for all studies so that it improves image reconstruction while not introducing any false structures from prior images. Results For one HF scan patient A (FIG.3) and one FF scan patient B (FIG.4), Table 1 has illustrated the comparison between gCBCT and ngCBCT regarding data acquisition time and CT dose index (CTDI). The FF gCBCT takes 2.8 min for only 200º rotation because patient B has a very short breathing period and uses a low gating duty cycle. In contrast, the FF ngCBCT acquisition is only 1 min * 200º / 360º = 0.56 min. The CTDI of 3D CBCT (HF / FF mode) was referred as 100%. The clinical gCBCT images (1st column of FIG.3 and 4) reconstructed from complete projections are used as ground truth for the evaluations. With the ngCBCT projections (e.g., non-uniform and under-sampled), the FDK, IR, and PIBR methods were implemented. All the images were reconstructed with 512x512x160 voxels and 1x1x1 mm3resolution. In FIG.3, the image margins were cropped for better visualization. In FIG.4, the images were cropped only to show the non- truncated ROI regions. Table 1: Comparisons between gCBCT and ngCBCT acquisitions for two selected patients. Patients Average Gating gCBCT Scan time gCBCT CTDI breathing duty vs. reduction vs. reduction period cycle ngCBCT with ngCBCT with scan time ngCBCT CTDI ngCBCT -18- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 patient A: 6.0 52% 2.4 min 58% 107% vs. 51% HF seconds vs.1 min 52% patient B: 3.9 34% 2.8 min 80% 130% vs. 74% FF seconds vs.0.56 34% min Severe streak artifacts were observed in FDK reconstructions, rendering them not clinically usable. The two iterative reconstruction methods demonstrated clear advantages in image quality. With the aid of patient-specific prior images, the PIBR method improves the image quality, overall texture, and bony structure details of the reconstructions compared to the IR method. In all scenarios, the PIBR images yield the best image quality in terms of peak signal-to-noise ratio (PSNR) and universal image quality index (UQI). Also importantly, PIBR reconstructions do not introduce any false structures, even if there are some anatomical differences (indicated with arrows on prior images) between the prior volume and the current scan. In this work, the prior CBCT volume is from a previous scan of the same patient. Synthesizing prior CBCT volume from planning CT images with cycleGAN for PIBR reconstruction may be further explored. The management of respiratory motion in radiotherapy is important for thoracic and abdominal cancer patients (e.g., lung, liver, pancreas). Managing tumor motion reduces margin around target and supports dose escalation, thereby improving local tumor control and decreasing the normal tissue complication probability. Compared to deep inspiration breath hold (DIBH) technique, respiratory gating allows free breathing (FB) and thus is less demanding for cancer patients. In respiratory gating radiotherapy, an external surrogate is usually employed to monitor patient breathing (e.g., RPM, RGSC) and allows delivery of radiation only during preset gating window of the breathing cycle (e.g., 30%-70%). The same motion management strategy is applied for treatment and pre-treatment CBCT imaging. For respiratory gating lung SBRT, free breathing gated CBCT is utilized clinically for tumor localization / alignment before treatment. -19- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 A new imaging paradigm is presented (named “nonstop gated CBCT”) which addresses the major challenges of current clinical gated CBCT by substantially reducing scan time and imaging dose. The image reconstruction of nonstop gated CBCT is more challenging (especially for HF scans) than traditional sparse-view CBCT, partially due to the non-uniform distribution of projections. Prior-image-based iterative reconstruction (PIBR) method can greatly improve image quality of nonstop gated CBCT in both HF and FF modes, as compared to the FDK algorithm and compressed sensing (CS)-based iterative reconstruction. A dual-domain (i.e., projection and image domains) neural networks may be used for image reconstruction of nonstop gated CBCT. B. Deep-Learning Empowered Fast-Gated CBCT Imaging for Respiratory Gating Radiotherapy Purpose: Gated CBCT is a motion-reduced imaging technique for guiding free-breathing respiratory gating radiotherapy of thoracic and abdominal cancer patients. However, the scan takes 2-7 minutes on the C-arm linear accelerators because the gantry movement is interrupted and resumed by the respiratory gating signal multiple times over the scan. To achieve faster imaging and reduce imaging dose, a “nonstop gated CBCT” technique and an Iterative-reconstruction-based Residual Network (IResNet) to reconstruct high-quality images from this novel acquisition technique was developed. Methods: Nonstop gated CBCT is achieved by allowing the gantry to rotate continuously and having the X-ray beam turned on only when the breathing signal is within the specified gating window of the respiratory cycle. This data acquisition technique results in non-uniform and under-sampled projections. Two independent IResNet for Half-fan (imager offset, larger field-of-view) and Full-fan (no imager offset, smaller field-of-view) nonstop gated CBCT were trained using 6720 and 9600 clinical images, respectively, and their outputs were compared to FDK, iterative reconstruction (IR), and a classical convolutional neural network (e.g., FBPCONVNet) approach. Results: The FDK method failed to yield clinically acceptable image quality for the nonstop gated CBCT acquisition. While IR can deliver improvements, the resulting images still contain some artifacts. The FBPCONVNet can remove most of the artifacts, -20- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 but the images are usually over-smoothed, and the network frequently introduces hallucinations. In contrast, the proposed IResNet is superior: great artifact reduction, excellent image texture preservation, and much lower hallucination occurrence. Conclusions: An IResNet to solve the challenge of image reconstruction in the presence of non-uniform and under-sampled projections in support of the novel nonstop gated CBCT technique was developed. IResNet’s ability to greatly improve image quality in both Half-fan and Full-fan modes was demonstrated. This feasibility study is critical for the clinical translation of the proposed nonstop gated CBCT for respiratory gating radiotherapy. Impact The deep-learning (DL) based convolutional neural network (CNN) was developed and tested, named Iterative-reconstruction-based Residual Network (IResNet), that has shown to be superior over Feldkamp-Davis-Kress (FDK) or penalized-likelihood iterative reconstruction (IR) methods in reconstructing the non-uniform and under-sampled projection data that is acquired when performing a novel acquisition technique that have been developed called “nonstop gated CBCT.” Nonstop gated CBCT is innovative because it is acquired in 1 minute, compared to the 2-7 min (depending on duty cycle and patient breathing period) needed in free-breathing gated CBCT on C-arm linacs (e.g., Varian’s TrueBeam®), reducing the opportunity for patient involuntary movement and breathing baseline shift. Therefore, a faster gated CBCT scan can improve not only patient experience but also treatment quality. In nonstop gated CBCT, the gantry rotates continuously at the maximal speed (6º / second on TrueBeam®) while only acquiring kV projections when the patient’s breathing signal is within the respiratory gate (FIG.17). Since the kV X-ray is not on all the time, the imaging dose to the patient is also reduced. Methods Clinical gated CBCT projections of 10 gating lung SBRT patients were retrospectively retrieved, which include 23 Half-Fan (HF, offset detector, 360º rotation, larger field-of-view (FOV)) scans and 30 Full-Fan (FF, no imager offset, 200º rotation, smaller FOV) scans. The gated CBCT projections were intentionally sampled based on real -21- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 respiratory cycles to emulate the nonstop gated CBCT acquisitions (because this proposed acquisition mode is not yet available on TrueBeam®), as in FIG.18. Conventional reconstruction methods such as FDK and IR are tested with these non-uniform and under- sampled projections. The FBPCONVNet (one neural network originally designed to denoise low-dose CT) was explored to reconstruct these projections. Based in part on FBPCONVNet and ResNet, the IResNet (FIG.19) was also developed for this particular application. Specifically, the projections are reconstructed with the IR method, and these reconstructions are used as the input for neural network training (in 2D). The gated CBCT images reconstructed from complete projections are used as the ground truth. The two independent networks were trained for the HF and FF modes. The 14 HF scans and 20 FF scans were utilized respectively, for the neural network training and the rest for validation and testing. Data augmentation with horizontal or vertical flips of the original image is performed for both cases. Thus, there are 6720 (=14x160x3) images for the HF training and 9600 (=20x160x3) images for the FF training; 960 images are used for HF validation, and 2160 images for FF validation. Both networks are trained for 50 epochs, with MSE as the loss function and NAdam as the optimizer. The test images (320 images for HF and 800 for FF) are obtained from patient scans independent of the ones used for training and validation. The results from the IResNet are compared with the other approaches. Results FIG.20 illustrates the reconstruction results of one HF scan (upper row) and one FF scan (lower row) for two gating lung SBRT patients. Figures 17 and 18 show the reconstruction results in sagittal and coronal views of HF and FF scans from two other patients. The CBCT images reconstructed with the conventional FDK or IR methods suffer from severe streak artifacts (FIG.20, panels b, c, g, h, and Fig.22, panel b) due to the non- uniform and under-sampled projections of nonstop gated CBCT. The IR method also introduces occasional fringe artifacts (FIG.21, panels c, and h). The FBPCONVNet is effective for sparse-view CT but does not perform well for nonstop gated CBCT: the reconstructions are mostly over-smoothed, with frequent hallucinations (e.g., anatomical structures that do not exist). The proposed IResNet outperforms the other methods, with -22- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 superior image quality for the tumor and general texture and detail preservation (FIG.20, panels e, j, FIG.21, panel j, FIG.22, panels e and j). These improvements are mainly due to the addition of the ResNet blocks. Also, during hyper-parameter tuning of IResNet, the NAdam optimizer (compared to SGD or Adam) was discovered that the training results are much improved, and the training also converges much faster. The performance of IResNet with both HF and FF scan modes was tested and observed superior results in all scenarios. Discussion The IResNet was developed to solve the challenge of image reconstruction in the presence of non-uniform and under-sampled projections in support of the novel nonstop gated CBCT technique, which can reduce scan time to 1 minute for an HF scan (only 0.56 min for an FF scan) and also substantially decrease the imaging dose to patients. The network’s ability was demonstrated to greatly improve the image quality of nonstop gated CBCT in both HF and FF modes. Future studies focus on developing dual-domain (e.g., projection and image domains) neural networks for image reconstruction of the nonstop gated CBCT. CNN approaches: Many CNNs developed to denoise low-dose CT, such as UNet based (e.g., FBPCONVNet). These CNNs are not specifically designed to handle scans reconstructed with non-uniform projections. FBPCONVNet results are not optimal, due to over-smoothed texture, inability to fully remove streak artifacts, and creation of many hallucinations, among others. To deal with the unique nonstop gated CBCT reconstructions, IResNet may be used. IR (PL or PIBR) may be used as inputs; the weights may be based on UNet with a skip connection between the full view and the nonstop gated view. ResNet Blocks may also be used between the encoding and decoding layers. For training of ResNet Blocks, two separate networks for Half-fan and Full-fan scans may be used. Samples (HF): Training set included 6720 images (5 patients). Validation set included 960 images (2 patients). Test set included 320 images (2 patients). Training: 50 epochs (~10 hours). For each scan, 160 slices (images) are used. The first and last 20 slices do not contain useful info and are not used. Data augmentation increase number of images by 3 (original + horizontal flip + vertical flip). -23- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 C. Scan Efficiency and Imaging Dose Analysis of Next-Generation Nonstop Gated CBCT for Respiratory Gating Lung Radiotherapy Purpose: Gated CBCT (gCBCT) is commonly employed for respiratory gating lung cancer patients to ensure precise patient setup. However, the scan is time- consuming on C-arm linear accelerators (LINAC) due to frequent interruptions in gantry rotation dictated by respiratory gating signals. This study aims to implement a novel nonstop gated CBCT (ngCBCT) technique on LINAC and quantitatively compare its scan efficiency and imaging dose to the current clinical gCBCT. Methods: gCBCT was acquired in the clinical mode of a C-arm LINAC, while ngCBCT was implemented via a customized XML file in the developer mode. Both techniques utilize the same thorax imaging protocol (half fan and full trajectory). The weighted Cone-Beam Dose Index (CBDI) was quantified using a standard CTDI body phantom and two pencil chambers that measure central and peripheral imaging doses. Respiratory patterns, including Cos4 motion (3-6 seconds cycles) and three clinical patient breathing patterns, were simulated using a CIRS surrogate motion platform. Different gating duty cycles (30%-60%) were tested on Cos4 motion, while one duty cycle was reproduced for each patient’s breathing pattern. Scan times were determined by analyzing the timestamps of the projection data. Results: Unlike gCBCT, where scan time ranges from 1.8 to 5 minutes— depending on the gating duty cycle and slightly on breathing period—ngCBCT scan times remain consistently around 1 minute. For imaging dose, the weighted CBDI (CBDIw) for ngCBCT is reduced to 26.7%-60.1% (nearly identical to selected gating duty cycle) of gCBCT (approximately 5.2 mGy across all gCBCT scans). Conclusion: The ngCBCT technique provides a transformative boost in scan efficiency and dose reduction over current clinical gCBCT. Empowered by our deep learning-based reconstruction framework, ngCBCT maintains high image quality without compromising precise patient setup. This innovation not only improves the patient experience but also expands access to advanced respiratory gating radiotherapy. -24- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 Innovation / Impact: A novel nonstop gated CBCT (ngCBCT) technique on C-arm LINAC was realized and thoroughly evaluated its performance in terms of scan time and imaging dose. Compared to current clinical gated CBCT (gCBCT), ngCBCT offers significant reductions in both scan time and imaging dose. This acceleration not only decreases overall patient on-table time but also alleviates patient discomfort, reduces the risk of patient movement, and minimizes breathing pattern drift during respiratory gating radiotherapy. Additionally, the reduced imaging dose lowers the potential risk of developing secondary cancers. When combined with our deep learning-based reconstruction framework, ngCBCT can maintain high image quality as gCBCT. These advancements are impactful for respiratory gating lung cancer patients who typically undergo 3-5 scans for SBRT and 20-35 scans for conventional fractionated treatments, further highlighting the clinical significance of this innovation. Methods: The ngCBCT is implemented via a customized XML file in the developer mode, while gCBCT is acquired in the clinical mode of a Varian TrueBeam LINAC. The same thorax imaging protocol (half-fan, full trajectory, x-ray voltage 125 kV, current 15 mA, pulse length 20 ms) was used for both techniques. One CTDI body phantom (32 cm diameter PMMA) and two RaySafe 100-mm pencil chambers (placed in the central and peripheral slots respectively) were utilized to quantify the Cone-Beam Dose Index (CBDI) of all the scans, as shown in Fig.27. The weighted CBDI was calculated as^^^^^^^^^^^^^^^^^^^^ =1 3 ^^^^^^^^^^^^^^^^^^^^+ 2 3 ^^^^^^^^^^^^^^^^^^^^. Both the ^^^^^^^^^^^^4 motion (with cycle periods of 3 to 6 seconds) and three clinical patients’ breathing patterns and were simulated using the CIRS Dynamic Thorax Phantom motion control software, as shown in Fig.25. The duty cycles of all motion patterns ranged from 30% to 60%, reflecting clinical settings. The scan time of ngCBCT and gCBCT was determined using the timestamps of projection data. Key Results: For the ^^^^^^^^^^^^4 motion patterns, gCBCT scan times ranged from 1.8 to 4.9 minutes, while ngCBCT scan times remained consistently around 1 minute. This represents a time reduction of 45%-80%, primarily influenced by the gating duty cycle and slightly by the cycle period. The ^^^^^^^^^^^^^^^^^^^^for gCBCT was 5.2 ± 0.08 mGy across all cycle periods and duty cycles, while the ^^^^^^^^^^^^^^^^^^^^of ngCBCT was reduced to 27.5% to 60.1% (nearly identical to the selected gating duty cycle). The results are shown in Fig.26. For -25- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 the three patient breathing traces, gCBCT scan time was 3.4, 2.4, and 5.0 minutes, respectively, while ngCBCT scan time remained at 1 minute, yielding time reduction of 71%, 58%, and 80%. Due to the irregular periodicity of the patient breathing traces, achieving a consistent gating duty cycle was challenging, even with identical upper / lower gating thresholds. Therefore, two ngCBCT scans were acquired for each patient breathing trace, with all the results displayed in Fig.27. The ^^^^^^^^^^^^^^^^^^^^of ngCBCT was reduced to 26.7%-52.7% of that of gCBCT, depending on the gating duty cycle of ngCBCT scan. The dose rate profiles (dose rate vs. scan time) measured at the central and peripheral locations for a single ngCBCT scan using a patient breathing trace are shown in Fig.28. D. Nonstop Gated CBCT for Respiratory Gating Lung SBRT with Image Reconstruction Using Prior-Image-Based Iterative Reconstruction Background: Free-breathing gated cone-beam computed tomography (gCBCT), which captures a specific anatomy coinciding with a preset gating window in the breathing cycle, is routinely prescribed to gating lung stereotactic body radiation therapy (SBRT) patients for pretreatment setup verification. However, a half-fan gCBCT scan can take 2-8 minutes (for a typical gating duty cycle of 30%-60% and patient breathing period of 3-6 seconds) on a C-arm linear accelerator because the gantry movement is interrupted and resumed by the respiratory gating signal multiple times over the scan. The long scan time increases patient on-table time, leading to discomfort and a higher likelihood of patient movement. Meanwhile, extra kV projections are acquired while the gantry is accelerating for the gCBCT scan, resulting in a higher imaging dose compared to 3D CBCT. Purpose: To investigate the feasibility of a novel imaging paradigm named “nonstop gated CBCT (ngCBCT)” that improves upon current clinical gCBCT by substantially reducing the scan time and imaging dose while retaining high-quality images. Methods: ngCBCT is implemented by allowing the gantry to rotate continuously, with the kV X-ray beam activated only when the breathing signal falls within the preset gating window. Raw gCBCT projections of two gating lung SBRT patients were retrospectively retrieved and intentionally sampled based on each patient’s respiratory cycle to emulate the ngCBCT acquisitions. The datasets include both half-fan and full-fan -26- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 acquisitions, representing the primary clinical scan geometries. Three reconstruction algorithms - FDK, penalized likelihood iterative reconstruction (PL), and prior-image-based iterative reconstruction (PIBR) - were applied to these ngCBCT emulations to evaluate reconstruction performances on the non-uniform and under-sampled projections resulting from this acquisition strategy. Results: The FDK reconstructions of ngCBCT are degraded with streak artifacts and have insufficient quality for clinical use. While PL yields improved reconstructions