Deep-RT: a universal quality control system to enhance the treatment quality and reduce the treatment inequity in radiotherapy of lung cancer patients

Abstract

Deep-RT is an artificial intelligence (AI)-based radiotherapy quality assurance system developed as a plug-in within a treatment planning system to enhance the quality of the lung radiotherapy and reduce the inter and intra-institution radiotherapy inequity. Deep-RT learns various radiotherapy class solution techniques developed to treat lung cancer at different stages of the disease each of them designed based on the physics of radiotherapy, tumor geometry and unique patient anatomy. Each class solution technique includes radiotherapy plan properties and 3D dose distribution. When a new patient comes in, Deep-RT generates the best radiotherapy planning setting and predicts the best personalized 3D dose distribution based on the tumor geometry and patient anatomy. The treatment planning system will then develop a personalized radiotherapy plan using the predicted plan properties and 3D dose distribution.

Description

DEEP-RT: A UNIVERSAL QUALITY CONTROL SYSTEM TO ENHANCE THE TREATMENT QUALITY AND REDUCE THE TREATMENT INEQUITY IN RADIOTHERAPY OF LUNG CANCER PATIENTS CROSS REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of U.S. Provisional Patent Application No. 63 / 748,244 filed on January 22, 2025, which is incorporated by reference, herein, in its entirety.FIELD OF THE INVENTION

[0002] The present invention relates generally to radiotherapy. More particularly, the present invention relates to a quality control to enhance treatment quality and reduce treatment inequity in radiotherapy for lung cancer patients.BACKGROUND OF THE INVENTION

[0003] Lung cancer is a leading cause of cancer death in the United States with estimated new cases of 253,000 and death of 130,000 in 2024. From all types of lung cancer, non-small cell lung cancer (NSCLC) comprises the majority of cases. For localized NSCLC, the overall 5-year relative survival rate is nearly 65%. For advanced disease, the 5-year relative survival rate may drop to 9%. Among various treatment modalities, radiotherapy (RT) has shown to be the only modality for which there is indication in all stages of lung cancer and can be part of the disease management in nearly 77% of the patients. Stereotactic body radiation therapy (SBRT), is the standard of care in patients with inoperable early-stage disease, including those with poor pulmonary function at baseline and has been suggested that may lead to a better overall survival (OS) compared to surgery in those with operable disease. RT may also be given concurrently or sequentially with chemotherapy in locally advanced disease or combined with targeted therapy such as immunotherapy. Patients with early-stage disease may also benefit from these treatment combinations as such patients still have high rates ofrelapse after definitive resection or SBRT, thus, there may be a need for post-resection or post-SBRT consolidation therapy.

[0004] In advanced disease, radiotherapy is mainly limited by radiation-induced lung injuries (RILI) which includes radiation-induced pneumonitis (RP) and the subsequent radiation-induced fibrosis (RF) which can affect 15-40% of patients. In early-stage disease where lesions are small and usually treated with ablative dose, the radiation-induced toxicity is minimal but cumulative dose from multiple treatments can increase the chance of lung injuries or limit the future treatments if needed. The total lung dose is the main factor in causing RP. Various radiation dosimetric parameters including the mean lung dose (MLD), V60Gy (volume of lung receiving 60 Gy), V20Gy, and low dose volume such as V5Gy have also been found to be correlated in development of radiation-induced toxicity.

[0005] Immunotherapy can also induce subsequent damage to normal tissue, a phenomenon called immune-related adverse events (irAEs). It is believed that the rate of immune related lung injuries can be increased in combination with RT as each treatment modality’ can manifest some level of toxicity and the synergistic effect of the two could increase the chance of lung damage. Several pharmacological agents are currently under investigation to prevent and / or treat RP and RF such as protectors, modifiers, and mitigators of radiation-induced lung toxicity.

[0006] Angiotensin-converting enzyme (ACE) inhibitors and angiotensin-II receptor subtype 1 (AT-1) antagonists have been shown to mitigate the radiation-induced damage by targeting inflammatory and fibrogenic pathways in preclinical studies. Amifostine is traditionally used to attenuate renal toxicity and / or xerostomia during anti-cancer chemoradiation therapy. Several clinical trials have shown its effectiveness to lower the rate of clinically apparentpneumonitis upon chemoradiotherapy of lung cancer patient. Prophylactic use of inhalative corticosteroids has been also suggested to prevent radiation-induced lung toxicity. However, despite encouraging preclinical results, clinical trials did not show efficacy of such agents in the prevention of RP and RF. In addition, studies have also shown that increase in the time to treatment can adversely affect the treatment outcome in both early stage and advanced disease requiring an administration of high-quality treatment with an appropriate timing. Therefore, the best strategy to address above issues could be investigating novel delivery techniques or standardized treatment planning approaches that can provide a robust and high-quality treatment in a timely fashion.

[0007] In recent years, improvement in treatment delivery techniques such as intensity-modulated radiotherapy (IMRT) or volumetric modulated arc therapy (VMAT) played a significant role to enhance the therapeutic index of RT and reducing the RILI in lung cancer patients by improving the dose conformality and reducing the high dose volume in lung tissue. In contrast, these approaches increase the low dose volume which may be associated with different levels of toxicity. This issue is mainly due to the increase in the number of beam entry angles into the body.

[0008] It would therefore be advantageous to provide a quality control to enhance treatment quality and reduce treatment inequity in radiotherapy for lung cancer patients.SUMMARY OF THE INVENTION

[0009] The foregoing needs are met, to a great extent, by the present invention, a system for radiation planning includes a processing device, wherein the processing device is programmed for assessing tumor geometry’. The processing device is also programmed for assessing patient anatomy. The processing device is programmed for determining a 3D dosedistribution based on the tumor geometry and patient anatomy. The processing device is also programmed for determining a personalized radiotherapy plan using predicted plan properties and the 3D dose distribution.

[0010] In accordance with an aspect of the present invention, the tumor is identified as lung cancer. The processor can be programmed for assessing a treatment stage for the patient. The system further includes a database of class solution techniques and predictive dose mapping models. The system further includes a processor programmed for generating predictive dose mapping models for different types and stages of tumor. The system further includes receiving user input related to the treatment. The system further includes determining a beam arrangement for the treatment. The system further includes predicting a 3D-dose distribution corresponding to the beam arrangement. The system further includes a planning dataset. The planning dataset is directed to a treatment modality7. The planning dataset is directed to a type and stage of cancer.

[0011] In accordance with a method of the present invention, a method for radiation planning includes assessing tumor geometry, with a processing device. The method includes assessing patient anatomy, with the processing device. The method includes determining a 3D dose distribution based on the tumor geometry and patient anatomy, with the processing device. The method includes determining a personalized radiotherapy plan using predicted plan properties and the 3D dose distribution, with the processing device.

[0012] In accordance with an aspect of the present invention, the method includes assessing a treatment stage for the patient. The method includes accessing a database of class solution techniques and predictive dose mapping models. The method includes generating predictive dose mapping models for different types and stages of tumor. The method includes receivinguser input related to the treatment. The method includes determining a beam arrangement for the treatment. The method includes predicting a 3D-dose distribution corresponding to the beam arrangement. The method includes generating a planning dataset. The planning dataset is directed to a treatment modality.BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The accompanying drawings provide visual representations, which will be used to more fully describe the representative embodiments disclosed herein and can be used by those skilled in the art to better understand them and their inherent advantages. In these drawings, like reference numerals identify corresponding elements and:

[0014] FIGS. 1A-1C illustrate image and graphical views of the principle that therapeutic gain for a beam angle is inversely proportional to the physical depth of a lesion to the beam entry at the patient surface.