over FDK, the PIBR method consistently delivers the best visual and quantitative results with the aid of patient-specific prior images. Conclusion: The proposed ngCBCT technique addresses the key limitations of current clinical gCBCT by substantially reducing data acquisition time and imaging dose. The ngCBCT with PIBR achieves adequate image quality and offers a promising opportunity for pretreatment setup verification in gating lung SBRT. 1. Introduction Stereotactic body radiation therapy (SBRT) utilizes precise image guidance and radiation delivery to achieve high tumor dose delivery. It has demonstrated favorable treatment outcomes for early-stage non-small-cell lung cancer (NSCLC) and lung metastases. However, one significant challenge in lung SBRT is managing tumor motion caused by the patient’s respiratory cycle. A conventional but suboptimal solution is to expand the planning target volume (PTV), which inadvertently increases radiation exposure to normal tissues. Over the past few decades, various motion management strategies have been developed to address respiratory-induced tumor displacement. For patients who cannot tolerate breath hold, respiratory gating (also known as “gating”) offers an effective alternative by allowing free breathing during treatment. In respiratory gating radiotherapy, radiation is delivered only within a specified gating window corresponding to a selected range of the respiratory cycle (often measured using an external surrogate placed on the chest). This approach aims to correlate chest movement with tumor motion. By reducing tumor movement during radiation delivery, respiratory gating minimizes toxicity to organs- at-risk (OARs) and facilitates dose escalation, enhancing the efficacy of SBRT. -27- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 Pretreatment imaging is essential to ensure the safe and effective delivery of respiratory gating lung SBRT. Specifically, a gated cone-beam computed tomography (gCBCT) scan can be performed to verify gating amplitudes and assist with patient positioning. This imaging approach captures the patient’s anatomy as it will appear during the subsequent respiratory gating radiotherapy because the same gating window is applied for both imaging and treatment delivery. Compared to conventional 3D CBCT, the gCBCT reduces image blurring and motion artifacts caused by respiratory motion, providing more precise visualization for lung tumors and surrounding OARs. With all the advantages of gCBCT, one significant drawback is the lengthy scan time. Prolonged scans can lead to patient discomfort and an increased likelihood of patient movement or breath pattern drift. Typically, clinical gCBCT scans on a Varian TrueBeam linear accelerator take 2-8 minutes (depending on the patient’s breathing period and gating duty cycle), compared to approximately 1 minute for a 3D CBCT scan on the same machine. This disparity highlights a clear unmet clinical need for a more time- efficient gCBCT method. To address this, a novel gCBCT acquisition technique was proposed: nonstop gated CBCT (ngCBCT). This technique retains the gating aspect of current gCBCT, acquiring projections only within the specified gating window. However, unlike current gCBCT, the gantry rotates continuously without interruptions from respiratory gating signals. The ngCBCT acquisition reduces the scan time on a Varian TrueBeam to approximately 1 minute for a half-fan / full-trajectory scan (360º gantry rotation) and 0.56 minutes for a full-fan / half trajectory scan (200º gantry rotation). The reduction in scan time can be substantial, given that each patient typically undergoes 3-5 treatment fractions, with 1-2 scans acquired per fraction. The ngCBCT acquisition results in non-uniform and under-sampled projections, which pose challenges for image reconstruction. Unlike conventional sparse- view CT or CBCT, where projections are uniformly under-sampled, the ngCBCT projections exhibit multiple strip gaps. Classical reconstruction methods, such as the Feldkamp-Davis-Kress (FDK) algorithm, penalized likelihood iterative reconstruction (PL), and prior-image-based iterative reconstruction (PIBR), have demonstrated the ability to reconstruct CBCT images from sparse-view (under-sampled and uniformly distributed) -28- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 projection data. In this study, an investigation and evaluation of the above reconstruction methods using ngCBCT projection data, which are under-sampled but non-uniform was run. This feasibility study can serve as a reference for future deep-learning-based reconstruction approaches to enhance ngCBCT image quality further. 2. Methods 2.1 Data Acquisitions Clinical gCBCT projections from two respiratory gating lung SBRT patients were retrospectively retrieved. Both patients had solid lung tumors and multiple gCBCT scans (required for the PIBR method). These scans were acquired using the onboard CBCT system on a Varian TrueBeam (Varian Medical Systems, Palo Alto, CA) linear accelerator. The scans of patient A were acquired in the half-fan (HF) mode with the detector offset and the gantry rotating 360º. The scans of patient B were performed in the full-fan (FF) mode, where the detector was not offset and the gantry rotated nearly 200º. The source-to- isocenter distance was 100 cm, while the source-to-detector distance was 150 cm. The raw projections were processed using the Varian iTools (Varian Medical Systems, Palo Alto, CA) for scatter and beam hardening corrections. Each projection consists of an image size of 1024 x 768 pixels with a resolution of 0.388 x 0.388 mm2. Finally, a 2x2 detector rebinning was applied to projections to reduce noise and computational load. Since the ngCBCT acquisition is not yet available on the Varian TrueBeam system, ngCBCT is emulated by down-sampling the clinical gCBCT projections based on the patient’s respiratory cycles. First, gCBCT projections corresponding to different breathing cycles are identified. Then, by selecting projections from every other breathing cycle (e.g., 1st, 3rd, 5th, etc.) or every third breathing cycle (e.g., 1st, 4th, 7th, etc.), ngCBCT acquisitions with various duty cycles can be simulated. In radiotherapy, the gating duty cycle is defined as the ratio of the beam-on treatment time to the total treatment time. FIG. 29 illustrates the ngCBCT projection data emulations using a clinical gCBCT scan of patient A. The gating amplitude of each projection in Fig.29, panel a was read from the raw projection data header. The sinogram data of one gCBCT slice is shown in Fig.29, panel b while the sinogram data of ngCBCT emulations are shown in Fig.29, panels c and d -29- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 respectively. The green dots in FIG.29, panels b-d represent the kV source positions. Notably, the kV source is activated only when the breathing signal falls within the preset gating window during ngCBCT. A comparison of the scan time and the CT dose index (CTDI) of emulated ngCBCT scans with that of the clinical gCBCT was run. The scan time is reduced by more than a factor of three compared to the clinical gCBCT scans, as summarized in Table 1. Specifically, the scan time is 1 minute for an HF scan and 0.56 minutes for an FF scan, independent of the gating duty cycle. The scan time and CTDI reduction positively correlate with the patient’s average breathing period and the gating duty cycle. FIG.30 shows the kV source on / off distributions for the HF (patient A, higher duty cycle) and FF (patient B, higher duty cycle) scanning geometry setups. Table 1: Comparison between gCBCT and ngCBCT emulations on scan time and CTDI for the two patient scans used in this study. Patient A (HF) Patient B (FF) Breathing Period 2.89 - 4.51 (mean 3.59) 2.54 - 3.55 (mean 2.91) (sec) Scan Method gCBCT ngCBCT ngCBCT gCBCT ngCBCT ngCBCT Gating Duty - 53% 32% - 49% 34% Cycle Scan Time 3.8 min 1 min 1 min 2.1 min 0.56 min 0.56 min CTDI 100% 53% 32% 100% 49% 34% The ngCBCT method reduces scan time and CTDI in both cases. 2.2 Image Reconstruction The three reconstruction algorithms for the unique ngCBCT projections were investigated: the Feldkamp-Davis-Kress (FDK) algorithm, penalized likelihood iterative reconstruction (PL), and prior-image-based iterative reconstruction (PIBR). -30- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 The offset detector setting for HF acquisitions leads to asymmetric patient coverage across different projection views and introduces data redundancy in the central region. Therefore, appropriate detector weighting is essential to mitigate ring artifacts in the FDK-reconstructed images. In this work, the same weighting function was applied for the FDK method. In the case of FF acquisitions, projection data can be truncated, which may result in significant truncation artifacts (e.g., bright rings) in the FDK-reconstructed images. To address these artifacts, A projection data extrapolation was performed to maintain continuity at the boundaries between the measured and extrapolated regions. The penalized-likelihood (PL) method optimizes an objective function, which can be formulated as follows: ^̂^^^ = arg max ^^^^ ( ^̂^^^;^^^^) − ^^^^^^^^||^^^^^^^^||1 (1)where L denotes the data fidelity term, a log-likelihood function derived from a Poisson noise model, y denotes the vector of projection measurements, ^^^^ and represents the vector of attenuation coefficients of the object to be reconstructed. The second term is a regularization which reflects prior knowledge or expectation of the characteristics of the image to be reconstructed. In this work, a standard l1 norm was utilized where ^^^^ denotes a pairwise voxel difference operator (e.g., first-order neighbors) and ^^^^^^^^control the smoothness of the reconstruction. The Prior-Image-Based iterative Reconstruction (PIBR) method utilizes patient-specific prior images, which is written as: ^̂^^^ = arg max ^^^^ ( ^̂^^^;^^^^) − ^^^^^^^^||^^^^^^^^||1 − ^^^^^^^^||^^^^ − ^^^^^^^^^^^^||1 (2)A second regularization term encourages similarity between the prior image (^^^^^^^^) and current anatomy (^^^^). Rigid registration T is performed between them to align the two volumes. The ^^^^^^^^factor controls the amount of prior information integrated into the reconstruction. 2.3 Image Quality Metrics -31- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 Two metrics were employed to evaluate the image quality of the reconstructions: the peak signal-to-noise ratio (PSNR) and the universal image quality index (UQI). Both metrics are relative, meaning that image quality is assessed compared to the original clinical gated CBCT image reconstructed by FDK. The PSNR is a widely used image quality metric. It compares pixel values between two images and is related to the mean squared error. Higher PSNR values indicate better image reconstruction quality. The UQI is a metric that evaluates the similarity between the original and reconstructed images. The UQI metric is defined as: This metric evaluates three factors between the two images: correlations, mean luminance, and contrast similarity. The UQI value is between 0 to 1, with higher values indicating better image reconstruction quality. While PSNR evaluates image quality based on pixel value differences, in cases of distortions such as noise, the UQI demonstrates advantages over traditional error summation-based metrics like PSNR. Therefore, combining PSNR and UQI provides a more comprehensive image quality evaluation. 3. Results 3.1 Higher gating duty cycle By selecting projections from every other breathing cycle (e.g., 1st, 3rd, 5th, etc.), ngCBCT acquisitions with a higher gating duty cycle were simulated. Specifically, the ngCBCT simulations include 483 projections (26 segments) for the HF scan and 241 projections (21 segments) for the FF scan. The under-sampled and non-uniform ngCBCT projections were reconstructed using the commonly employed FDK and PL methods. Additionally, PIBR was implemented by leveraging patient-specific prior images. It is important to note that these prior images are from a previous gCBCT scan of the same patient, although the anatomical structures do not perfectly match the current scan. The reconstruction results from ngCBCT emulations are compared with the clinical gCBCT -32- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 images (reconstructed from complete projections), as shown in FIG.31 and FIG.32. All the reconstructed images have a resolution of 512x512x160, with a voxel size of 1 mm x 1 mm x 1 mm. Both FDK and PL reconstructions from ngCBCT projections exhibited artifacts, primarily due to the non-uniformity in the projection data. In particular, the FDK and PL reconstructions appear worse for the HF scan than the FF scan. However, improved visual image quality was observed in the PIBR reconstructions. Although the prior images used by the PIBR method exhibit anatomical differences compared to the current anatomy, these differences do not lead to the introduction of false structures. The ^^^^^^^^factor in the PIBR objective function regulates the degree of prior information incorporated during reconstruction. For all the studies, The ^^^^^^^^was set to 103.5, while ^^^^^^^^was set to 102.5. PIBR reconstructions demonstrated improved image quality without introducing anatomical artifacts from the prior images. 3.2 Lower gating duty cycle The number of acquired ngCBCT projections is directly related to the gating duty cycle. Therefore, what is hypothesized is that reconstructed image quality correlates with the duty cycle. To test this, additional ngCBCT acquisitions with lower duty cycles were simulated by selecting projections from every third breathing cycle (e.g., 1st, 4th, 7th, etc.). For the HF scan, the duty cycle was reduced from 53% (FIG.31) to 32% (FIG.33). Similarly, for the FF scan, the duty cycle was lowered from 49% (FIG.32) to 34% (FIG. 34). There are 289 projections for the HF scan and 167 projections for the FF scan. It is evident that FDK and PL reconstructions substantially reduce image quality with lower duty cycles. Meanwhile, PIBR can still yield adequate image quality with lower gating duty cycle for patient setup. 3.3 Quantitative evaluations Across all simulated duty cycles and scan scenarios, PIBR consistently outperformed FDK and PL, with noticeable visual improvements. The PSNR and UQI values for the reconstructions shown in Figures 33–36 are summarized in Table 2. Additionally, the average PSNR and UQI values for all the 160 slices in the reconstructed -33- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 volumes are listed in Table 3. These results indicate that PIBR achieved the highest PSNR and UQI for all the reconstructions, demonstrating superior reconstruction accuracy. Table 2: PSNR and UQI of the reconstructed images in Figures 33 - 36. The proposed PIBR method yields the highest PSNR and UQI in all cases. FDK Patient Duty Metric Transverse Coronal Sagittal Cycle PSNR 18.64 19.88 18.98Patient A53%(HF) UQI 0.86 0.88 0.89 32%PSNR 14.43 14.58 14.93UQI 0.68 0.68 0.62PSNR 25.28 25.16 24.26Patient B49% UQI 0.90 0.90 0.91(FF)34%PSNR 22.13 22.07 21.46UQI 0.81 0.80 0.85PL Patient Duty Metric Transvers Corona Sagittal Cycle e l PSNR 24.75 24.15 24.55UQI 0.96 0.94 0.96 PSNR 22.12 21.83 22.42UQI 0.92 0.89 0.9249%PSNR 26.46 26.59 25.18Patient B (FF) UQI 0.90 0.91 0.92 34%PSNR 24.59 24.56 23.66UQI 0.84 0.83 0.89PIBR Patient Duty Metric Transvers Coronal Sagittal Cycle e 53PSNR 25.23 24.83 24.94Patient A%UQI 0.96 0.95 0.96 32%PSNR 23.64 23.33 23.79UQI 0.95 0.92 0.95Patient B PSNR 26.72 26.81 25. F)466 (F9%UQI 0.91 0.91 0.93 -34- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 PSNR 25.28 25.24 23.85UQI 0.87 0.87 0.89Table 3: Average PSNR and UQI for all the 160 slices in the reconstructed CBCT volumes. Patient Duty Cycle Metric FDK PL PIBR PSNR 18.67 23.94 24.28 53% Patient A UQI 0.87 0.95 0.96 (HF) PSNR 14.19 22.02 23.22 32% UQI 0.67 0.92 0.94 PSNR 24.82 25.10 25.90 49% Patient B UQI 0.89 0.91 0.92 (FF) PSNR 21.37 23.66 23.72 34% UQI 0.82 0.85 0.86 4. Discussions The proposed ngCBCT acquisition method presents significant challenges for achieving high-quality image reconstruction suitable for clinical use. Filtered back- projection methods, by design, cannot handle the non-uniformity of projection data, particularly when large gaps exist between projections. As a result, severe streak artifacts are observed in FDK-reconstructed images, rendering them clinically unusable. In contrast, the iterative nature of the PL method makes it more robust to non-uniform projections. In our study, PL demonstrated notable improvements over the FDK method, with far fewer streak artifacts. Overall, PL resulted in higher increase of PSNR and UQI for the HF scan than that for the FF scan. The smaller improvements for FF scans might be attributed to the same sampling region of interest (ROI) because of no detector offset and smaller projection gaps caused by shorter breathing periods. However, one limitation of PL-reconstructed images is their over-smoothed texture and patchy artifacts. The PIBR method offers a solution by using prior images to produce textures closer to those seen in FDK-reconstructed CBCT images. While the PSNR and UQI improvements may not be substantial, what was observed that the textures of all PIBR reconstructions closely resemble those of the gCBCT images. Future studies will involve -35- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 radiation therapists to evaluate the usability of ngCBCT images for patient setup tasks, providing a more clinical perspective on their quality and applicability. The current implementation of PIBR relies on patient-specific prior images, typically obtained from previous gCBCT scans of the same patient. This approach is only feasible if the patient undergoes multiple scans. To expand the applicability of the PIBR method to all cases— including first-time or sole scan for a patient— alternative methods for generating prior images are required. One possible solution is using deep-learning techniques, such as Generative Adversarial Networks (GANs) or CycleGAN, to synthesize prior gCBCT images from the patient’s planning CT. Our study demonstrated that anatomical differences between prior images and the current scan can be minimized by adjusting the factor in the PIBR formalism that controls the incorporation of prior information. In this proof-of-concept study, only two patient datasets were used, and two gating duty cycles were simulated. Clinically, gating duty cycle between 30% and 60% are typically employed. Values above this range offer limited dosimetric benefits, while those below can lead to excessively long treatment time. That is why the simulated duty cycles in Table 1 all fall within this range. Additionally, the gaps in the projection data of ngCBCT are directly influenced by the breathing period; shorter breathing intervals result in smaller gaps, while longer intervals lead to larger gaps, which can be as wide as 80–100 projections, corresponding to a gantry rotation of 30º to 40º. Although the respiratory gating technique allows patients to breathe freely, introducing a minimal level of breathing coaching—such as encouraging patients to take shorter, consistent breaths and avoid prolonged breath holds—could reduce the projection gaps and ease the challenges of image reconstruction, ultimately improving image quality. 5. Conclusion A novel imaging paradigm called the ngCBCT for pretreatment imaging verification in respiratory gating lung SBRT patients has been introduced. This innovation addresses the major challenges of current clinical gCBCT by substantially reducing scan time and imaging dose, thereby enhancing patient comfort and experience. More importantly, the ngCBCT technique has the potential to expand access to high-quality imaging for a broader patient population undergoing respiratory gating lung SBRT -36- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 treatments. Although ngCBCT poses greater challenges for image reconstruction compared to conventional sparse-view reconstruction—primarily due to the non-uniform distribution of projections—how the PIBR method can effectively overcome these challenges has been demonstrated. With PIBR, significant improvements in image quality were achieved in both HF and FF acquisition modes, highlighting its robustness and clinical potential. This study establishes the feasibility of ngCBCT as a viable replacement for current gCBCT techniques in clinical practice, paving the way for its future integration into respiratory gating SBRT workflows. Referring now to Figs.37–39, shown are comparison of FDK reconstruction for nonstop gated CBCT and sparse-view CBCT. The image reconstruction for ngCBCT presents significantly greater challenges compared to conventional sparse-view CBCT, primarily due to the non-uniform distribution of projections. As a result, despite having more projections, the image quality of ngCBCT is generally inferior to that of sparse-view CBCT. The sparse-view projection data was emulated by uniformly downsampling the same gCBCT projection data for patient A (HF) and patient B (FF), and the resulting FDK reconstructions are shown in FIG.35. For patient A, even the FDK image from 115 sparse-view projections are better than the FDK image from 483 ngCBCT projections in FIG.31. Similarly, for patient B, the FDK image from 62 sparse-view projections are better than the FDK image from 241 ngCBCT projections in FIG.32. E. Toward Nonstop Gated CBCT Imaging via Dual-Domain Reconstruction Network: Application in Pretreatment Setup Verification of Respiratory Gating Lung SBRT Purpose: To introduce an innovative nonstop gated CBCT technique that reduces gated CBCT scan time from several minutes to just one minute on C-arm linear accelerators and facilitates pretreatment setup verification for gating lung SBRT patients. Also present, is an efficient dual-domain reconstruction network to handle the challenging under-sampled and non-uniform projection data from this acquisition strategy. -37- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 Methods: During nonstop gated CBCT acquisition, the gantry rotates continuously without interruption while respiratory signals gate the kV beam on and off. Retrospectively retrieved was the clinical gated CBCT projections of gating lung SBRT patients and then down-sampled the projections based on respiratory signals to emulate the nonstop gated CBCT scans for this study. Our proposed dual-domain convolutional neural network (DDCNN) consists of a projection-domain network to complete the missing projection data and an image -domain network to reduce reconstruction artifacts. The DDCNN results are compared with those of the conventional FDK algorithm and an iterative-reconstruction-based residual network (IResNet) that was developed previously. Results: The FDK method fails to produce acceptable image quality due to under-sampled and non-uniform projections in nonstop gated CBCT. The proposed dual- domain network (DDCNN) outperforms the image-domain network (IResNet) in several ways: 1) much shorter reconstruction time since DDCNN does not require iterative reconstruction inputs; 2) small-size tumors and structures handled poorly by IResNet can be better reconstructed since the projection domain network in DDCNN completes missing projections; 3) fewer artifacts since the image domain network in DDCNN has better image quality inputs. Within DDCNN, the projection and image domain networks supplement each other to achieve improved image quality under a unified framework. Conclusion: Accelerated data acquisition and rapid DDCNN reconstruction offers a promising opportunity for clinical adoption of nonstop gated CBCT, which enables shorter on-table time and improved patient comfort for gating lung SBRT patients and helps increase clinical throughput. Innovation / Impact: Proposed was an accelerated gated CBCT acquisition method called nonstop gated CBCT, which takes only 1 minute on C-arm linacs (e.g., Varian’s TrueBeam®, maximal gantry speed of 6º per second). Nonstop gated CBCT allows continuous gantry rotation while gating the kV beam with patient respiratory gating signals. This technique shortens the lengthy pretreatment setup verification time in the respiratory gating SBRT treatment workflow, in which the clinically gated CBCT scan can take 2-8 minutes. This innovative imaging technique is believed to improve clinical workflow, patient experience, and treatment outcomes. The acquisition time, CT dose -38- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 index, and projection distribution for free-breathing 3D CBCT, gated CBCT, and nonstop gated CBCT are summarized in FIG.39. The biggest challenge with nonstop gated CBCT is reconstructing high-quality images from the acquired under-sampled and non-uniform projection data in real time for patient alignment. Previously studied, was an image-domain network called iterative-reconstruction-based residual network (IResNet), which requires time-consuming iterative reconstruction as input and thus impedes its clinical application. To overcome this issue, a more efficient dual-domain reconstruction network was developed, which consists of a projection-domain convolutional neural network (CNN) and an image-domain CNN in a unified framework, to achieve rapid reconstruction (under 1 minute) and superior image quality for the nonstop gated CBCT. Methods: Collected was the gated CBCT projection data of 13 patients who received gating lung SBRT treatments. This study used 34 half-fan scans (360º rotation, offset detector) and 30 full-fan scans (200º rotation, no detector offset). Since the nonstop gated CBCT is not yet available clinically, the projection data used for this study was down- sampled from the clinical scans based on patient-specific respiratory signals. The nonstop gated CBCT projections with the FDK algorithm was reconstructed, the iterative- reconstruction-based residual network (IResNet), and the newly developed dual-domain CNN (DDCNN). As illustrated in FIG.40, the dual-domain reconstruction network consists of a projection-domain CNN and an image-domain CNN. The CNNs were modified from U-Net architecture. First, the incomplete projections were preprocessed with linear interpolation and then trained in pairs with the ground truth gated CBCT projections. Then, the nonstop gated CBCT projections and the projection-domain predicted synthetic projections were reconstructed using the FDK method. Next, the images were used for the image-domain CNN training. The half-fan and full-fan geometry modes require separately trained networks. The data used for the projection-domain and image-domain training were also independent of each other (from different patients). For the projection-domain network, a 12 half-fan (9216 paired training sinograms) and 8 full-fan scans (6144 sinograms) were used for training. For the image-domain network training, a 13 half-fan (6240 paired training images with data augmentation) and 11 full-fan scans (5280 images) was used. -39- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 Both networks shared 4 half-fan and 6 full-fan scans for validation. A 4 half-fan and 4 full- fan scans were reserved for testing. The networks were trained for 50 epochs, using MSE as the loss function and NAdam as the optimizer. Key Results: An improved reconstruction was achieved that resulted in the proposed dual-domain reconstruction network. Specifically observed, were improvements in aspects such as the image texture, artifact reduction, and small tumor size. FIG.41 compares the clinically gated CBCT with various nonstop gated CBCT reconstructions: the FDK algorithm, the IResNet, and the DDCNN. The DDCNN results outperform the other methods for both half-fan (FIG.41, panel d) and full-fan (FIG.41, panel h) modes. Also, the missing projection data makes small-size tumors hard to reconstruct for IResNet. Often, these tumors are either faint or completely missing (FIG.42, panel c). In contrast, the DDCNN approach can handle small-size tumors much better. Additionally, even with iterative reconstruction as input, the large gaps in the projection data cause streak artifacts, and the IResNet can have difficulty dealing with them. With the DDCNN, what was observed from FIG.43 was that the streak artifact suppression is better. Lastly, the DDCNN is so powerful at removing artifacts that some of those in the ground truth gated CBCT images are removed or improved, as exemplified in FIG.44. F. Systems and Methods for Performing Nonstop Gated Cone Beam Computed Tomography (CBCT) Imaging and Reconstructing Images Using Projection Data from Nonstop Gated CBCT Imaging Referring now to FIG.47, depicted is a block diagram of a system 100 for processing projection data acquired via non-stop gated cone beam computed tomography (CBCT). In brief overview, the system 100 may include at least one data processing system 105, at least one cone-beam computed tomography (CBCT) scanner 110, at least one scanner controller 115, and at least one administrative device 120, among others, communicatively coupled with one another, via at least one network 125. The data processing system 105 may include at least one dataset indexer 130, at least one model trainer 132, at least one model applier 134, at least one output evaluator 136, at least one machine learning (ML) architecture 138, and at least one database 140, among others. The ML architecture 138 may include at least one projection synthesizer 142, at least one -40- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 domain converter 144, and at least one image reconstructor 146, among others. The CBCT scanner 110 may include at least one gantry 150 with at least one beam emitter 152 and at least one detector 154, and at least one linear accelerator 156, among others. The scanner controller 115 may include at least one gantry guider 160, at least one motion detector 162, at least one beam manager 164, and at least one data acquirer 166, among others. Each of the components of the system 100 can be implemented using the computing system as described in Section D. The system 100 may be used to implement the functionalities detailed herein in Sections A–E. In further detail, the data processing system 105 can be any computing device comprising one or more processors coupled with memory and software capable of performing the various processes and tasks described herein. The data processing system 105 may be housed within a computing system (e.g., laptop, PC, smart device) or within a server group (e.g., a data center, a branch office, or a server site), and include instructions. The data processing system 105 may be associated with an entity to handle processing of raw projection data to reconstruct CT images. The data processing system 105 may be in communication with the CBCT scanner 110, the scanner controller 115, administrative device 120, and the database 140, among others. In some embodiments, the data processing system 105 may include the scanner controller 115, and perform the functionalities ascribed to the scanner controller 115 as detailed herein. The data processing system 105 may have one or more components, modules, processes, and threads to perform the various processes and tasks described herein. On the data processing system 105, the dataset indexer 130 may retrieve, identify, or otherwise obtain projection data from the CBCT scanner 110 acquired in accordance with non-stop gated CBCT (e.g., including gaps in projection acquisitions in the projection data). The model trainer 132 may initialize, train, or otherwise establish the ML architecture 138. The model applier 134 may provide, input, or otherwise apply the projection data to the ML architecture 138 to generate a reconstructed CT image. The output evaluator 136 may generate and provide an output based on the reconstructed CT image. The database 140 may store and maintain any data generated in the operation of the data processing system -41- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 105, the CBCT scanner 110, the scanner controller 115, and the administrative device 120, among others. The ML architecture 138 may be any type of artificial intelligence (AI) algorithm or ML model to generate CT images from raw projection data. The ML architecture 138 may include a deep learning artificial neural network (ANN) based architecture, such as a transformer architecture, an encoder-decoder model with one convolution neural networks (CNN), a diffusion model, or a recurrent neural network (RNN) among others. The ML architecture 138 may, for example, be an instance of the model architecture in accordance with the description in herein in conjunction with FIGs.17 or 36. In general, the ML architecture 138 may include inputs and outputs related to one another via a set of weights. The set of weights may be in accordance with the AI algorithm or ML model (e.g., encoder-decoder model with one convolution neural networks) used to implement the ML architecture 138. The input may include the projection data acquired in accordance with the non-stop gated CBCT. In some embodiments, the set of weights of the ML architecture 138 may be distributed or arranged across the projection synthesizer 142 and the image reconstructor 146, among others. The projection synthesizer 142 may include deep learning artificial neural network (ANN) based architecture, such as a transformer architecture, an encoder- decoder model with one convolution neural networks (CNN), a diffusion model, or a recurrent neural network (RNN) among others. The projection synthesizer 142 may include inputs and outputs related to one another via a set of weights. The input may include the raw projection data acquired in accordance with non-stop gated CBCT. The output may include the raw projection data in higher quality (e.g., lacking gaps in projections). The domain converter 144 of the ML architecture 138 may include any number of functions to convert or transform the projection data to an initial CT image. The function may include, for example, Feldkamp-Davis-Kress (FDK) algorithm, filtered back projection (FBP), algebraic reconstruction technique (ART), simultaneous iterative reconstruction technique (SIRT), and ordered subset expectation maximization (OSEM), among others. The image reconstructor 146 may include deep learning artificial neural network (ANN) based architecture, such as a transformer architecture, an encoder-decoder -42- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 model with one convolution neural networks (CNN), a diffusion model, or a recurrent neural network (RNN) among others. The image reconstructor 146 may include inputs and outputs related to one another via a set of weights. The input may include the initial CT image as generated by the ML architecture 138. The output may include the raw projection data in higher quality (e.g., lacking gaps in projections). In some embodiments, the ML architecture 138 set of weights of the ML architecture 138 may be distributed or arranged across an encoder and a decoder. The encoder and the decoder may be part of a deep learning artificial neural network (ANN) based architecture (e.g., U-Net architecture). The encoder may have the raw projection data as the input. The encoder may produce or generate a set of embeddings with features as the output. The decoder may have the set of embeddings as the input. The decoder may produce or generate the reconstructed CT image from the set of embeddings. The CBCT scanner 110 (sometimes referred herein as a CT scanner or generally referred to as an imaging device) may be any device to perform a CBCT imaging of a volume of a subject. The imaging may be performed in accordance with non-stop gated CBCT, in which the CBCT scanner 110 selectively activates the radiation beam (e.g., X-ray beam) based on motion of the volume while the gantry 150 continues to spin during the acquisition time period. The CBCT scanner 110 may be associated with an entity, such as a clinic, hospital, or site for imaging subjects under evaluation. The CBCT scanner 110 may be in communication with the data processing system 105, the scanner controller 115, administrative device 120, and the database 140, among others. The CBCT scanner 110 may administer, delivery, or provide radiotherapy to the subject. The CBCT scanner 110 may include any number of components, including the gantry 150, the beam emitter 152, the detector 154, and the linear accelerator 156, as shown, among others. In the CBCT scanner 110, the gantry 150 may include a ring or a cylindrical structure within which the subject is positioned, situated, or placed. The gantry 150 may contain, house, or otherwise include the beam emitter 152 and the detector 154. The gantry 150 may rotate at least partially about a volume of the subject while the CT imaging is performed. The beam emitter 152 may output, generate, or otherwise produce a radiation beam (e.g., in the form of cone-shaped X-rays) to image within the volume of the -43- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 subject. The detector 154 may accept, obtain, or receive the radiation beam produced by the beam emitter 152 and traversing through the volume of the subject. The detector 154 may be situated, positioned, or otherwise disposed in the gantry 150 opposite of the beam emitter 152. In addition, the linear accelerator 156 can provide, deliver, or administer radiotherapy to a subject. The linear accelerator 156 may include a particle source to provide ions (e.g., electrons); an accelerating waveguide (e.g., resonating cavities) to accelerate beams to high energy; one or more bending magnets to direct ions toward a target; a target to convert the ions to the radiation beam; and a collimation system to shape and modulate the radiation beam used for the radiotherapy, among others. In some embodiments, the linear accelerator 156 may also be integrated or included in the gantry 150. In some embodiments, the linear accelerator 156 may be on another portion of the CBCT scanner 110. The radiotherapy provided by the linear accelerator 156 may include, for example, at least one of a stereotactic body radiation therapy (SBRT), an intensity- modulated radiation therapy (IMRT), a volumetric modulated arc therapy (VMAT), a conformal radiation therapy (CRT), or a proton beam therapy (PBT), among others. The scanner controller 115 may be any computing device comprising one or more processors coupled with memory and software and capable of performing the various processes and tasks described herein. The scanner controller 115 may be in communication with the data processing system 105, the CBCT scanner 110, and the administrative device 120, via the network 125. The scanner controller 115 may manage, administer, or otherwise control various operations of the CBCT scanner 110 and its components. The scanner controller 115 may have one or more components, modules, processes, and threads to perform the various processes and tasks described herein. On the scanner controller 115, the gantry guider 160 may administer the rotation of the gantry 150 about the volume of the subject. The motion detector 162 may identify a motion or oscillation of the volume of the subject. The beam manager 164 may control activation or deactivation of the beam emitter 152 based on the motion. The data acquirer 166 may generate projection data based on readings from the detector 154. In some embodiments, the scanner controller 115 may be part of the data processing system 105, or vice-versa. -44- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 The administrative device 120 may (sometimes herein referred to as a client device, a client, or an end user computing device) may be any computing device comprising one or more processors coupled with memory and software and capable of performing the various processes and tasks described herein. The administrative device 120 may be in communication with the data processing system 105, the CBCT scanner 110, and the scanner controller 115, via the network 125. The administrative device 120 may have at least one display. The administrative device 120 may be associated with an entity (e.g., a clinician) examining the CT scan of a subject or administering radiotherapy to the subject. The display may present information about the subject provided by the data processing system 105. Referring now to FIG.48, depicted is a block diagram of a process 200 for acquiring projection data acquired via non-stop gated cone beam computed tomography (CBCT). For the process 200, the CBCT scanner 110 may be used to image at least one volume 208 of at least one subject 205. The subject 205 may be a human or animal subject. The subject 205 may situated, placed, or otherwise arranged along a surface (e.g., a top surface) of an examination table 210 within the gantry 150. The volume 208 may correspond to a three-dimensional region or section within the subject 205 to be imaged using the CBCT scanner 110. The subject 205 may be at risk of or diagnosed with a cancer associated with an organ within the volume 208. The volume 208 may include the organ to be evaluated for cancer or radiotherapy. The cancer may affect the organ in an abdominal or thoracic region (e.g., chest) of the subject 205. The cancer may include, for example, at least one of lung