[0015] FIG. 2 illustrates image and graphical views of dose distribution between the initial and plan with beam optimization.

[0016] FIG. 3 illustrates a graphical view of dose volume histograms (DVH) between the two plans shown in FIG. 2.

[0017] FIG. 4 shows the scatter plot of %Δd̄ph vs %ΔIDLV x, x = 25, 20, 15, 10, and 5Gy between the initial and new plans for all twenty-five lesions.

[0018] FIG. 5 illustrates graphical views of scatter plots of %Δd̄phvs %ΔMLD. %AV20Gy, ° / oV10Gy, and %ΔV5Gy between the initial and new plans for all twenty-five lesions in this study.

[0019] FIG. 6 shows the boxplots of dosimetric indices between the initial and plans with beam optimization (new) for all 25 cases.

[0020] FIG. 7 illustrates image views of initial and plan with beam optimization for cases with different changes in lesion depth with respect to beam-eye-view.

[0021] FIGS. 8, 9, and 10 illustrate flow diagrams of the dose prediction model of the present invention.DETAILED DESCRIPTION

[0022] The presently disclosed subject matter now will be described more fully hereinafter with reference to the accompanying Drawings, in which some, but not all embodiments of the inventions are shown. Like numbers refer to like elements throughout. The presently disclosed subject matter may be embodied in many different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will satisfy applicable legal requirements. Indeed, many modifications and other embodiments of the presently disclosed subject matter set forth herein will come to mind to one skilled in the art to which the presently disclosed subject matter pertains having the benefit of the teachings presented in the foregoing descriptions and the associated Drawings. Therefore, it is to be understood that the presently disclosed subject matter is not to be limited to the specific embodiments disclosed and that modifications and other embodiments are intended to be included within the scope of the appended claims.

[0023] Deep-RT is an artificial intelligence (Al)-based radiotherapy quality assurance system developed as a plug-in within a treatment planning system to enhance the quality of the lung radiotherapy and reduce the inter and intra-institution radiotherapy inequity. Deep-RT learns various radiotherapy class solution techniques developed to treat lung cancer at different stages of the disease each of them designed based on the physics of radiotherapy, tumor geometry and unique patient anatomy. Each class solution technique includes radiotherapy plan properties and 3D dose distribution. When a new patient comes in. deep RT generates the best radiotherapy planning setting and predicts the best personalized 3D dose distributionbased on the tumor geometry and patient anatomy. The treatment planning system will then develop a personalized radiotherapy plan using the predicted plan properties and 3D dose distribution.

[0024] To quantify beam optimization for stereotactic body radiotherapy (SBRT) of peripheral lung lesions, a new beam optimization approach is based on maximizing the therapeutic gain of the beam set by minimizing the average physical depth of the lesion with respect to the beam’s eye view (BEV). The new approach was evaluated by replanning the twenty -five SBRT lesions, retrospectively to assess if a better plan is achievable in all aspects. Difference in 25Gy isodose line volume (IDLV25Gy), IDLV20Gy, IDLVisGy, IDLVioGy and IDLVsGy between the two plan cohorts were calculated as a measure of plan size and fitted in a linear regression model against the changes in the lesion depth with respect to the BEV to assess the relationship between the changes in the treatment depth and that of the plan size. Beam optimization achieved a better plan in all cases by lowering the depth of treatment with an average of %20.03 ± 12.30 (3.66 - 45.78). As the depth of treatment decreases, the size of the plan also decreases. A reduction of %4.64 ± 4.55 (0.02-21.58, p < 3.8 x 10’5), %5.16 ± 5.54 (0.03-24.68, p < 0.005), %6.46 ± 6.95 (-1.35-29.05, p < 0.009), %12.83 ± 9.06 (0.89-37.65, p < 0.0001), and %14.01 ± 9.87 (1.43- 41.84, p < 4.5 x IO’6) was observed in IDLV25Gy, IDLV20Gy, IDLVisGy, IDLVioGy, and IDLVsGy, respectively. Physical depth of the lesion with respect to the beam’s eye view is inversely proportional to the therapeutic gain of a beam-set and can be used as a robust and standard metric to select an appropriate beam-set for SBRT of the peripheral lung lesions. Further evaluation warrants the utility of such concept in routine clinical use.

[0025] Therefore, optimization of beam entry angles prior to plan optimization can help reduce the low-dose volume in such treatment. In fact, due to the presence of large tissue heterogeneity’ in the lung region and variation in the target location, various beam angles havedifferent impact on the therapeutic index of a treatment plan. This variation is more pronounced for non-central lesions as the physical and water-equivalent distance of the lesion to the beam entry varies significantly for various beam angles. When the target is small, the isocenter or the lesion center can be a good estimate of the lesion distance from the beam entry and the beam optimization can be achieved by focusing on the distance of the isocenter to the beam entry. However, this assumption is not valid for large lesion, hence more sophisticated metric may be needed for such purpose.

[0026] Here, a quantitative metric is used to achieve a robust and standard beam arrangement to develop a high-quality treatment plan for ablative radiotherapy of peripheral lung nodules. Small lesions are usually treated with ablative dose. Therefore, attention is focused on the optimization of the beam angles for the SBRT of the peripherally located lesions. Several references have shown the importance of optimal beam selection in reducing the organs at risk (OARs) dose in SBRT of the lung lesions. However, most of these works provide qualitative assessment such as use of couch kicks or non-coplanar beams to reduce OARs dose. Therefore, the final beam arrangement could still depend on the planner’s experience, vary person-to-person, and also need several attempts to find the best setting which can prolong the planning process as well. In contrast, the present invention is designed to eliminate these qualitative assessments to provide high-quality, repeatable results regardless of experience, perspective, setting, etc.

[0027] In conventional SBRT with volumetric modulated arc therapy (VMAT), depending on the location of the lesion, a set of partial or full arcs is designed by the planner in an optimization setting and then several clinical objectives are defined and updated interactively based on the clinical goals to come up with a desired treatment plan. It has been shown thatoptimal beam selection prior to plan optimization can help reduce the OARs dose and achieve a better plan. To quantify this process, the following procedure is used.

[0028] For an isocentric treatment, the dose deposition and fall-off are formulated by the tissue maximum ratio (TMR). A lesion gets the highest relative dose compared to its surrounding organs at risk (OARs) when it is located at the depth of maximum TMR (dmax). Hence, the closer a lesion to the dmax, the higher the therapeutic gain (TG) of a beam where the TG is defined as follows:∫TMRtumor(bθ)dlTG(bθ) = (1)∫TMROARs(bθ)dl

[0029] In the above, bθdenotes the beam at the angle θ and OARs represent any voxel beyond the tumor boundary on the beam trajectory line. Similarly, the therapeutic gain for an arc is defined as follows:TG(Arc(θstart, θstop)) = ∫TG(bθ)dθ (2)In treating the peripheral lung lesions, partial arcs are usually preferred to avoid entering the contralateral lung. Also, to achieve acceptable conformality around the target, the length of the arcs is often set to be equal or greater than a half arc. Hence, a set of half arcs with maximum TG for a lesion are found.