cancer, esophageal cancer, stomach cancer, colorectal cancer, liver cancer, pancreatic cancer, small intestine cancer, kidney cancer, prostate cancer, testicular cancer, ovarian cancer, uterine cancer, or cervical cancer, among others. The organ may include, for example, lung, esophagus, stomach, colon, rectum, large intestine, small intestine, liver, pancreas, kidney, prostate, testes, ovaries, uterus, or cervix, among others. The subject 205 may be imaged as part of pre-treatment process to verify the positioning of the subject 205, prior to administration of radiotherapy for the cancer. -45- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 To image, the gantry guider 160 on the scanner controller 115 may trigger, configure, or otherwise cause the CBCT scanner 110 to turn, spin, or otherwise rotate the gantry 150 at least partially about the volume 208 of the subject 205. The gantry guider 160 may convey, transmit, or otherwise send a control signal to the CBCT scanner 110 to rotate the gantry 150. The gantry 150 may rotate along at least one arc 215 about the volume 208 of the subject 205 during a time window. The time window may correspond to a time during which the CT imaging is performed or the gantry 150 rotates about the volume 208 for performing imaging. The time window may range between 45 seconds to 3 minutes. Once initiated, the gantry 150 may continue to rotate without any interruptions or stops throughout the duration of the time window. The arc 215 along which the gantry 150 turns, revolves, or rotates may range between 140 degrees to 360°, such as 140°, 160°, 180°, 200°, 220°, 240°, 260°, 280°, 300°, 320°, 330°, 340°, or 360°, among others. In addition, the detector 154 of the CBCT scanner 110 may be activated for measuring or accepting radiation beams from the beam emitter 152, during the acquisition time window. The motion detector 162 on the scanner controller 115 may determine, gauge, or identify a motion 212 of the volume 208 of the subject 205 in relation to the examination table 210. The motion 212 may correspond to a relative distance between the surface (e.g., top surface) of the examination table 210 and a distal surface (e.g., top, outer surface) of the volume 208 of the subject 205. As the gantry 150 rotates along the arc 215, the motion detector 162 may measure or monitor the motion 212 of the volume 208 of the subject 205 during the time window. The volume 208 may contract or expand, depending on whether the subject 205 is inhaling, exhaling, or any other number of causes. As a result of the contraction and expansion, the motion 212 of the volume 208 may fluctuate or oscillate through the duration of the time window. To identify the motion 212, the motion detector 162 may use a position of at least one structure 220. The structure 220 may arranged, positioned, or otherwise situated externally on the side (e.g., along the top as depicted) of the subject 205. The structure 220 may function as a tool (e.g., an external surrogate) to keep track of the motion 212 of the volume 208 of the subject 205. The structure 220 may be any type of component, such as a block (e.g., as depicted), an instrument, or a marker. For instance, the structure 220 may be -46- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 a Linear Accelerator (LINAC) block used to shape and direct the radiation beam produced by a LINAC as part of the radiotherapy administered to the subject 205. In some embodiments, the structure 220 may be part of an external breath monitor. The motion detector 162 may track the position of the structure 220 relative to the surface (e.g., top side) of the examination table 210 using image data from an external camera or position data from within the structure 220. In some embodiments, the motion detector 162 may use image data of the subject 205 on the external table 210 to identify the motion 212. The image data may be obtained via a camera (e.g., an optical camera) positioned along a side (e.g., on frontal or coronal plane) of the subject 205. The image data may be used to identify the motion 212, with or without the positioning of the structure 220 on the side of the subject 205. The image data may include one or more image frames (e.g., as part of a video) taken by the camera. The motion detector 162 may use any number of computer vision techniques, such as feature points tracking, inter-frame difference algorithm, motion tracking algorithm, image segmentation, or objection detection algorithm, among others. Using the computer vision technique on the image data, the motion detector 162 may determine the motion 212 of the volume 208 of the subject 205 relative to the examination table 210. The motion detector 162 may calculate, determine, or otherwise generate at least one gating threshold 225 for the motion 212. The gating threshold 225 may specify, identify, or define a range of values (or gating window) for the motion 212, within which the volume 208 is identified as being in a first state 230A and outside of which the volume 208 is identified as being in a second state 230B. The gating threshold 225 may define the subject 205 or the volume 208 as switching between the first state 230A and the second state 230B. The gating window may correspond to the range of values of the motion 212 during which the beam emitter 152 is to be activated to image the volume 208 of the subject 205. For example, the gating window may correspond to when the subject 205 is inhaling, with the objective of ensuring the volume 208 of the subject 205 in end product CT image is consistent. In some embodiments, the motion detector 162 may generate the gating threshold 225, prior to the time window for imaging. The motion detector 162 may track -47- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 the motion 212 of the subject 205 for a period of time, before the start of imaging. The period of time may correspond to one or more breathing cycles by the subject 205. The tracking of the motion 212 may be used to generate time-series data of the relative distance between the top side of the volume 208 relative to the top surface of the examination table 210. Using the tracked motion 212, the motion detector 162 may determine the gating threshold 225. For example, the motion detector 162 may calculate the gating threshold 225 as a function (e.g., set percentage) of the maximum or minimum amplitude of the motion 212. In some embodiments, the motion detector 162 may generate multiple gating thresholds 225. For instance, one gating threshold 225 may define a minimum value of the motion 212 (or relative distance) for the gating window and another gating threshold 225 may define a maximum value of the motion for the gating window. In some embodiments, the motion detector 162 may obtain or identify the gating threshold 225 pre-defined for the CBCT scanner 110. For example, the gating threshold 225 may be pre-defined as a fixed value for the relative distance between the top surface of the examination table 210, independent of the specific subject 205. Based on the motion 212 of the volume 208 during the time window, the motion detector 162 may identify or determine whether the volume 208 of the subject 205 is in the first state 230A or the second state 230B. As the gantry 150 rotates along the arc 215, the motion detector 162 may continue to determine whether the volume 208 is in the first state 230A or the second state 230B. To determine the state of the volume 208, the motion detector 162 may compare the motion 212 with the gating threshold 225. If the motion 212 satisfies (e.g., is less than or within) the gating threshold 225, the motion detector 162 may determine that the volume 208 is in the first state 230A instead of the second state 230B. In some embodiments, if the motion 212 is within the gating window defined by the one or more gating thresholds 225, the motion detector 162 may determine that the volume 208 is in the first state 230A instead of the second state 230B. Conversely, if the motion 212 does not satisfy (e.g., more than or outside) the gating threshold 225, the motion detector 162 may determine that the volume 208 is in the second state 230B instead of the first state 230A. In some embodiments, if the motion 212 is within the gating window defined by the one or more gating thresholds 225, the motion detector 162 may determine that the volume 208 is in the first state 230A instead of the second state 230B. -48- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 The beam manager 164 on the scanner controller 115 may configure or control the activation or deactivation of the beam emitter 152 in accordance with whether the volume 208 of the subject 205 is in the first state 230A or the second state 230B. When the volume 208 is determined to be in the first state 230A instead of the second state 230B, the beam manager 164 may cause the beam emitter 152 to be enabled or activated (e.g., by sending a command signal to activate). Upon activation, the beam emitter 152 may emit, output, or otherwise produce a radiation beam to image the volume 208 of the subject 205. The radiation beam may be produced in accordance with CBCT, and may have a cone- shaped beam geometry. The radiation beam may be produced and outputted by the beam emitter 152 in a continuous or pulsed manner while activated. In some embodiments, the radiation beam may be produced in accordance with a full-fan configuration. The full-fan configuration may correspond to the radiation beam being fully collimated with the detector 154. In some embodiments, the radiation beam may be produced in accordance with a half- fan configuration. The half-fan configuration may correspond to the radiation beam being partially collimated with the detector 154. On the other hand, when the volume 208 is determined to be in the second state 230B instead of the first state 230A, the beam manager 164 may cause the beam emitter 152 to be disabled or deactivated (e.g., by sending a command signal to deactivate). Upon deactivation, the beam emitter 152 may cease or terminate production of the radiation beam for imaging the volume 208 of the subject 205. As the beam emitter 152 is activated or deactivated, the gantry 150 of the CBCT scanner 110 may continue to rotate along one or more segments 235A–N (hereinafter generally referred to as segments 235) along the arc 215. Each segment 235 may correspond to a respective radial portion of the arc 215, where the beam emitter 152 is either activated or deactivated. For example, while the beam emitter 152 is activated (e.g., when the volume 208 is in the first state 230A), the gantry 150 of the CBCT scanner 110 may continue to rotate along one segment 235 (e.g., the segment 235A) of the arc 215 about the volume 208. Conversely, while the beam emitter 152 is deactivated (e.g., when the volume 208 is in the second state 230B), the gantry 150 of the CBCT scanner 110 may continue to rotate along another segment 235 (e.g., the segment 235B) of the arc 215 about the volume 208. If the beam emitter 152 is subsequently activated (e.g., as a result of to the volume 208 switching from the second state 230B to the first state 230A), the gantry 150 of the CBCT -49- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 scanner 110 may continue to rotate along yet another segment (e.g., segment 235C) of the arc 215 about the volume 208. In conjunction, the detector 154 may remain activated, and may accept, obtain, or receive the radiation beams traversing through the volume 208 of the subject 205, when the radiation beams are produced by the beam emitter 152. The data acquirer 166 on the scanner controller 115 may create, produce, or otherwise generate at least one projection dataset 240 (sometimes referred herein as raw projection data) corresponding to the imaging of the volume 208 of the subject 205. To generate the projection dataset 240, the data acquirer 166 may obtain, accept or otherwise acquire a set of projections 245A–N (hereinafter generally referred to as projections 245) via the detector 154. The projections 245 may correspond to the radiation beam produced by the beam emitter 152 and traversing through the volume 208 of the subject 205. For example, each projection 245 may capture the attention of the radiation beam (e.g., X-ray beam with cone beam geometry) through the volume of the subject 205. The data acquirer 166 may receive the set of projections 245 via the detector 154, while the beam emitter 152 is activated. The projections 245 may by extension correspond to the segment 235 (e.g., the segment 235A) when the beam emitter 152 is activated and the volume 208 is in the first state 230A within the gating window. With receipt, the data acquirer 166 may add or inject data corresponding to the set of projections 245 to the projection data 240. In generating the projection dataset 240, the data acquirer 166 may insert or add one or more gaps 250A–N (hereinafter generally referred to as gaps 250). Each gap 250 may correspond to a corresponding segment 235 (e.g., the segment 235B) during which the beam emitter 152 is deactivated. By extension, each gap 250 may correspond to the segment 235 (e.g., the segment 235B) when the beam emitter 152 is deactivated and the volume 208 is in the first state 230B outside the gating window. The data acquirer 166 may obtain, accept, or otherwise receive a null reading (corresponding to no radiation beam) from the detector 154, when the beam emitter 152 is deactivated. The data acquirer 166 may add or insert the null reading as a corresponding gap 250 into the projection dataset 240. As the volume 208 continues to fluctuate between the first state 230A and the second state 230B, the data acquirer 166 may add successive sets of projections 245 and gaps 250 in an alternating manner. As shown in the example, the projection dataset 240 may include -50- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 one set of projections 245, followed by a gap 250, and then another set of projections 245. Within the projection dataset 240, the set of projections 245 may be interleaved with the set of gaps 250. As the gantry 150 rotates along the arc 215, the gantry guider 160 may identify or determine whether the gantry 150 has completed the rotation of the arc 215. If the gantry 150 has not completed the rotation, the gantry guider 160 may cause the CBCT scanner 110 to continue to rotate the gantry 150 further along the arc 215, and the processes detailed herein may be repeated. For instance, the motion detector 162 may determine the volume 208 of the subject 205 as switching back to the first state 230A form the second state 230B. With this determination, the beam manager 165 may cause the beam emitter 152 to be activated to produce a radiation beam to image the volume 208 of the subject 205, while another segment 235. The data acquirer 166 may acquire another set of projections 245 corresponding to the radiation beam traversing through the volume 208 of the subject 205. The data acquirer 166 may also add to and generate the projection dataset 240 to add the set of projections 245. This may be repeated any number of times until the completion of the arc 215. In contrast, if the gantry 150 has completed the rotation, the gantry guider 160 may cause the CBCT scanner 110 to cease or terminate rotation of the gantry 150. With the completion of the rotation of the arc 215, the data acquirer 166 may store and maintain an association between the subject 205 and the projection dataset 240 using one or more data structures on the database 140. The data structure may include, for example, a linked list, a tree, an array, a matrix, a table, a queue, a stack, or a data class object, among others. For instance, the data acquirer 166 may store the projection dataset 240 as a file in a Digital Imaging and Communications in Medicine (DICOM) format, and a data structure may index the file with an identifier for the subject 205. In some embodiments, the data acquirer 166 may send, transmit, or otherwise provide the projection dataset 240 to the data processing system 105 for processing using the ML architecture 138. By providing the projection dataset 240, a reconstructed CT image may be obtained (e.g., by the scanner controller 115 or the data processing system 105). Referring now to FIG.49, depicted is a block diagram of a process 300 for training a machine learning (ML) architectures for reconstructing CT images. Under the -51- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 process 300, the dataset indexer 130 on the data processing system 105 may retrieve, receive, or otherwise identify training data 305. The training data 305 may be stored and maintained on the database 140. The training data 305 may be used to initialize, train, and establish the ML architecture 138, including its subcomponents, such as the projection synthesizer 142 and the image reconstructor 146. The training data 305 may identify or include a set of examples with which the ML architecture 138 is to perform learning (e.g., supervised or weakly supervised learning). Each example of the training data 305 may be for at least one sample subject. The sample subject may differ from the subject 205. In the training data 305, each example may identify or include at least one sample projection dataset 340. The sample projection dataset 340 may be similar to the projection dataset 240 generated by the CBCT scanner 110. The sample projection dataset 340 may be used as an input to the ML architecture 138, such as to the projection synthesizer 142. In general, the sample projection dataset 340 may be a set of projections corresponding to at least a portion of an arc about a sample volume of the sample subject over a sample time window. The sample projection dataset 340 may be acquired using the same beam collimation configuration as the projection dataset 240, such as the half-fan configuration or the full-fan configuration. The arc about which the gantry of the CT scanner used to generate the sample projection dataset 340 may be the same or substantially similar (e.g., within 80–90%) as the range of the arc 215 used in generating the projection dataset 240, and may range between 140 and 360 degrees. The sample time window for the sample projection dataset 340 may be the same or substantially similar (e.g., within 80– 90%) as the time window used to acquire the projection data 240. In some embodiments, the sample projection dataset 340 may be a non-stop gated sample projection dataset, acquired in accordance with non-stop gated CBCT as detailed herein. The sample projection dataset 340 may include the sample set of projections corresponding to segments of the arc. Each segment for the projections may correspond to when the sample volume of the sample is in the first state 235A (e.g., within the gating window). In addition, the sample projection dataset 340 may include a set of gaps corresponding to other segments (e.g., interleaved or adjacent) of the arc. Each segment for a corresponding gap may correspond to when the sample volume of the sample -52- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 is in the first state 235B (e.g., outside the gating window). In some embodiments, the sample projection dataset 340 may be a modification of non-stop gated sample projection dataset, acquired in accordance with non-stop gated CBCT. The modification may be an application of interpolation to remove the gaps using the projections in the original set of projections. In some embodiments, the sample projection dataset 340 may be a gated sample projection dataset with stops. The sample projection dataset 340 may include the sample set of projections throughout the data, corresponding to an entirety of the sample arc during the sample time window. The sample projection dataset 340 may lack any gaps. The training data 305 may include examples with non-stop gated CBCT sample data, and other examples with gated CBCT sample data with stops. In some embodiments, at least one example may identify or include at least one expected projection dataset 340’. The expected projection dataset 340’ may correspond to the expected output for the projection synthesizer 142 in the ML architecture 138, when provided with the sample projection dataset 430 as input. The expected projection dataset 340’ may correspond to the sample projection dataset 430 with higher quality (e.g., higher signal-to-noise (SNR), higher contrast-to-noise (CNR), sharpness, point spread, less artifacts, lower number of streaks). For example, the expected projection dataset 340’ may be of sample volume of the same sample subject, acquired in accordance with gated CBCT imaging with stops. The expected projection dataset 340’ may include