[0030] Because the chest wall thickness is often more than 2 cm, it can provide adequate buildup to establish electron equilibrium for a photon beam with an energy of 6MV. This also results in treating the lesion with a TMR < 1. Therefore, the closer a lesion to the chest wall. the higher the therapeutic gain of the beam. In other word, the therapeutic gain of a beam is inversely proportional to the physical distance of the lesion to the beam entrance. For small lesions, where the distance of lesion center or isocenter to the beam entrance can be a good estimate of lesion distance to the beam entrance,TG(Vbg0')7oc - - - (3dph,e(iso)V)7

[0031] Where in the above, dPh,e (iso), denotes the physical distance of the isocenter at an angle 0 to the patient surface. Please note that due to the presence of lung tissue, water equivalent depth of a beam at certain angle could be lower than that of the other angles but this does not guarantee that the therapeutic gain at that angle is also higher since the lung tissue is initially treated before that beam reaches the target and this lower the therapeutic gain of that beam, as illustrated in FIG. 1 A. Finally, the therapeutic gain of a beam set is inversely proportional to the average physical depth of the isocenter to the beam entry as follows and as illustrated in FIG. 1B:TG(Arc(θstart, θstop)) ∝ (4)∫θstartdph,θ(iso)

[0032] FIGS. 1 A-1C illustrate image and graphical views of the principle that therapeutic gain for a beam angle is inversely proportional to the physical depth of a lesion to the beam entry at the patient surface. FIG. 1 A illustrates various beam angles with their corresponding physical and water equivalent depths; bo and b180have the highest and lowest therapeutic gain among these three beams. FIG. 1B illustrates that although b130has higher water equivalent depth compared to b180, it has higher therapeutic gain than b180as it treats less lung tissue before the beam reaches the target. FIG. 1C illustrates a graphical view which shows that to determine the best arc arrangement, the plot of physical depth vs beam angle for the lesion is estimated by placing 72 evenly spaced beams around the lesion. The distance of the lesion center to the patient surface indicates the physical depth at that angle. The optimal arc set to treat the lesion is the 180 arc-set with the lowest area under the curve which will be a set of 295-115 partial arcs for this case.

[0033] FIG. 1 A shows a parallel opposed beams and one independent beam angle with the corresponding values of physical and water equivalent depths. As shown, bo has the highest therapeutic gain and b180has the lowest therapeutic gain for the lesion shown in this Figure.Also, although b130has higher water equivalent depth than b180, it has higher TG due to its lower physical depth. FIG. 1C shows the plot of physical depth vs beam angles for this lesion. FIG. 1C is estimated by placing certain number of (72 in this work) evenly-spaced beams, FIG. 1B, around the lesion and computing the distance of the lesion center to the patient surface at each direction. The optimal arc set to treat the lesion is the set of half arc with the lowest area under this curve which will be a set of 295-115 partial arcs for this case.

[0034] Twenty-five patients who had peripherally located lung lesions and were recently treated with SBRT were included in this IRB-approved study, Table 1. Planning target volume (PTV) ranged between 5.11 - 71.52 cc (24.20 cc ± 17.11). Lesions were treated with a prescription dose of Rx= 10 Gy* 5 (N = 20), 12 Gy* 4 (N = 4) and 12.5 Gy* 4 (N = 1). For each target, a plot of physical depth vs beam angle was generated and used an in-house developed script to provide the 180 arc-set with minimum area under the curve as the optimal beam set. Table 1 also shows the average physical depth of the lesion with respect to the original and optimal beam sets for each lesion. Each case was re-planned using the new beam-set with the aim that a superior plan is achieved compared to the clinical plan used to treat the patient. A superior plan is a plan with better coverage, conformity index and OARs sparing for all normal tissue compared to the clinically approved plan. Improvement in one aspect is not the result of compromise on other aspects of the plan. Also, a plan is only accepted if the total number of monitor unit (MU) for the new plan was lower than the initial plan. Because each round of optimization can increase the total MUs and plan improvement could happen with the cost of overmodulation, the new plan was limited with the original MUs to assure that any possible improvement is the result of optimal beam selection rather than overmodulation.Tsfele 1: Characteristic of the patent and lesions useh iri th& stvciy. The average physical depth of the festen for the initial and Prtirrwl beam arrartspptent are aiso Xnctwcs for es<#» lesion... No. F / M Location Primary Diagnosis TNM Staging Dose (Fix) PTV loo) initial Optimal beam-set beam-set 1 F RLL SCCa of epiglottis Metastatic lung 1000 * S 16,01 12.56 7.38 2 F RU Lung Asientrcarcinoma T2NXM8 1090 * 5 28.23 8.4 5.31 3 F RUL NSCLC TlbNOMO 1000* 5 30.45 10.88 7.66 4 M RU Thymic -carcinoma Metastatic tang 2000 * 5 25.06 15.28 13.25 5 F RLl SCIC adenocarcinoma T1N2M0 1090* 5 9. S4 17.44 12.22 5 F Ul SPNs Solitary Nodule 1000 » 5 11.5 9.28 8.13? F LUI Lung Adenocarcinoma T1NXM0 1000 * 5 49.39 9.23 8.13 8 M LUI SONS Solitary Nodule 1090 * 5 25.31 11.46 18.3 9 F LUI umg Adenocarcinoma T1NXM0 1080 * 5 29.31 12.28 10.75 10 F LUI lung Adenocarcinoma TlfiOMO 2280 * 4 28.84 8.24 6.8S 11 M RUl Colorectal cancer Metastatic lung WOO * 5 10.26 10.66 S.4? 12 M RUl SCLC T1N0M0 1200 * 4 17.04 9.3 8.81 13 M LUI Sarcona Metastatictang 2080* 5 8.71 9.37 6.39 14 F I spica! Lung Adenocarcinoma T4N2Mla 1090* 5 10.02 11.2 10.75 15 F LIL LungAdenocatctaoma T2aN2M2b 1000 * 5 7.68 6.41 6,083 IS F RMl SPNs Solitery Module 1000 * 5 24.74 10.12 9 17 M LUI Lung Adenocarcinoma T2bN8M0 1000 * 5 53.77 11.05 8.6S6 IS M RUl NSCLC T2N0MO 1.000 * 5 51.48 9.48 8.41 19 M LUI lung Adenocarcinoma T1N0M0 2250* 4 29.83 9.92 7. OS 20 F RU. Umg Adenocarcinoma T1N0M0 2200* 4 22. Q6 6.93 6.38 21 M RLL N& XC T1CN8M8 1080* 5 71.52 15.39 11.66 22 F RU N / A Sditery ftoctale 1000 * 5 5.11 7.26 4 23 M RU IWMiBi carcinoma Metastatic tang 1080 * 5 28.3 8.. S 4.5 24 M I. UL GE Adenocarcinoma Metastatic tang 1290 * 4 2’2.71 13.1 10.83 25 M LU. SPNs Mtary Nodule 1W 5 42.08 7.29 5.69 FTt" pferwirrg target vofome, ■: female anil te. Mile. fii;:.. r ipht u pper lobe, lt.lt- left upper -. Li • right fewer fcb-e and ILU Wt tower iobe, SCCs; $«tuam<n» ceil carcinoma, MSCI. C: non- trrisil cefi lung cancer, SCtC: small cell lung cancer, SiNsr solitary pulmenery oodufes, GE. gastrointestinal, TNM: tumc-t, node, metastaslc. a»:prescriptive.,r:, • Aveiege physicai rteptb

[0035] Table 2 shows the clinical objectives used for this study.Table 2: dosimetric indices used to define the clinical goals inthis study. PTV: planning target volume,. GTV: gross tumorvolume,Structure Dosimetric index (Unit}fiTOG Quality of CoverageRTOG Homogeneity IndexMeso (eGy)GTV V100(%)RTOG Conformity IndexConformity Indices Gradient indexRatio of Dmax at 2cm to Rx doseMean Lung Dose (MID), cGyLung V20Gy (cc)VlOGy(cc)V5Gy (cc).... „ V30Gy (cc)Chest wail.V4SGy (cc)Spinal Cord Max Dose (cGy[Skin Max Dose (cGy)Pericardium Max Dose (cGy)Esophagus Max Dose (cGy)Airways Max Dose (cGy)Also. RTOG conformity indices are used to assess the plan conformality as follow:CIRTOG=^~, RT OG Conformity IndexQRTOG=-^-, RTOG Quality of Coverage IndexGI —Vso%rx, Gradient Index (5) Vrx HRTOG= j I2^ > RTOG Homogeneity Index

[0036] Where in the above, TV denotes the target volume, VRI is the prescription isodose volume, Imin and Imax is the minimum and maximum dose in the target, respectively, RI indicates the prescription dose, Vso%rxis the volume of 50% prescription dose and Vrxshows the volume of the prescription dose. Planning was done using RayStation (RaySearch Laboratories Inc. Clinical Version 2023B) treatment planning system. The clinicallyapproved plans were done using a set of tw o partial arcs (clockwise and counter clockwise) with 2° gantry spacing. 6MV flattening free filter (fff) energy was used in all cases. In the new' plan, each arc is replaced with the new arc but with the initial setting. Dosimetrists who did the clinical plans were blinded about this study.

[0037] To evaluate the performance of the beam optimization approach, assessed each new' plan w as initially assessed to verify that if all clinical objectives achieved a comparable or better value compared to the initial plan. Then, the isodose line volumes of 25 Gy (IDLV25Gy), 20 Gy, 15 Gy, 10 Gy, and 5 Gy were calculated for both initial and new plans as a measure of plan size for each isodose line level. The volumetric change of each isodose line level was then calculated between the initial (i) and new' (n) plans as follow s:%ΔIDLVX=, DLV^~, DL* ioox= 25, 20, 15, 10 and 5 (6)For each patient, the difference between the average physical depth of the lesion with respect to the treating beam set was also calculated between the initial (i) and new plan (n) as follows:%Δd̄ph=dphCdphn* 100 (7)VdphiFor each isodose line level, the scatter plot of %ΔIDLVXvs %Δd̄phwas drawn and fitted to a linear regression model to assess the relationship between changes in the depth of lesion from the beam’s eye view and changes in the plan size at that particular dose level. Similar analysis was also performed for the lung tissue in a way that changes of mean lung dose between the initial and new plans (%ΔMLD), V20Gy (%ΔV20), VlOGy (%ΔV10), and V5Gy (%ΔV5) were also plotted against the %Δd̄phand fitted to a linear regression model to assess the relationship between the changes in the depth of lesion with respect to the beam’s eye view and changes in the lung dosimetric indices. Dosimetric statistics of the above metrics were also compared betw een the two planning cohorts. Statistical student paired t-test was performed when comparing the two planning cohorts.

[0038] Beam optimization achieved a better plan for all patients in the study. In other words, with the same or better coverage, a plan with equal or better OARs sparing and lower MU was achieved for all lesions using the beam optimization approach. FIGS. 2 and 3 illustrate the initial and plan with beam optimization along with the dose volume histogram (DVH) comparison between the two plans for the lesion shown in FIG. 1 A along w ith the position of the arc set for each beam arrangement on the physical depth vs beam angle cur e. Using beam optimization, the average physical depth of the lesion with respect to the beam set decreased by -31% (from 9.37 cm to 6.39 cm). Table 3 also show's a dosimetric comparison between the initial and new plan for this case. Please note that some of the OARs dose such as Spinal Cord max dose was very low' for both plans but attempt w as made to make sure that all OARs dose in the new plan w as lower than the initial plan to assure that any possible improvement in the plan is not the result of dosimetric compromise to any other structures. It’s worthwhile noting that better plan w as achieved for this case while the total MU also decreased by -22%. Also, please note that a shorter arc set with fewer number of control points was used in the new' plan.Table 3, Ooilmeirfe comparison between the initial and pfer® with beam optimisation. The comparison iiwSuttes a typical example shown: jp Figure 2 and the group comparison between all the twenty-five patients Example shown in fig 2 Group Comparison Structure Dosimetric lnd«x (Unit) F Value Initial Plan W / 80 Initial Plan W / BO VWOffij 95.00 95,13 95.5411,10 95.64 * 1.22 0.021 «96 Suai-iy oiCoverage 0.923 8-827 9.8971 0.9110035 0.11 FTV 0.0S7Homogeneity Index l.t.72 1.569 5.53 l ow 1.6010.99 £3.4? Mean |cGy) 8095 8086 6G41.71134 6084.8 t. lUS 0.001 GTV WO&ptj 1< W 100 100 100 N / A. STOGConfarmirytnder 1.043 1.011 1.04 i 004 1,0310.4 0.13 Gradient Index 5.82 5.515 4-6630 73 4.4710.6 0.00? ” *“* 8«t<oofbmaxat2cmtoftiidos«(^ 485 48.5 50.1616.9 47.616.9 O.4X3X Mean Lung Oosa JIWLO}. xGy 15* 13* 2601105.8 244.1198 3 £-0'1 V206V M 68.07 54.83 113.8155 106160 0.0002 •w-« V108y{C<) 182,45 14053 274.41129 33551112 7.68-09VSGyfcej 373.3 291-95 477.61 ISO 433. S U73 3.46-87 „.. vswed s:?i * & IO. SIU MH 13.8 a.0Q41WSy lcrj 0.33 023 2.7613.9 2.613.8 9.90S Spinal Cord Max Dose (c6y) 375 381 7591480 6921435 6 £-05: Skin Max Bess {oGy> 1538 1459 15541499 15051395 3.364® Pericardium Max Dose |cGy| 1100 1074 715.111028 666.181824 0.003 W Max Dose IcGyl 810 760 392,81608 732.41460 00006 Airways Max Dona fcGyl 522 452 71® 1698 6351609 0.002 O. V;:-.(iy(cc) 52.21 48.6 399,2159 195.2187 3.86-0$ IOLV. X,., («} 87.34 174.41111 ISS. J i W 9.90® IDLVs-zn 166-SS 138.13 321.58220 3911202 0.0009 lOt. Vj. Bv (Cd 395.53 25S. SB 699.41480 W.41376 0.0001 lOlVx., (C<) 943.09 677.83 1514.71771 13212664 4. S6-95MU 3923 SOU 3.388.417850 3001.51582 0.0001

[0039] FIG. 2 illustrates image and graphical views of dose distribution between the initial and plan with beam optimization. The physical depth of the lesion with respect to beam set was decreased from 9.37 to 6.39 in the new plan. In the initial plan, IDLV25Gy= 52.21 cc, IDLV20Gy= 87.34 cc, IDLVisGy =166.55 cc, IDLV ioGy — 395.53 cc, and IDLVsGy — 943.09 cc; In the new plan IDLV25Gy= 48.6 cc, IDLV20Gy= 78.33 cc, IDLV 15Gy “ 138.13 CC, IDLV ioGy — 268.89 cc, IDLVsGy — 677.83 cc. The initial plan was done with a beam set of 340-179 (cw, ccw) and the new plan was done with a beam set of 295-115 (cw, ccw). cw: clockwise, ccw: counter clockwise.

[0040] FIG. 3 illustrates a graphical view of dose volume histograms (DVH) between the two plans shown in FIG. 2. To better illustrate different rates of dose reduction as a function of distance from the target in the two plans, the DVH of inter-connected rings wi th diameters of5 mm, 1 cm, 2 cm and 3 cm are also shown. RingO.5mm encompasses the target while Ring 1cm encompasses Ring 0.5cm and so on. The numeric value of different dosimetric indices used for plan comparison of the two plans are also shown in Table 3.