the sample set of projections throughout the data, corresponding to an entirety of the sample arc during the sample time window. The expected projection dataset 340’ may lack any gaps. In some embodiments, the examples of the training data 305 may lack the expected projection dataset 340’. Each example may also identify or include at least one computed tomography (CT) image 345. The CT image 345 may correspond to the sample volume of the sample subject. The CT image 345 may be a three-dimensional cross-sectional representation of the sample volume of the subject that is imaged. The CT image 345 may be the expected output of the image reconstructor 146 when provided with the expected projection dataset 340’ as input. The CT image 345 may also be the expected output of the oval ML architecture 138 when provided with the sample projection dataset 340 as input. -53- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 The CT image 345 may be generated via a reconstruction algorithm using the sample projection dataset 340 or the expected projection dataset 340’. The reconstruction algorithm used in generating the CT image 345 may include, for example, Feldkamp-Davis-Kress (FDK) algorithm, filtered back projection (FBP), algebraic reconstruction technique (ART), simultaneous iterative reconstruction technique (SIRT), and ordered subset expectation maximization (OSEM), among others. With the identification, the model trainer 132 on the data processing system 105 may provide, input, or otherwise apply the ML architecture 138 to the sample projection dataset 340. In some embodiments, the model trainer 132 may create, produce, or otherwise generate an interpolated projection dataset using the sample projection dataset 340, in accordance with an interpolator. The interpolator may include a function to estimate values for projections in the gaps within the sample projection dataset 340, using the values of the projections already included in the sample projection dataset 340. The interpolation function may include, for example, a linear interpolation, a bilinear interpolation, spline interpolation, bicubic interpolation, nearest-neighbor interpolation, frequency-domain based interpolation, or radon-space interpolation, among others. With the generation of the interpolated projection dataset, the model trainer 132 may apply the interpolated projection dataset to the ML architecture 138. In applying, the model trainer 132 input, feed, or otherwise provide the sample projection dataset 340 (or the interpolated projection dataset) to the projection synthesizer 142 of the ML architecture 138. The projection synthesizer 142 may process the input sample projection dataset 340 in accordance with the set of weights of the projection synthesizer 142. From processing the input sample projection dataset 340, the projection synthesizer 142 may output, produce, or otherwise generate at least one projection dataset 340”. The projection dataset 340” may correspond to an instance of the sample projection dataset 340 (or interpolated sample projection dataset), with higher quality (e.g., higher signal-to-noise (SNR), higher contrast-to-noise (CNR), sharpness, point spread, less artifacts, or lower number of streaks). With the generation, the model trainer 132 may input, feed, or otherwise provide the projection dataset 340” to the domain converter 144. The domain converter 144 -54- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 may apply a function (e.g., FDK algorithm or FBP) to convert or transform the projection dataset 340” to output, produce, or otherwise generate at least one initial CT image 345’. The initial CT image 345’ may be the initial estimate of the CT imaging of the sample volume of the sample subject. The initial CT image 345’ may be of poorer quality (e.g., lower SNR, lower CNR, lower sharpness, more artifacts, or higher number of streaks). The model trainer 132 may input, feed, or otherwise provide the initial CT image 345’ to the image reconstructor 146 of the ML architecture 138. The image reconstructor 146 may process the input sample projection dataset 340 in accordance with the set of weights of the image reconstructor 146. From processing the input initial CT image 345’, the image reconstructor 146 may output, produce, or otherwise generate at least one reconstructed CT image 345”. The reconstructed CT image 345” may correspond to the sample volume of the sample subject. The reconstructed CT image 345” may correspond to three-dimensional cross-sectional representation of the sample volume of the subject that is imaged. The reconstructed CT image 345” may correspond to the output of the overall ML architecture 138. In some embodiments, the model trainer 132 may input, feed, or otherwise input the sample projection dataset 340 to the encoder of the ML architecture 138. The encoder may process the input sample projection dataset 340 in accordance with the set of weights of the encoder. From processing the input sample projection dataset 340, the encoder may output, produce, or otherwise generate a set of embeddings corresponding to relevant features in the input. The model trainer 132 may input, feed, or otherwise input the set of embeddings to the decoder of the ML architecture 138. The decoder may process the input set of embeddings in accordance with the set of weights of the decoder. From processing the set of embeddings, the decoder may output, produce, or otherwise generate the reconstructed CT image 345”. With the generation of outputs in the ML architecture 138, the model trainer 132 may compare the outputs to the expected outputs from the training data 305. Based on the comparisons, the model trainer 135 may calculate, generate, or otherwise determine one or more loss metrics 310A or 310B (hereinafter generally referred to as loss metrics 310). The loss metric 310 may be calculated in accordance with any number of loss functions, -55- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 such as a norm loss (e.g., L1 or L2), mean squared error (MSE), a quadratic loss, a cross- entropy loss, and a Huber loss, among others. In some embodiments, the loss metric 310 may be calculated in accordance with a similarity function, such as a Dice score, Jaccard index, correlation coefficient, or Minkowski distance, among others. The model trainer 132 may compare the projection dataset 340” generated by the projection synthesizer 142 with the expected projection datasets 340’. Based on the comparison, the model trainer 132 may determine the loss metric 310 (e.g., the loss metric 310A as depicted). The loss metric 310 may indicate a degree of deviation between the projection dataset 340” outputted by the projection synthesizer 142 with the expected projection dataset 340’. The model trainer 132 may also compare the reconstructed CT image 345” generated by the image reconstructor 146 with the CT image 345 of the training data 305. In some embodiments, the model trainer 132 may compare the reconstructed CT image 345” generated by the decoder in the ML architecture 138 with the CT image 345. Based on the comparison, the model trainer 132 may determine the loss metric 310 (e.g., the loss metric 310B as depicted). The loss metric 310 may indicate a degree of deviation between the reconstructed CT image 345” outputted by the expected CT image 345 as identified in the example of the training data 305. Based on the comparisons, the model trainer 132 may modify, change, or otherwise update one or more weights of the ML architecture 138 or its components, such as the projection synthesizer 142, the image reconstructor 146, encoder, or decoder, among other. Using the loss metrics 310, the model trainer 132 may update one or more the set of weights arranged across the ML architecture 138. In some embodiments, the model trainer 132 may update one or more of the set of weights of the ML architecture 138 based on one or more of the comparison between the projection dataset 340’ and the projection dataset 340 and the comparison between the reconstructed CT image 345” and the CT image 345, among others. In some embodiments, the model trainer 132 may update one or more of the set of weights of the projection synthesizer 142 using the loss metric 310 (e.g., the loss metric 310A) determined for the projection synthesizer 142. In some embodiments, the model trainer 132 may update one or more of the set of weights of the image reconstructor -56- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 146 using the loss metric 310 (e.g., the loss metric 310B) determined for the image reconstructor 146. In some embodiments, the model trainer 132 may update one or more of the set of weights of the overall ML architecture 138 (e.g., the projection synthesizer 142 and the image reconstructor 146 or the encoder and decoder) using the loss metric 310. In some embodiments, the model trainer 132 may omit end-to-end training of the overall ML architecture 138 using the loss metrics 310. The updating of the weights of the ML architecture 138 may be in accordance with an optimization function (or an objective function). The optimization function may define one or more learning rates or parameters at which the weights of the ML architecture 138 are to be updated. The optimization function may include, for example, adaptive moment estimation (Adam), Adam with weight decay, or stochastic gradient descent (SGD), among others. For example, the weights of the projection synthesizer 142, the image reconstructor 146, the encoder, or the decoder in the ML architecture 138 may be modified or updated using the respective loss metrics 310 in accordance with the respective objective function (e.g., Adam). By updating the weights, the model trainer 132 may further train the ML architecture 138. The training may be iteratively repeated using the examples of the training data 305 until a convergence condition for the ML architecture 138 to complete the training and establishment of the ML architecture 138. Referring now to FIG.50, depicted is a block diagram of a process 400 for reconstructing CT images using ML architectures. Under the process 400, the dataset indexer 130 may retrieve, identify, or otherwise receive the projection dataset 240. In some embodiments, the dataset indexer 130 may receive the projection dataset 240 from the CBCT scanner 110 or the scanner controller 115. In some embodiments, the dataset indexer 130 may access the database 140 to retrieve, obtain, or otherwise identify the projection dataset 240. As discussed above, the projection dataset 240 may be generated by the CBCT scanner 110 as the gantry 150 rotates along the arc 215 at least partially about the volume 208 of the subject 205 during the acquisition time window. The projection dataset 240 may include the set of projections 245 corresponding to segments 235 of the arc 215, with each segment 235 corresponding to when the volume 208 of the subject 205 is in the first state -57- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 230A (e.g., within the gating window). In addition, the projection dataset 240 may include the set of gaps 250 corresponding to other segments 235 of the arc 215, with each segment 235 corresponding to when the volume 208 of the subject 205 is in the first state 230B (e.g., outside the gating window). With the identification, the model applier 134 on the data processing system 105 may provide, input, or otherwise apply the ML architecture 138 to the projection dataset 240. In some embodiments, the model applier 134 may create, produce, or otherwise generate an interpolated projection dataset using the projection dataset 240, in accordance with an interpolator. The interpolator may include a function to estimate values for projections in the gaps within the projection dataset 240, using the values of the projections already included in the projection dataset 240. The interpolation function may include, for example, a linear interpolation, a bilinear interpolation, spline interpolation, bicubic interpolation, nearest-neighbor interpolation, frequency-domain based interpolation, or radon-space interpolation, among others. With the generation of the interpolated projection dataset, the model applier 134 may apply the interpolated projection dataset to the ML architecture 138. In applying, the model applier 134 may input, feed, or otherwise provide the projection dataset 240 (or the interpolated projection dataset) to the projection synthesizer 142 of the ML architecture 138. The projection synthesizer 142 may process the input projection dataset 240 in accordance with the set of weights of the projection synthesizer 142. From processing the input projection dataset 240, the projection synthesizer 142 may output, produce, or otherwise generate at least one projection dataset 240’. The projection dataset 240’ may correspond to an instance of the projection dataset 240 (or interpolated projection dataset), with higher quality (e.g., higher signal-to-noise (SNR), higher contrast- to-noise (CNR), sharpness, point spread, less artifacts, or lower number of streaks). With the generation, the model applier 134 may input, feed, or otherwise provide the projection dataset 240’ to the domain converter 144. The domain converter 144 may apply a function (e.g., FDK algorithm or FBP) to convert or transform the projection dataset 240’ to output, produce, or otherwise generate at least one initial CT image 405. The initial CT image 405 may be the initial estimate of the CT imaging of the volume 208 -58- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 of the subject 204. The initial CT image 405 may be of poorer quality (e.g., lower SNR, lower CNR, lower sharpness, more artifacts, or higher number of streaks). The model applier 134 may input, feed, or otherwise provide the initial CT image 405 to the image reconstructor 146 of the ML architecture 138. The image reconstructor 146 may process the input projection dataset 240 in accordance with the set of weights of the image reconstructor 146. From processing the input initial CT image 405, the image reconstructor 146 may output, produce, or otherwise generate at least one reconstructed CT image 405’. The reconstructed CT image 405’ may correspond to the volume 208 of the subject 205. The reconstructed CT image 405’ may correspond to three-dimensional cross-sectional representation of the volume 208 of the subject 205 that is imaged by the CBCT scanner 110. The reconstructed CT image 405’ may correspond to the output of the overall ML architecture 138. In some embodiments, the model applier 134 may input, feed, or otherwise input the projection dataset 240 to the encoder of the ML architecture 138. The encoder may process the input projection dataset 240 in accordance with the set of weights of the encoder. From processing the input projection dataset 240, the encoder may output, produce, or otherwise generate a set of embeddings corresponding to relevant features in the input. The model applier 134 may input, feed, or otherwise input the set of embeddings to the decoder of the ML architecture 138. The decoder may process the input set of embeddings in accordance with the set of weights of the decoder. From processing the set of embeddings, the decoder may output, produce, or otherwise generate the reconstructed CT image 405’. Referring now to FIG.51, depicted is a block diagram of a process 500 for providing outputs based on the reconstructed CT images. Under the process 500, the output evaluator 136 on the data processing system 105 may store and maintain an association between the subject 205 (e.g., using an anonymized identifier) and the reconstructed CT image 405’ generated by the ML architecture 138, using one or more data structures on the database 140. The data structure may include, for example, a linked list, a tree, an array, a matrix, a table, a queue, a stack, or a data class object, among others. For instance, the data acquirer 166 may store the projection dataset 240 as a file in a Digital Imaging and -59- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 Communications in Medicine (DICOM) format, and a data structure may index the file with an identifier for the subject 205. In addition, the output evaluator 136 may create, produce, or otherwise generate at least one output 505 based on the reconstructed CT image 405’ corresponding to the volume 208 of the subject 205. The output 505 may include information about the reconstructed CT image 405’. The information may be used to determine whether to confirm, corroborate, verify whether at least one target 510 with the volume 208 of the subject 205, as indicated in the reconstructed CT image 405’. The target 510 may correspond to one or more tumors associated with the cancer in the organ of the subject 205. The information may include other data about the subject 205, such as patient information, diagnosis, and treatment history among others. The output 505 may include instructions for presenting the information, for example, defining user interface elements of a user interface 515 to display the information. With the generation, the output evaluator 136 may send, transmit, or otherwise provide the output 505 for presentation via the user interface 515 on the administrative device 120. The administrative device 120 may retrieve, identify, or otherwise receive the output 505 from the processing system 105. With receipt, the administrative device 120 may render, display, or otherwise present the information of the output 505 via the user interface 515. For example, the administrative device 120 may render the reconstructed CT image 405’ on a user interface element in the user interface 515. The information and the CT image 405’ of the output 505 may be used by the user (e.g., a clinician or radiologist) to determine whether to provide a radiotherapy to the target 510 within the volume 208 of the subject 205. The user interface 515 may include one or more user interface elements to indicate whether the target 510 is within the volume 208 of the subject 205 corresponding to the reconstructed CT image 405’ presented via the user interface 515, and matches or substantially matches (e.g., within 80–90% similarity) a location as identified in a therapy plan. The therapy plan may define or identify a location (or region) within the organ of the subject 205 to which radiotherapy is to be applied. The therapy plan may have been previous generated by the clinician examining the subject 205, during a treatment planning phase for the cancer affecting the subject 205. -60- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 Using the interactions on the user interface 515, the administrative device 120 may create, produce, or otherwise generate at least one indication 520. When the interactions are to indicate that the target 510 is within the volume 208 and matches the location, the indication 520 may identify that the target 510 within the volume 208 matches the location identified by the therapy plan. On the other hand, when the interactions are to indicate that the target 510 within the volume 208 and does not match the location, the indication 520 may identify that the target 510 is within the volume 208 does not match the location identified by the therapy plan. With the generation, the administrative device 120 may send, transmit, or otherwise provide the indication 520 to the data processing system 105. The output evaluator 136 may in turn retrieve, identify, or otherwise receive the indication 520 from the administrative device 120. In accordance with the indication 520, the output evaluator 136 may produce or generate at least one instruction 525. When the indication 520 identifies that the target 510 within the volume 208 matches the location identified by the therapy plan, the output evaluator 136 may generate the instruction 525 to indicate that the target 510 of the volume 208 of the subject 205 is to be administered with the radiotherapy. Conversely, when the indication 520 identifies that the target 510 within the volume 208 does not match the location identified by the therapy plan, the output evaluator 136 may generate the instruction 525 to indicate that the subject 205 is to not be administered with the radiotherapy. In some embodiments, the output evaluator 136 may refrain from providing the instruction 525 to administer radiotherapy to the target 510. With the generation, the output evaluator 136 may send, transmit, or otherwise provide the instruction 525 to the administrative device 120. The administrative device 120 may retrieve, identify, or otherwise receive the instruction 525. With the receipt, the administrative device 120 may render, display, or otherwise present the instruction 