[0041] FIG. 4 shows the scatter plot of %Δd̄phvs %ΔIDLVX, x = 25, 20, 15, 10, and 5Gy between the initial and new plans for all twenty -five lesions. The linear regression model parameters fitted to each plot were also shown for each isodose line level. As shown, as the difference between the physical depth of the lesion with respect to the two beam set increases, the difference between the volume of the lower isodose lines also increases. In other words, the beam optimization treats the lesion in a shallower depth leading to a decrease in the low er isodose lines’ volume. FIG. 4 also shows the impact of beam arrangement is more pronounced with more distance from the lesion. On average, a reduction of %4.64 ± 4.55 (0.02-21.58, p < 3.8 x 10’5), %5.16 ± 5.54 (0.03-24.68, p < 0.005), %6.46 ± 6.95 (-1.35-29.05, p < 0.009), %12.82 ± 9.06 (0.89-37.65, p < 0.0001), and %14.01 ± 9.87 (1.43-41.84, p < 4.5 x 10'6) was observed in the volume of IDLV25Gy, IDLV20Gy, IDLVisGy, IDLV ioGy, and IDLVsGy, respectively. An average reduction of %20.03 ± 12.30 (3.66 to 45.78) was observed in the dph between the initial and optimized beam sets.

[0042] FIG. 4 illustrates graphical views of scatter plots of %Δd̄phvs %ΔIDLVX, x = 25, 20, 15, 10, and 5Gy between the initial and new plans for all twenty lesions in this study. As shown, as the difference between the physical depth of the lesion with respect to the two beam set increases, the difference between the volume of the low er isodose lines also increases. Hence, the beam optimization brings the lesion to a shallower depth leading to decrease the size of different isodose line levels.

[0043] Similarly, FIG. 5 shows the scatter plot of %Δd̄phvs MLD, V20Gy, VlOGy and V5Gy, and the regression model parameters between the initial and new plans, respectively.As shown, similar trends can be seen for various isodose line levels in the lung tissue as well. On average, a reduction of %6.25 ± 2.95 (0-12.98, p < 2.23 x IO’9), %6.98 ± 5.21 (1.14-21.2, p < 0.0002), %15.03 ± 6.94 (2.59-31.50, p < 7.6 x IO’9), and %9.86 ± 7.54 (1.11-32.49, p < 3.4 x IO’7) was observed in the MLD, V20Gy, VlOGy and V5Gy, respectively.

[0044] FIG. 5 illustrates graphical views of scatter plots of %Δd̄phvs %ΔMLD, %ΔV20Gy. %ΔV10Gy, and %ΔV5Gy between the initial and new plans for all twenty-five lesions in this study. It can be inferred from this figure that as depth of the lesion with respect to the beam set increases, the difference in the mean lung dose, V20 and V10 also increases.

[0045] FIG. 6 shows the boxplots of dosimetric indices between the initial and plans with beam optimization (new) for all 25 cases. Statistical comparison between the two plan cohorts is also shown in Table 3.

[0046] Finally, examples of new plans with different %A dpi, are shown in FIG. 7. FIG. 7 illustrates image views of initial and plan wi th beam optimization for cases with different changes in lesion depth with respect to beam-eye-view. As shown, as changes in the lesion depth between the tw o plans decreases, the difference between the size of the two plans also decrease. The IDLVsoy for each case is as follows: A-Initial = 1415 cc, A-WBO = 884 cc, B-Initial = 1430 cc, B-WBO = 1208 cc, C-Initial = 1143 cc, CWBO = 1034 cc, D-Initial = 736 cc, D-WBO = 726 cc.

[0047] FIGS. 8 and 9 illustrate flow diagrams of the dose prediction model of the present invention. Both show an iterative AI driven process for modifying the treatment plan based on plan properties and outcome goals for 3D dose distribution and prediction. Variable properties include, but are not limited to beam setting, isocenter, energy, and optimization setting. As shown in FIG. 8, a set of class-solution radiotherapy techniques are applied at different stages of the disease to build a standard plan database as the backbone of training the AI models. The present invention provides a semantic approach to develop a new' AI-driven system using physics principles embedded in a set of standard planning database. For each stage of the disease, a class-solution technique is developed and used to provide sufficient cases in the plan repository'. The plan repository is then used to train the AI-driven dose mapping model to predict the appropriate three-dimensional (3D) dose distribution for new cases. When a new patient comes in, the components of radiotherapy plan such as treatment modality (IMRT vs VMAT), beam characterization, optimization setting etc. are calculated based upon the class-solution principles and used in the standard optimization setting to achieve the 3D-dose distribution predicted by the AI model.

[0048] FIG. 10 illustrates a schematic diagram of dose optimization steps according to the present invention. After selection of a proper treatment modality using the beam optimization technique, target and OARs doses are optimized in a two-step process: MLC initialization and prioritized optimization. The AI-based dose mapping model of the present invention is a modulation learning tool to receive an initial plan generated by initialization of optimal beam set to treat the target and retrieve a modulated plan that optimize the target and all OARs dose.

[0049] The present invention provides an AI-based system and method to standardize the radiotherapy of lung cancer at various stages with applications in clinics with different levels of expertise and resources. This present invention is distinguished from others by several distinct features.

[0050] First and foremost, the present invention provides a set of unprecedentedclass-solution techniques as the fundamental building blocks of the system of the present invention. Given the dosimetric goals by the clinicians for a patient, each class-solution technique will provide a personalized set of plan properties including the choice of treatment modality either as VMAT or IMRT, optimal beam angles, predicted 3D-dose distribution corresponding to the beam arrangement and the planning objectives needed to achieve thedesirable treatment plan. This information can be used by either planners or automated systems to generate the final plan. Secondly, the present invention includes a set of standard planning datasets for various treatment modalities at different stages of lung cancer using the principles of the newly developed class-solution techniques. One of the major drawbacks of developing knowledge-based and automated system is the lack of a standard planning dataset. The new datasets associated with the present invention have the potential to mitigate and even eliminate the existing inconsistency of treatment plan quality among various dosimetrists and clinics leading to enhance the quality and robustness of the newly developed automated systems. Thirdly, the AI-driven technology of the present invention is capable of predicting the actual three-dimensional (3D) dose distribution in addition to 2D dose volume histogram (DVH). Therefore, the clinicians are able to assess the predicted organs at risk (OARs) dose before starting the planning process giving them an ability to modify their clinical objectives much earlier in time. This feature will have the potential to facilitate and expedite the treatment planning significantly. Lastly, the present invention will be a hybrid modulated technology integrating the radiotherapy physics principles into an AI-assisted predictive model wherein each individual component of this new system can be improved individually. Therefore, while the entire framework of the underlying technology remains unchanged, each institution has the ability to customize it per their need. For example, if an institution wishes to develop a new set of class solution techniques, the new technology can be adapted by replacing the standard dataset with a new one based upon their new principles.

[0051] In the radiotherapy workflow, the patient initially meets with the radiation oncologist to assess the disease. Depending on the tumor type, stage and volume, different dose and fractionation regimens are used to treat the patient. Stereotactic body radiotherapy (SBRT), conventionally-fractionated radiotherapy (CFRT). hypo-fractionated radiotherapy (HFRT).and twice-a-day regimen (BID) are the main definitive treatment schemes commonly used to treat lung cancer at various stages, as illustrated in FIG. 8. 3D-conformal radiotherapy (3DC-RT) is another treatment scheme. However, 3DC-RT is mostly used in a palliative setting with lower doses. Because the goal of the present invention is to reduce toxicity and enhance the treatment outcome, with focus on definitive treatments typically given to patients through advanced radiotherapy techniques such as intensity modulated radiotherapy (IMRT) or volumetric modulated arc therapy (VMAT). These techniques involve a time-consuming and complex process known as plan optimization where a planner works to balance the radiation dose among various OARs while giving a tumoricidal dose to the target. In IMRT, the radiation dose is conformed to the target by spatially modulating the fluence map at specific angles via independent but concurrent sliding of multi-leaf collimators (MLC) across the treatment field. In VMAT, the radiation dose is conformed to the target by continuously reshaping the MLC to match the target geometry while rotating around the patient. Therefore, in cases where the direction of irradiation differs substantially for treatment planning, the IMRT is the preferred choice. In contrast, where no significant angular preference exists, VMAT should be used. Studies have shown that both IMRT and VMAT are effective in radiotherapy of lung cancer; however, no established criteria currently exist for making a personalized selection between the two modalities. This is primarily due to the variation in lesion location, the presence of substantial tissue heterogeneity and the involvement of various organs at risk within the thorax all of which make the treatment process complex. The present invention standardizes lung radiotherapy by establishing a set of personalized class solution techniques and disseminating these techniques through Al-models.