525. When the instruction 525 indicates that the target 510 of the volume 208 of the subject 205 is not to be administered with the radiotherapy, the administrative device 120 may present the indication of the same via the user interface 515. In response to the presentation of the instruction 525, there may be a refraining from administration of the radiotherapy. For instance, the clinician examining the subject 205 -61- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 may opt to forego administration of the radiotherapy when the instruction 525 indicates that the target 510 is not to be administered with the radiotherapy and that the target 510 does not match the location. When the instruction 525 indicates that the target 510 of the volume 208 of the subject 205 is to be administered with the radiotherapy, the administrative device 120 may present the indication of the same via the user interface 515. In response to the presentation of the instruction 525, the target 510 within the volume 208 of subject 205 corresponding to the reconstructed CT image 405’ may be administered with radiotherapy 530 (e.g., under the guidance of the clinician examining the subject 205) in accordance with the therapy plan. The linear accelerator 156 of the CBCT scanner 110 may administer or deliver radiotherapy to the target 510 within the subject 205. For instance, having verified that the target 510 matches the location as indicated in the therapy plan, the clinician examining the subject may initiate the delivery and administration of the radiotherapy 530. The clinician can interact with the CBCT scanner 110 or a computing device coupled therewith to cause the linear accelerator 156 to initiate delivery of the radiotherapy 530. The radiotherapy 530 may include, for example, at least one of a stereotactic body radiation therapy (SBRT), an intensity-modulated radiation therapy (IMRT), a volumetric modulated arc therapy (VMAT), a conformal radiation therapy (CRT), or a proton beam therapy (PBT). The radiotherapy 530 may be administered via the linear accelerator 156 on the CBCT scanner 110 or another device. The beam from the radiotherapy 530 may at least partially coincide with the volume 208 (e.g., about the target 510). In this manner, by performing a non-stop gated CBCT technique, the CBCT scanner 110 may acquire the projection dataset 240 with a substantial reduction in scan time and imaging dosage. This may also further reduce the amount of time that the subject 205 is on the examination table 210, thus alleviating discomfort during the process of CBCT imaging, and may also decrease the likelihood that the movement (e.g., due to breathing) drifts or changes during the administration of the radiotherapy 530, especially with thoracic and abdominal cancers. From a computer perspective, the resultant projection dataset 240 from the non-stop gated CBCT may result in a much smaller file size, with null readings in the form of gaps, thereby providing savings in terms of storage. In addition, the data -62- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 processing system 105 can use the ML architecture 138 to quickly generate high-quality reconstructed CT images 405’ from the projection dataset 240, despite the gaps 250 in the data. The improvement in image quality may also provide for more precise target or tumor localization for the purposes of delivering radiotherapy 530 to the target within the organs of the subject 205. Referring now to FIG.52, depicted is a flow diagram of a method 600 of performing CBCT to acquire projection data. The method 600 may be implemented or performed by any of the components detailed herein, such as the system 100 or the system 800. Under the method 600, a computing system (e.g., the scanner controller 115) may initiate rotation of a gantry of a CT scanner about a volume of a subject along an arc (605). The computing system may identify a motion of the volume of the subject (610). The computing system may determine whether the volume is in a first state (e.g., within a gating window) or a second state (e.g., outside the gating window) (615). If the volume is determined to be in the first state, the computing system may activate a beam emitter to produce a radiation beam to image the volume (620). The computing system may acquire a set of projections corresponding to the radiation beam via a detector to add to a projection dataset (625). Conversely, if the volume is determined to be in the second state, the computing system may deactivate the beam emitter to refrain from production of the radiation beam (630). The computing system may add a gap to the projection dataset to the radiation beam (635). The computing system may determine whether the arc of rotation has completed (640). When the arc has not completed, the method may be repeated from (610). On the other hand, the computing system may store and maintain the projection dataset (645). Referring now to FIG.53, depicted is a flow diagram of a method 700 of reconstructing images using projection data from CBCT. The method 700 may be implemented or performed by any of the components detailed herein, such as the system 100 or the system 800. Under the method 700, a computing system (e.g., the data processing system 105) may receive a projection dataset of a volume of a subject (705). The computing system may apply a machine learning (ML) architecture to the projection dataset (710). Based on applying the ML architecture, the computing system may generate -63- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 a reconstructed CT image corresponding to the volume of the subject (715). The computing system may provide an output based on the reconstructed CT image (720). With the presentation of the output, the computing system may receive an indication of whether a target is within the volume corresponding to the CT image (725). The computing system may determine whether the target is verified as being within the volume (730). If the target is determined to be within the volume, the computing system may provide an instruction to administer radiotherapy (735). On the other hand, if the target is determined to be outside the volume, the computing system may provide an instruction to refrain from administration (740). G. Computing and Network Environment Various operations described herein can be implemented on computer systems. FIG.54 shows a simplified block diagram of a representative server system 800, client computing system 814, and network 826 usable to implement certain embodiments of the present disclosure. In various embodiments, server system 800 or similar systems can implement services or servers described herein or portions thereof. Client computing system 814 or similar systems can implement clients described herein. The system 100 described herein can be similar to the server system 800. Server system 800 can have a modular design that incorporates a number of modules 802 (e.g., blades in a blade server embodiment); while two modules 802 are shown, any number can be provided. Each module 802 can include processing unit(s) 804 and local storage 806. Processing unit(s) 804 can include a single processor, which can have one or more cores, or multiple processors. In some embodiments, processing unit(s) 804 can include a general-purpose primary processor as well as one or more special-purpose co- processors such as graphics processors, digital signal processors, or the like. In some embodiments, some or all processing units 804 can be implemented using customized circuits, such as application specific integrated circuits (ASICs) or field programmable gate arrays (FPGAs). In some embodiments, such integrated circuits execute instructions that are stored on the circuit itself. In other embodiments, processing unit(s) 804 can execute instructions stored in local storage 806. Any type of processors in any combination can be included in processing unit(s) 804. -64- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 Local storage 806 can include volatile storage media (e.g., DRAM, SRAM, SDRAM, or the like) and / or non-volatile storage media (e.g., magnetic or optical disk, flash memory, or the like). Storage media incorporated in local storage 806 can be fixed, removable or upgradeable as desired. Local storage 806 can be physically or logically divided into various subunits such as a system memory, a read-only memory (ROM), and a permanent storage device. The system memory can be a read-and-write memory device or a volatile read-and-write memory, such as dynamic random-access memory. The system memory can store some or all of the instructions and data that processing unit(s) 804 need at runtime. The ROM can store static data and instructions that are needed by processing unit(s) 804. The permanent storage device can be a non-volatile read-and-write memory device that can store instructions and data even when module 802 is powered down. The term “storage medium” as used herein includes any medium in which data can be stored indefinitely (subject to overwriting, electrical disturbance, power loss, or the like) and does not include carrier waves and transitory electronic signals propagating wirelessly or over wired connections. In some embodiments, local storage 806 can store one or more software programs to be executed by processing unit(s) 804, such as an operating system and / or programs implementing various server functions such as functions of the system 100 of FIG.36 or any other system described herein, or any other server(s) associated with system 100 or any other system described herein. “Software” refers generally to sequences of instructions that, when executed by processing unit(s) 804 cause server system 800 (or portions thereof) to perform various operations, thus defining one or more specific machine embodiments that execute and perform the operations of the software programs. The instructions can be stored as firmware residing in read-only memory and / or program code stored in non-volatile storage media that can be read into volatile working memory for execution by processing unit(s) 804. Software can be implemented as a single program or a collection of separate programs or program modules that interact as desired. From local storage 806 (or non-local storage described below), processing unit(s) 804 can retrieve program instructions to execute and data to process in order to execute various operations described above. -65- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 In some server systems 800, multiple modules 802 can be interconnected via a bus or other interconnect 808, forming a local area network that supports communication between modules 802 and other components of server system 800. Interconnect 808 can be implemented using various technologies including server racks, hubs, routers, etc. A wide area network (WAN) interface 810 can provide data communication capability between the local area network (interconnect 808) and the network 826, such as the Internet. Technologies can be used, including wired (e.g., Ethernet, IEEE 802.3 standards) and / or wireless technologies (e.g., Wi-Fi, IEEE 802.11 standards). In some embodiments, local storage 806 is intended to provide working memory for processing unit(s) 804, providing fast access to programs and / or data to be processed while reducing traffic on interconnect 808. Storage for larger quantities of data can be provided on the local area network by one or more mass storage subsystems 812 that can be connected to interconnect 808. Mass storage subsystem 812 can be based on magnetic, optical, semiconductor, or other data storage media. Direct attached storage, storage area networks, network-attached storage, and the like can be used. Any data stores or other collections of data described herein as being produced, consumed, or maintained by a service or server can be stored in mass storage subsystem 812. In some embodiments, additional data storage resources may be accessible via WAN interface 810 (potentially with increased latency). Server system 800 can operate in response to requests received via WAN interface 810. For example, one of modules 802 can implement a supervisory function and assign discrete tasks to other modules 802 in response to received requests. Work allocation techniques can be used. As requests are processed, results can be returned to the requester via WAN interface 810. Such operation can generally be automated. Further, in some embodiments, WAN interface 810 can connect multiple server systems 800 to each other, providing scalable systems capable of managing high volumes of activity. Other techniques for managing server systems and server farms (collections of server systems that cooperate) can be used, including dynamic resource allocation and reallocation. -66- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 Server system 800 can interact with various user-owned or user-operated devices via a wide-area network such as the Internet. An example of a user-operated device is shown in FIG.12 as client computing system 814. Client computing system 814 can be implemented, for example, as a consumer device such as a smartphone, other mobile phone, tablet computer, wearable computing device (e.g., smart watch, eyeglasses), desktop computer, laptop computer, and so on. For example, client computing system 814 can communicate via WAN interface 810. Client computing system 814 can include computer components such as processing unit(s) 816, storage device 818, network interface 820, user input device 822, and user output device 824. Client computing system 814 can be a computing device implemented in a variety of form factors, such as a desktop computer, laptop computer, tablet computer, smartphone, other mobile computing device, wearable computing device, or the like. Processing unit(s) 816 and storage device 818 can be similar to processing unit(s) 804 and local storage 806 described above. Suitable devices can be selected based on the demands to be placed on client computing system 814; for example, client computing system 814 can be implemented as a “thin” client with limited processing capability or as a high-powered computing device. Client computing system 814 can be provisioned with program code executable by processing unit(s) 816 to enable various interactions with server system 800. Network interface 820 can provide a connection to the network 826, such as a wide area network (e.g., the Internet) to which WAN interface 810 of server system 800 is also connected. In various embodiments, network interface 820 can include a wired interface (e.g., Ethernet) and / or a wireless interface implementing various RF data communication standards such as Wi-Fi, Bluetooth, or cellular data network standards (e.g., 3G, 4G, 5G, LTE, etc.). User input device 822 can include any device (or devices) via which a user can provide signals to client computing system 814. The client computing system 814 can interpret the signals as indicative of particular user requests or information. In various -67- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 embodiments, user input device 822 can include any or all of a keyboard, touch pad, touch screen, mouse or other pointing device, scroll wheel, click wheel, dial, button, switch, keypad, microphone, and so on. User output device 824 can include any device via which client computing system 814 can provide information to a user. For example, user output device 824 can include a display to display images generated by or delivered to client computing system 814. The display can incorporate various image generation technologies, e.g., a liquid crystal display (LCD), light-emitting diode (LED) including organic light-emitting diodes (OLED), projection system, cathode ray tube (CRT), or the like, together with supporting electronics (e.g., digital-to-analog or analog-to-digital converters, signal processors, or the like). Some embodiments can include a device such as a touchscreen that function as both input and output device. In some embodiments, other user output devices 824 can be provided in addition to or instead of a display. Examples include indicator lights, speakers, tactile “display” devices, printers, and so on. Some embodiments include electronic components, such as microprocessors, storage and memory that store computer program instructions in a computer-readable storage medium. Many of the features described in this specification can be implemented as processes that are specified as a set of program instructions encoded on a computer-readable storage medium. When these program instructions are executed by one or more processing units, they cause the processing unit(s) to perform various operation indicated in the program instructions. Examples of program instructions or computer code include machine code, such as is produced by a compiler, and files including higher-level code that are executed by a computer, an electronic component, or a microprocessor using an interpreter. Through suitable programming, processing unit(s) 804 and 816 can provide various functionality for server system 800 and client computing system 814, including any of the functionality described herein as being performed by a server or client, or other functionality. It will be appreciated that server system 800 and client computing system 814 are illustrative and that variations and modifications are possible. Computer systems used in connection with embodiments of the present disclosure can have other capabilities -68- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 not specifically described here. Further, while server system 800 and client computing system 814 are described with reference to particular blocks, it is to be understood that these blocks are defined for convenience of description and are not intended to imply a particular physical arrangement of component parts. For instance, different blocks can be but need not be located in the same facility, in the same server rack, or on the same motherboard. Further, the blocks need not correspond to physically distinct components. Blocks can be configured to perform various operations, e.g., by programming a processor or providing appropriate control circuitry, and various blocks might or might not be reconfigurable depending on how the initial configuration is obtained. Embodiments of the present disclosure can be realized in a variety of apparatus including electronic devices implemented using any combination of circuitry and software. While the disclosure has been described with respect to specific embodiments, one skilled in the art will recognize that numerous modifications are possible. Embodiments of the disclosure can be realized using a variety of computer systems and communication technologies including but not limited to the specific examples described herein. Embodiments of the present disclosure can be realized using any combination of dedicated components and / or programmable processors and / or other programmable devices. The various processes described herein can be implemented on the same processor or different processors in any combination. Where components are described as being configured to perform certain operations, such configuration can be accomplished, e.g., by designing electronic circuits to perform the operation, by programming programmable electronic circuits (such as microprocessors) to perform the operation, or any combination thereof. Further, while the embodiments described above may make reference to specific hardware and software components, those skilled in the art will appreciate that different combinations of hardware and / or software components may also be used and that particular operations described as being implemented in hardware might also be implemented in software or vice versa. Computer programs incorporating various features of the present disclosure may be encoded and stored on various computer-readable storage media; suitable media include magnetic disk or tape, optical storage media such as compact disk (CD) or DVD -69- 4938-3168-7971.1 Atty. Dkt. No.: 115872-3198 (digital versatile disk), flash memory, and other non-transitory media. Computer-readable media encoded with the program code may be packaged with a compatible electronic device, or the program code may be provided separately from electronic devices (e.g., via Internet download or as a separately packaged computer-readable storage medium). Thus, although the disclosure has been described with respect to specific embodiments, it will be appreciated that the disclosure is intended to cover all modifications and equivalents within the scope of the following claims. -70- 4938-3168-7971.1

Claims

Atty. Dkt. No.: 115872-3198 WHAT IS CLAIMED IS:

1. A method of reconstructing images using projection data from cone beam computed tomography (CBCT), comprising: receiving, by one or more processors, a projection dataset of at least a portion of a volume of a subject, the projection dataset generated via a gantry of a CBCT scanner rotating along an arc at least partially about the volume during a time window, the volume of the subject switching between a first state and a second state during the time window, the projection dataset comprising: (i) a plurality of projections corresponding to a first plurality of segments of the arc, each of the first plurality of segment corresponding to when the volume of the subject is in the first state, and (ii) a plurality of gaps corresponding to a second plurality of segments of the arc, each of the second plurality of segment corresponding to when the volume of the subject is in the second state, applying, by the one or more processors, a machine learning (ML) architecture on the projection dataset, wherein the ML architecture is established using training data comprising a plurality of examples, each of the plurality of examples having: (i) a sample projection dataset comprising a sample plurality of projections corresponding to at least a portion of a sample arc about a sample volume of a sample subject over a sample time window and (ii) a sample CT image corresponding to the sample volume of the sample subject; generating, by the one or more processors, based on applying the ML architecture on the projection dataset, a CT image corresponding to the volume of the subject; and storing, by the one or more processors, using one or more data structures, an association between the subject and the CT image.

2. The method of claim 1, further comprising providing, by the one or more processors, for presentation via a user interface, an output based on the CT image corresponding to the volume of the subject.

3. The method of claim 2, further comprising: -71- 4938-3168-7971.1Atty. Dkt. No.: 115872-3198 receiving, by the one or more processors, via the user interface, an indication that a target within the volume corresponding to the CT image as matching a location identified in a therapy plan; and providing, by the one or more processors, responsive to the indication, an instruction to administer radiotherapy to the target in accordance with the therapy plan.

4. The method of claim 3, wherein the target within the volume of the subject is administered with the radiotherapy, wherein the radiotherapy comprises at least one of a stereotactic body radiation therapy (SBRT), an intensity-modulated radiation therapy (IMRT), a volumetric modulated arc therapy (VMAT), a conformal radiation therapy (CRT), or a proton beam therapy (PBT).

5. The method of claim 2, further comprising: receiving, by the one or more processors, via the user interface, an indication that a target within the volume corresponding to the CT image as not matching a location identified in a therapy plan; and refraining, by the one or more processors, responsive to the indication, from providing an instruction to administer radiotherapy to the target.

6. The method of claim 1, further comprising generating, by the one or more processors, using the projection dataset in accordance with an interpolator, a second projection dataset comprising (i) the plurality of projections corresponding to the first plurality of segments and (ii) a second plurality of projections corresponding to one or more of the second plurality of segments; and wherein applying the ML architecture further comprises applying the ML architecture to the second projection dataset to generate the CT image.

7. The method of claim 1, wherein the ML architecture further comprises: a projection synthesizer configured to generate, using the projection dataset, a second projection dataset lacking one or more of the plurality of gaps; -72- 4938-3168-7971.1Atty. Dkt. No.: 115872-3198 a domain converter configured to generate, from the second projection dataset, an initial CT image corresponding to the volume of the subject; and an image reconstructor configured to generate, using the initial CT image, the CT image corresponding to the volume of the subject.

8. The method of claim 1, wherein the ML architecture further comprises: an encoder configured to generate, using the projection dataset, a plurality of embeddings corresponding to features; and a decoder configured to generate, using the plurality of embeddings, the CT image corresponding to the volume of the subject.