[0052] A new approach to quantify the beam optimization prior to plan optimization in stereotactic body radiotherapy (SBRT) of the peripheral lung lesions is described herein. The new approach is based on maximizing the therapeutic gain of the beam set by minimizing theaverage physical depth of the lesion with respect to the beam's eye view. This approach is assessed primarily in lesions treated with stereotactic body radiotherapy because SBRT is usually applied to small lesions for which lesion center or isocenter is a good estimation of the lesion distance to the patient surface where beams enter. Lowering the depth of the treatment with respect to the beam’s eye view7also lowers the overall integral dose of the plan. This is the first attempt to develop a quantitative metric to determine the optimal beam arrangement for lung SBRT. Therefore, to overcome the lack of prior metrics for comparison, a set of plans that were used clinically and approved through a very7rigorous peer review7process were used as the base cohort. Therefore, the base-plan set had no bias regarding the value of the proposed metric.

[0053] The effect of beam optimization is assessed by measuring the size of the plans at different isodose line volumes rather than comparing the dose in various organs at risk (OARs). This was mainly done because the OARs dose is a subjective metric for plan comparison and varies peruser discretion and patient to patient. In other words, two plans might have similar size but substantially different dose distribution per planner’s or physician’s demand. Also, to make sure that beam optimization does not compromise the safety of any OAR in the new plan, the initial dosimetric objectives were used as the clinical goals and only accepted the new plan if these clinical goals were met. For example, it may be believed that new beam arrangements may increase the skin dose for some condition, but the results showed that beam optimization does not compromise the safety of any7OARs in the new plan. In fact, the beam optimization decreases the size of the plan at different isodose line levels and it’s the planner role to guide the optimizer how to distribute the dose. In the current dataset, an average reduction of ~%3.07 ± 2.69 (0.05-8.94, p < 3.3 x 1 O’5) was observed in the skin maximum dose. Lastly, it's worthwhile to note that the new plans were accepted only if their total monitor units (MUs) were less than the initial plans. Because eachround of optimization can increase the total number of MUs and could potentially improve the plan, this assumption was made to assure that any improvement in the plan quality7was achieved due to beam optimization rather than overmodulation. An average reduction of ~%10.24 ± 10.24 (0.18-46.06, p < 0.0002) was observed in the total MUs of the new plans. While reduction of MU may not sound as important as OARs dose reduction, this could significantly increase the machine throughput in slow-deli very machine like MRI-Linac and Cobalt-device machine. The increased throughput has the potential to improve access to treatment in low- and middle-income countries with limited resources. These countries are also projected to have a higher burden of cancer in the coming years. Efforts are underway to bring IMRT-like treatment plans using cobalt devices at these resource-limited settings.

[0054] One important feature of the beam optimization approach proposed in this work is that minimizing the average physical depth of the lesion with respect to the beam entrance substantially decreases the lung tissue exposure before the beam reaches the target as well, as illustrated in FIGS. 2 and 7. Because radiation-induced lung injuries are the main limiting toxicity7in radiotherapy of the lung lesions, this approach has the potential to lower the lung dose, FIG. 5, and injuries systematically. Also, accurate dose calculation has been a historical challenge for lung SBRT due to the low density of lung structure and difficulty of small field dosimetry associated with the small targets. Studies have shown that different algorithms such as analytical anisotropic algonthm (AAA), collapse cone convolution (CCC) and Acuros XB (AXB) can show substantial dosimetric discrepancies in case of lung SBRT compared to Monte Carlo simulation and measurements. One crucial point to consider in such cases is that the increased volume of lung in the beam trajectory will increase the dosimetric difference, an expected decrease in PTVD 99% BED of 1.6% per 500 cc. Because in the proposed beam optimization, the beam geometry is defined in a way that lower lung tissue is exposed before the beam reaches the target, this may also help reduce the uncertaintyof dose calculation for PTV coverage. In addition, since many of the lung patients are treated with free breathing with possible effect of diaphragm or even pericardium movement in dose calculation, limiting the lung exposure could also mitigate the impact of internal motion in PTV dose calculation’s uncertainty as beams hit the target before reaching the dynamic components of the thorax. Quantifying this feature is part of future study. One of the main challenges to obtain the ideal beam arrangement in the proposed approach is when the lesion seats posteriorly, as illustrated in FIG. 7, a set of full arcs with avoidance sector or split the fields into two half arcs may be needed to treat the lesion. The first scenario increases the chance of collision and for the second scenario, the couch may need to be moved between the arcs, and this may add extra minutes to the delivery time or may require extra imaging scans. However, if the gantry can move freely around the patient, both ways can be followed seamlessly with an efficient delivery time. Therefore, developing a collision metric sounds essential for the posteriorly located lesions. In cases where the gantry can’t rotate around the patient freely, beam optimization should include this as part of the process and make a balance between the collision and optimal beam selection. Luckily, the chance of collision increases when the lesion gets closer to the lateral part of the lung but in this case the role of posterior field in forming the optimal beam set decreases. In contrast, when the posterior lesion seats more medially, the chance of collision decreases and the impact of posterior field in forming the optimal beam set increases. One other solution for such a problem is to use an off-axis isocenter for such lesions. In new treatment modalities such as MRI-Linac, PET-Linac, and Tomotherapy where off-axis treatment is part of the routine practice, this approach can be easily applied. However, in regular Linac where off axis treatment is not routinely practiced, the potential drawbacks of such treatments should be evaluated comprehensively. While the plan quality of off-axis treatment might be comparable to on-axis treatments, the accuracy of the dose calculation of the treatment planning system should be entirely assessedin radiotherapy of the thoracic malignancy where substantial tissue heterogeneity exists. Developing the collision metrics and assessing the off-axis treatment for such lesions are currently underway. It's worthwhile to note that in the current study, a significantly bigger change (p < 0.044) was observed in the treatment depth of the posterior lesions (%24.68 ± 13.24) in contrast to anterior lesions (%15.00 ± 2.07) after beam optimization which can justify the effort of developing collision metrics and off-axis treatment.

[0055] One other aspect of the proposed beam optimization approach is to use a set of half arcs for SBRT of the peripheral lung lesions. Such selection is important as ahemi arc includes beam angles in all directions which is crucial to achieve appropriate conformality for SBRT treatment. In one hand, any arc greater than 180°can increase the depth of treatment, due to inclusion of beam angles with larger physical depth into the beam set, and the plan size subsequently with no actual benefit. In other words, increasing the number of control points to a set of hemi-arcs may not necessarily result in a better plan but could increase the total monitor units (MU), the deliver}' time and more importantly the plan size. In the other hand, using arc shorter than 180 can compromise the conformality of the plan with an increase in the high dose volumes around the target. In addition, several studies have shown the benefit of couch kicks in the case of SBRT treatment. Since applying the couch rotation can also increase the chance of collision, a nsk / benefit analysis can be performed by measuring the change in the lesion depth with respect to the beam’s eye view and make an appropriate decision in such cases. It’s worthwhile to note that all of these can be achieved by eliminating the need to test different beam arrangement settings in the actual planning process which has the potential to reduce the planning time significantly. In addition, as automation now plays a key role in radiotherapy practice and especially in busy clinics, having a quantified metric in selecting the appropriate beam arrangement is a major step in developing a robust automated workflow. In this study, the beam optimization approach was assessed using a small samplesize. Reassessing this concept with a larger plan cohort warrants the appropriateness of this technique for development of automated workflow and routine clinical use.