9. The method of claim 1, wherein the sample projection dataset in at least one of the plurality of examples comprises at least one of: (i) a non-stop, gated sample projection dataset comprising: (a) the sample plurality of projections corresponding to a first sample plurality of segments of the sample arc, each of the first sample plurality of segment corresponding to when the sample volume of the sample subject is in the first state, and (b) a sample plurality of gaps corresponding to a sample second plurality of segments of the sample arc, each of the second plurality of segment corresponding to when the volume of the subject is in the second state, or (ii) a gated sample projection dataset comprising sample plurality of projections corresponding to an entirety of the sample arc during the sample time window.

10. The method of claim 1, wherein the subject is at risk of or diagnosed with cancer associated with an organ within the volume, wherein the first state corresponds to the volume of the subject being within a gating window and the second state corresponds to the volume of the subject being outside the gating window, wherein the cancer comprises at least one of lung cancer, esophageal cancer, stomach cancer, colorectal cancer, liver cancer, pancreatic cancer, small intestine cancer, kidney cancer, prostate cancer, testicular cancer, ovarian cancer, uterine cancer, or cervical cancer. -73- 4938-3168-7971.1Atty. Dkt. No.: 115872-3198 11. A system for reconstructing images using projection data from cone beam computed tomography (CBCT), comprising: one or more processors coupled with memory, configured to: receive a projection dataset of at least a portion of a volume of a subject, the projection dataset generated via a gantry of a CBCT scanner rotating along an arc at least partially about the volume during a time window, the volume of the subject switching between a first state and a second state during the time window, the projection dataset comprising: (i) a plurality of projections corresponding to a first plurality of segments of the arc, each of the first plurality of segment corresponding to when the volume of the subject is in the first state, and (ii) a plurality of gaps corresponding to a second plurality of segments of the arc, each of the second plurality of segment corresponding to when the volume of the subject is in the second state, apply a machine learning (ML) architecture on the projection dataset, wherein the ML architecture is established using training data comprising a plurality of examples, each of the plurality of examples having: (i) a sample projection dataset comprising a sample plurality of projections corresponding to at least a portion of a sample arc about a sample volume of a sample subject over a sample time window and (ii) a sample CT image corresponding to the sample volume of the sample subject; generate, based on applying the ML architecture on the projection dataset, a CT image corresponding to the volume of the subject; and store, using one or more data structures, an association between the subject and the CT image.

12. The system of claim 11, wherein the one or more processors are further configured to provide, for presentation via a user interface, an output based on the CT image corresponding to the volume of the subject.

13. The system of claim 12, wherein the one or more processors are further configured to -74- 4938-3168-7971.1Atty. Dkt. No.: 115872-3198 receive, via the user interface, an indication that a target within the volume corresponding to the CT image as matching a location identified in a therapy plan; and provide, responsive to the indication, an instruction to administer radiotherapy to the target in accordance with the therapy plan.

14. The system of claim 13, wherein the target within the volume of the subject is administered with the radiotherapy, wherein the radiotherapy comprises at least one of a stereotactic body radiation therapy (SBRT), an intensity-modulated radiation therapy (IMRT), a volumetric modulated arc therapy (VMAT), a conformal radiation therapy (CRT), or a proton beam therapy (PBT).

15. The system of claim 12, wherein the one or more processors are further configured to receive, via the user interface, an indication that a target within the volume corresponding to the CT image as not matching a location identified in a therapy plan; and refrain, responsive to the indication, from providing an instruction to administer radiotherapy to the target.

16. The system of claim 11, wherein the one or more processors are further configured to: generate, using the projection dataset in accordance with an interpolator, a second projection dataset comprising (i) the plurality of projections corresponding to the first plurality of segments and (ii) a second plurality of projections corresponding to one or more of the second plurality of segments; and apply the ML architecture to the second projection dataset to generate the CT image.

17. The system of claim 11, wherein the ML architecture further comprises: a projection synthesizer configured to generate, using the projection dataset, a second projection dataset lacking one or more of the plurality of gaps; a domain converter configured to generate, from the second projection dataset, an initial CT image corresponding to the volume of the subject; and an image reconstructor configured to generate, using the initial CT image, the CT image corresponding to the volume of the subject. -75- 4938-3168-7971.1Atty. Dkt. No.: 115872-3198 18. The system of claim 11, wherein the ML architecture further comprises: an encoder configured to generate, using the projection dataset, a plurality of embeddings corresponding to features; and a decoder configured to generate, using the plurality of embeddings, the CT image corresponding to the volume of the subject.

19. The system of claim 11, wherein the sample projection dataset in at least one of the plurality of examples comprises at least one of: (i) a non-stop, gated sample projection dataset comprising: (a) the sample plurality of projections corresponding to a first sample plurality of segments of the sample arc, each of the first sample plurality of segment corresponding to when the sample volume of the sample subject is in the first state, and (b) a sample plurality of gaps corresponding to a sample second plurality of segments of the sample arc, each of the second plurality of segment corresponding to when the volume of the subject is in the second state, or (ii) a gated sample projection dataset comprising sample plurality of projections corresponding to an entirety of the sample arc during the sample time window.

20. The system of claim 11, wherein the subject is at risk of or diagnosed with cancer associated with an organ within the volume, wherein the first state corresponds to the volume of the subject being within a gating window and the second state corresponds to the volume of the subject being outside the gating window, wherein the cancer comprises at least one of lung cancer, esophageal cancer, stomach cancer, colorectal cancer, liver cancer, pancreatic cancer, small intestine cancer, kidney cancer, prostate cancer, testicular cancer, ovarian cancer, uterine cancer, or cervical cancer.

21. A method of performing cone beam computed tomography (CBCT), comprising: causing, by one or more processors, a CBCT scanner to rotate a gantry along an arc at least partially about a volume of a subject during a time window, the subject arranged along a surface of an examination table; -76- 4938-3168-7971.1Atty. Dkt. No.: 115872-3198 identifying, by the one or more processors, during the time window, a motion of the volume of the subject relative to the surface of the examination table; determining, by the one or more processors, based on the motion of the volume, the volume of the subject as in a first state instead of a second state; causing, by the one or more processors, responsive to determining the volume as in the first state, a beam emitter of the CBCT scanner to be activated to produce a radiation beam while the gantry is rotating along a segment of the arc; acquiring, by the one or more processors, via a detector of the CBCT scanner, a plurality of projections corresponding to the radiation beam traversing through the volume of the subject via the segment; generating, by the one or more processors, a projection dataset comprising the plurality of projections corresponding to the segment of the arc about the volume of the subject; and storing, by the one or more processors, using one or more data structures, an association between the subject and the projection dataset.

22. The method of claim 21, further comprising: determining, by the one or more processors, based on the motion of the volume, the volume of the subject as in the second state instead of the first state; causing, by the one or more processors, responsive to determining the volume as in the second state, the beam emitter of the CBCT scanner to be deactivated to cease production of the radiation beam while the gantry is rotating along a second segment of the arc; and continuing, by the one or more processors, to cause the CBCT scanner to rotate the gantry along the second segment of the arc about the volume of a subject, wherein generating the projection dataset further comprises generating the projection dataset to include a gap corresponding to the second segment of the arc.

23. The method of claim 21, further comprising: determining, by the one or more processors, based on the motion of the volume, the volume of the subject as switching back to the first state from the second state; -77- 4938-3168-7971.1Atty. Dkt. No.: 115872-3198 causing, by the one or more processors, responsive to determining the volume as switching back to the first state, the beam emitter of the CBCT scanner to be activated to produce the radiation beam while the gantry is rotating along a third segment of the arc; and acquiring, by the one or more processors, via the detector of the CBCT scanner, a second plurality of projections corresponding to the radiation beam traversing through the volume of the subject via the third segment; and wherein generating the projection dataset further generating the projection dataset to include the first plurality of projections corresponding to the first segment, the gap corresponding to the second segment, and the third plurality of projections corresponding to the third segment.

24. The method of claim 21, further comprising: providing, by the one or more processors, the projection dataset to a machine learning (ML) architecture, wherein the ML architecture is established using training data comprising a plurality of examples, each of the plurality of examples having: (i) a sample projection dataset comprising a sample plurality of projections corresponding to at least a portion of a sample arc about a sample volume of a sample subject over a sample time window and (ii) a sample CT image corresponding to the sample volume of the sample subject; and obtaining, by the one or more processors, based on applying the ML architecture, a CT image corresponding to the volume of the subject.

25. The method of claim 24, wherein the subject is administered with radiotherapy to a target within the volume in accordance with a therapy plan, wherein the radiotherapy comprises at least one of a stereotactic body radiation therapy (SBRT), an intensity- modulated radiation therapy (IMRT), a volumetric modulated arc therapy (VMAT), a conformal radiation therapy (CRT), or a proton beam therapy (PBT).

26. The method of claim 21, further comprising generating, by the one or more processors, using the motion of the volume of the subject prior to the time window, a threshold value -78- 4938-3168-7971.1Atty. Dkt. No.: 115872-3198 for the motion to define the subject as switching between the first state and the second state; and wherein determining the volume as in the first state further comprises determining the volume as in the first state instead of the second state based on a comparison of the motion with the threshold value.

27. The method of claim 21, wherein identifying the motion further comprises identifying the motion of the volume relative to the surface of the examination table based on at least one of (i) image data of the subject on the examination table or (ii) a position of a structure situated on a side of the subject distal from the surface.

28. The method of claim 21, wherein causing the beam emitter to produce further comprises causing the beam emitter of the CBCT scanner to produce the radiation beam in accordance with one of a full-fan configuration or a half-fan configuration, the full-fan configuration corresponding to the radiation beam being fully collimated with the detector, the half-fan configuration corresponding to the radiation beam being partially collimated with the detector.

29. The method of claim 21, wherein causing the CBCT scanner to rotate the gantry further comprises causing the CBCT scanner to rotate the gantry along the arc without interruption during the time window, wherein the arc along which the gantry rotates ranges between 140° to 360°.

30. The method of claim 21, wherein the subject is at risk of or diagnosed with cancer associated with an organ within the volume, wherein the first state corresponds to the volume of the subject being within a gating window and the second state corresponds to the volume of the subject being outside the gating window, wherein the cancer comprises at least one of lung cancer, esophageal cancer, stomach cancer, colorectal cancer, liver cancer, pancreatic cancer, small intestine cancer, kidney cancer, prostate cancer, testicular cancer, ovarian cancer, uterine cancer, or cervical cancer. -79- 4938-3168-7971.1Atty. Dkt. No.: 115872-3198 31. A system for performing cone beam computed tomography (CBCT), comprising: one or more processors coupled with memory, configured to: cause a CBCT scanner to rotate a gantry along an arc at least partially about a volume of a subject during a time window, the subject arranged along a surface of an examination table; identify, during the time window, a motion of the volume of the subject relative to the surface of the examination table; determine, based on the motion of the volume, the volume of the subject as in a first state instead of a second state; cause, responsive to determining the volume as in the first state, a beam emitter of the CBCT scanner to be activated to produce a radiation beam while the gantry is rotating along a segment of the arc; acquire, via a detector of the CBCT scanner, a plurality of projections corresponding to the radiation beam traversing through the volume of the subject via the segment; generate a projection dataset comprising the plurality of projections corresponding to the segment of the arc about the volume of the subject; and store, using one or more data structures, an association between the subject and the projection dataset.

32. The system of claim 31, wherein the one or more processors are configured to: determine, based on the motion of the volume, the volume of the subject as in the second state instead of the first state; cause, responsive to determining the volume as in the second state, the beam emitter of the CBCT scanner to be deactivated to cease production of the radiation beam while the gantry is rotating along a second segment of the arc; and continue to cause the CBCT scanner to rotate the gantry along the second segment of the arc about the volume of a subject, generate the projection dataset to include a gap corresponding to the second segment of the arc. -80- 4938-3168-7971.1Atty. Dkt. No.: 115872-3198 33. The system of claim 31, wherein the one or more processors are configured to: determine, based on the motion of the volume, the volume of the subject as switching back to the first state from the second state; cause, responsive to determining the volume as switching back to the first state, the beam emitter of the CBCT scanner to be activated to produce the radiation beam while the gantry is rotating along a third segment of the arc; and acquire, via the detector of the CBCT scanner, a second plurality of projections corresponding to the radiation beam traversing through the volume of the subject via the third segment; and generate the projection dataset to include the first plurality of projections corresponding to the first segment, the gap corresponding to the second segment, and the third plurality of projections corresponding to the third segment.

34. The system of claim 31, wherein the one or more processors are configured to: provide the projection dataset to a machine learning (ML) architecture, wherein the ML architecture is established using training data comprising a plurality of examples, each of the plurality of examples having: (i) a sample projection dataset comprising a sample plurality of projections corresponding to at least a portion of a sample arc about a sample volume of a sample subject over a sample time window and (ii) a sample CT image corresponding to the sample volume of the sample subject; and obtain, based on applying the ML architecture, a CT image corresponding to the volume of the subject.

35. The system of claim 34, wherein the subject is administered with radiotherapy to a target within the volume in accordance with the therapy plan, wherein the radiotherapy comprises at least one of a stereotactic body radiation therapy (SBRT), an intensity- modulated radiation therapy (IMRT), a volumetric modulated arc therapy (VMAT), a conformal radiation therapy (CRT), or a proton beam therapy (PBT).

36. The system of claim 31, wherein the one or more processors are configured to: -81- 4938-3168-7971.1Atty. Dkt. No.: 115872-3198 generate, using the motion of the volume of the subject prior to the time window, a threshold value for the motion to define the subject as switching between the first state and the second state; and determine the volume as in the first state instead of the second state based on a comparison of the motion with the threshold value.

37. The system of claim 31, wherein the one or more processors are configured to identify the motion of the volume relative to the surface of the examination table based on at least one of (i) image data of the subject on the examination table or (ii) a position of a structure situated on a side of the subject distal from the surface.

38. The system of claim 31, wherein the one or more processors are configured to cause the beam emitter of the CBCT scanner to produce the radiation beam in accordance with one of a full-fan configuration or a half-fan configuration, the full-fan configuration corresponding to the radiation beam being fully collimated with the detector, the half-fan configuration corresponding to the radiation beam being partially collimated with the detector.

39. The system of claim 31, wherein the one or more processors are configured to cause the CBCT scanner to rotate the gantry further comprises causing the CBCT scanner to rotate the gantry along the arc without interruption during the time window, wherein the arc along which the gantry rotates ranges between 140° to 360°.

40. The system of claim 31, wherein the subject is at risk of or diagnosed with cancer associated with an organ within the volume, wherein the first state corresponds to the volume of the subject being within a gating window and the second state corresponds to the volume of the subject being outside the gating window, wherein the cancer comprises at least one of lung cancer, esophageal cancer, stomach cancer, colorectal cancer, liver cancer, pancreatic cancer, small intestine cancer, kidney cancer, prostate cancer, testicular cancer, ovarian cancer, uterine cancer, or cervical cancer. -82- 4938-3168-7971.1

Images