[0056] Beam optimization was done in this work based on the assumption that lesion center is a good estimate of the lesion distance to the patient surface where beams enter. This assumption can be true for small lesions but as the lesion size increases more sophisticated metric will be needed to measure the distance of the lesion to the site of beam entry. In contrast, beam optimization might have bigger impact on larger lesions since the radiotherapy plans of such lesions are much bigger and might have shallower dose fall-off. Therefore, a small reduction of plan size could generate bigger reduction in absolute volume of the plan compared to SBRT cases.

[0057] It is also worthwhile to note that collision is of less concern for larger lesions due to flexibility of isocenter placement and higher tolerance of error on setup margin. Beam optimization of large lesions and non-SBRT cases is underway. Also, it was assumed that all voxels beyond the PTV are equally important from the radiation-induced toxicity perspective and lowering the overall integral dose and lung tissue exposure was the main goal of beam optimization.

[0058] Approximation of therapeutic gain with the treatment depth was based on the assumption that the distance of the lesion center to the patient surface is a good estimate of the treatment depth from the beam’s eye view. This assumption can be true for small lesion but as the lesion size increases, more sophisticated metric is needed to estimate the therapeutic gain of a beam for a particular lesion. Also, in SBRT. all voxels assumed to be beyond the PTV are equally weighted in estimation of the therapeutic gain and lowering the overall integral dose and lung tissue exposure was the main goal. However, in treating large lesions, different OARs might be at different risk for each patient requiring personalizedunequally -weighted optimization to achieve the optimal beam setting for each case.Therefore, the therapeutic gain of a beam for a large lesion is estimated as follows:Tr,,.= / H TMRTumor(be)dv[Wlik 6)and i G {Lungs, Heart, Esophagus, Airways, Spinal Cord, Chest wall and... } wherein Wi is associated with the relative toxicity of the ith OAR in the patient and ranged between 0 and 1. e.g., if the receiving dose of an OAR has minimal effect on the organ’s toxicity, w:would be 0. In case where OAR's dose imposes significant toxicity, it would be set to 1. The main objective in beam optimization of the large lesions is to find {w for each patient.

[0059] When a patient meets with Radiation Oncologist, the patient’s disease and health status are thoroughly assessed by the clinical team to select an appropriate treatment for the patient, as illustrated in FIG. 9. The patient is then sent for treatment planning and the clinical team defines a set of dosimetric constraints (clinical goals) for each OAR to assure no or minimal toxicity is associated with the patient’s treatment. These dosimetric constraints are used in the beam optimization approach to obtain {iv,}. To this end, a set (180) of evenly- spaced beams, 2° apart, is initially arranged around the patient. Assuming each beam delivering the same monitor unit (MU) initially, a dummy 3D conformal plan is generated to achieve the tumoricidal dose for the lesion. The dose volume histogram (DVH) for each OAR is then calculated to create an optimization objective function, / (OAR, C), as shown in the following equation in which OAR1j is the / th constraint dose for zth OAR and Cij is the jth dosimetric constraint for zth OAR defined by the clinical team. Initially set to 1 for all OARs, {iv,} are updated iteratively using the following equation until no further changes are achieved for any w,. Using the updated {w,}, the therapeutic gain for each beam angle is also updated. The updated TG{wi}(be) at each angle is used to update the corresponding MU as also shown below. This procedure is repeated until no further update is achievable for {w;},as illustrated in FIG. 5. It’s worthwhile to note that as the lesion size decreases, the chance of violating the dose constraints also decreases for various OARs. and a point may be reached that no weighting preference would be achieved through this approach for the lesion. In that scenario, the beam optimization is equivalent to the beam optimization of small lesion and can be achieved by minimizing the treatment depth of the lesion from the beam’s eye view.

[0060] Studies have shown that both VMAT and IMRT are effective in the radiotherapy of lung cancer. In a retrospective study, Wei et al demonstrated that VMAT is a preferred choice for lung SBRT. In a separate study, Li et al. reported that both IMRT and VMAT offer advantages in lung radiotherapy. Therefore, it is essential to develop a robust metric for personalized decision making between the two modalities for each patient. The therapeutic gain curve can serve as a reliable mechanism for such decision-making.

[0061] When the lesion is small, the contributions of all OARs to the therapeutic gain curve are roughly equivalent and shown to be approximated as a function of treatment depth vs beam angle. In such cases, the therapeutic gain (TG) reaches a global minimum at a specific angle and increases as the angle deviates from the optimal one. In this scenario, VMAT is the preferred approach, with the optimal beam set consisting of half-arcs centered around this optimal angle. In contrast, using IMRT with specific orientations for such conditions comprises only a subset of optimal beam angles used in VMAT reducing degree of freedom in the beam set hence leading to an inferior plan compared to VMAT.

[0062] As the lesion size increases, the depth-angle curve no longer represents the therapeutic gain-angle curve accurately primarily because the lesion boundaries may be close to the patient's surface at different orientations. Additionally, the denominator of the therapeutic gain equation may need to incorporate different weightings for various OARs. As a result, the beam optimization approach introduced for large lesions will likely achieve multiple extrema indicating that IMRT could be a better choice compared to VMAT. In suchcases, using split arcs around these local maxima could also be considered as a substitute for the VMAT modality. However, if the contribution of various OARs in the beam optimization does not introduce any new local extremum, VMAT would still be a preferred modality.

[0063] In situations where the therapeutic gain cur e exhibits multiple local extrema around the patient, a hybrid approach using both VMAT and IMRT can be employed to create a base and complementary plan, with VMAT and IMRT serving these roles, respectively. While the VMAT component ensures that all angles contribute to generating the optimal plan for the patient, the IMRT component ensures that sufficient control points are dedicated to angles with the best therapeutic gain, hypothetically leading to a better OARs sparing.

[0064] IMRT and VMAT use inverse planning multi-objective optimization (MOO) techniques to determine complex beam configurations that distribute radiation doses to the patient, delivering the tumoricidal dose to the target while minimizing the radiation exposure to the surrounding OARs. While an increasing number of algorithms have been introduced for MOO, the most common strategy is to use a single-objective function constructed from the weighted sum of multiple objectives associated with the clinical goals, where the weights denote the importance of different clinical objective. Traditionally, gradient-based algorithms have been used for optimization of single-objective function requiring planner intervention in a trial-and-error manner until an acceptable trade-off is achieved. The main disadvantage of this approach is its subjectivity which can vary from planner to planner. One way to address this issue is to use prioritized optimization (PO) which clinical objectives cannot be traded off against each other, at least not in a numerically traceable way. Instead, PO considers finite number of objective functions to be optimized on a feasible set of solution in a quantitatively prioritized order. In other words, in PO, lower priority objectives are optimized only as long as they do not interfere with the optimization of higher-priority objectives. The main advantage of this approach is that it generates an optimal solution at each sequential stageindependent of planner intervention as long as the prioritization remains unchanged. In contrast, PO is criticized for preventing large gains in low-priority goals when trading off a small loss in high-priority7goals. To address this issue, the slip factor allows a small degradation in higher-priority7goals to benefit low-priority7objectives. In other words, the constraints on the algorithms for high-priority7goals are relaxed in order to provide more flexibility' for optimizing lower-priority' goals. Additionally, the large-scale constrained optimization, such as that in radiotherapy planning with hundreds of variables (e.g., beamlet intensities in IMRT and multi-leaf collimators in VMAT), is highly computationally intensive. Therefore, dimensionality' reduction techniques 160 are also essential for reducing both the complexity and time required to solve the constraint optimization.

[0065] Assume that Dr denotes the prescribed dose to the PTV and = {^i, y2. ••• < / „ 1 represent the set of clinical goals in the optimization setting where the subscript indicates the importance of each objective with yi being the most important objective and ^k-i being always more important than yk. The solution1= {is better than y2= { g- if and only' if <and = g1; = min(^j) for certain i < n and all j < i. To solve this problem, using beam optimization approach explained above, the modality of treatment and the optimal beamset are initially selected. With an unconstraint optimization, the selected beam set is hen initialized by minimizing the following single-objective function:(Di ~ DT) finittr, OAR) = arg min<or2NTy where NT denotes the number of voxels in the target and Dj ) is the mean dose of OAR j and o)Tand a>j are relative weights for the PTV and OARs, respectively and Z < = l The main purpose of this step is to initialize the fluence map and MLC set to reduce the search space to beamlets and MLC configuration that treat the target and minimally spare the OARs. Thisstep is very similar to a 3DC-RT plan but with slight degree of modulation. In the next steps, the maximum and volumetric dose for OAR1, OAR2,... and OARmwill be minimized according to their priority order defined by the clinical team.

[0066] Depending on the tumor type, stage, and history of prior radiation, different treatment regimens and prescriptions (Rx) are used to treat the thoracic malignancy. The tolerance dose are also used as guidelines to manage the normal tissue complications. Common Terminology Criteria for Adverse Events (CTCAE) guidelines are also followed to report and record the side effects of radiation therapy. To create standard datasets, the radiotherapy treatments are categorized into 7 major groups based on the treatment regimens and prescriptions. The BID regimens are classified into two subcategories: 1)-initial treatment for SCLC and 2) re-RT for all other types of lung cancer. The conventionally fractionated regimen is also classified into three classes including the stage III NSCLC, esophageal cancer and the rest of lung cancer types. The classification is mainly done based on the similarity of the normal tissue constraints used to design the treatment plan. For each category of treatment, plans are generated using the treatment planning process explained above.

[0067] Due to the significant variation in the geometry of the target and its surrounding OARs in thoracic lesions, the radiotherapy plan for such cases differs substantially from one another, often making it impossible to meet all dose constraints. When considering the patient's health status and vanations in lung volume, the complexity of the issue increases further. Therefore, superior knowledge and expertise are required to develop an appropriate treatment plan for thoracic lesions. It has been shown that knowledge-based planning and dose prediction models are valuable tools for improving the quality of radiotherapy plans. Using a dose prediction model developed with a standard dataset can not only expedite and facilitate the radiotherapy process but also enhance the quality of radiotherapy plans for thoracic patients. Hence, using the standard datasets developed in this project and comprisevariety of high-quality plans, seven artificial intelligence (AI)-driven dose mapping models for different categories of radiotherapy plans explained in the previous section. In short, the dose prediction models of the present invention map the unmodulated plan achieved by initializing the fluence map / MLC set in the dose optimization step to a modulated plan, providing a recommended 3-dimensional dose distribution for each case. In other words, the AI-based dose prediction models of the present invention take the optimal beam set as input, create an unmodulated plan using fluence map / MLC set initialization, and then provide a 3D modulated plan as a recommended-dose distribution for each case.

[0068] While modern treatment planning systems offer powerful tools to help planners create the desired plan, they may not always provide sufficient guidance in developing the optimal treatment for every patient. In contrast, the dose prediction model, developed using a standard dataset and comprising of variety of high-quality plans, is hypothesized to offer three dimensional dosimetric guidance to all planners, regardless of their expertise, and provides a template to guide the planner in achieving the desired dose distribution. To this end, the dose volume histogram of the predicted dose will define the plan objectives for new patients, and the planner's role will be to replicate the predicted dose or improve upon it if possible. This will be accomplished while the proposed beam optimization module of the present invention provides the appropriate beam arrangement and treatment modality of each individual patient.

[0069] While the primary goal of the present invention is to provide a solution for standardization of lung radiotherapy, automating the entire process will also help expedite planning, reduce time to treatment and provide assistive technology in clinics with limited dosimetry and physics personnel. Given that all components of the above technology are designed to require minimal user input, the ultimate objective is to fully automate the planning process. To achieve this, the system of the present invention receives prescription and clinical goal information from the medical team, providing a personalized radiotherapyplan through a series of automated tasks, including beam optimization, treatment modality selection, AI-guided plan objective generation, and plan optimization for each patient.

[0070] The present invention consists of three levels of validation. The first step is to assess the impact of beam optimization in designing an appropriate treatment for each patient, and this will be done retrospectively. The second step will evaluate the suitability of dose prediction models to predict the appropriate dose distribution for each case and study whether such models can improve the quality of the plan. In the final stage, end-user validation is performed to assess the quality of the automated workflow in designing an appropriate plan for each patient and to understand the strengths and limitations of the new technology.

[0071] However, different organs at risk (OARs) might be at different risk requiring unequally -weighting optimization which can also be more pronounced in radiotherapy of large lesions where toxicity of other organs such as heart, esophagus, etc. is a limiting factor. One other important note in that regard is that while VMAT is the treatment modality for SBRT cases, intensity modulated radiotherapy (IMRT) could outperform VMAT for larger lesions. Therefore, a smart decision on the selection of VMAT vs. IMRT is also essential to design the best treatment for lung lesions. Depth-angle curve is a good approximation of the therapeutic gain-angle curve for small lesions. Because the patient surface is approximately elliptical, the depth-angle curve usually reaches a global minimum as shown in FIGS. 1A-1C and FIG. 2 and the depth increases as distance from this angle increases. Therefore, for small lesions with a unique global extremum, choosing a shortest half arc around that point will provide the best beam-set. Hence, using IMRT with certain orientations for small lesions is just a subset of optimal beam angles used in VMAT with reduced degree of freedom in the beam-set leading to an inferior plan compared to VMAT. In contrast, as the lesion size increases, the depth-angle curve cannot represent the therapeutic gain-angle curve anymore as the lesion boundaries could be close to the patient surface at different orientations. Inaddition, the denominator of equation (1) may also need to include different weighting for various OARs as well for larger lesions. Hence, the therapeutic gain-angle curve will likely have multiple extrema for large lesions for which IMRT could be a better choice compared to VMAT. In such cases, using split arcs around various local extrema could also be used for VMAT modality. In close, the introduction of therapeutic gain concept not only provides the best beam arrangement for lung RT but also has the potential to guide us to select an appropriate choice of treatment modality leading to a more personalized radiotherapy for lung cancer patients.

[0072] In lung radiotherapy, the treatment outcome is mainly limited by the radiation induced lung injuries which is highly correlated with its radiation exposure. This could also depend on the planner’s skill and experience and varies significantly among different dosimetrists and even treating physicians. Treatment standardization has the potential to remove inter variability among various RT teams by generating a robust treatment plan in a timely fashion. The present invention provides a new metric, treatment depth with respect to beams’ eye view, for optimal beam selection as the first step in standardization of lung stereotactic body radiotherapy to minimize the radiation-induced lung toxicity and potentially decrease the time to treatment. The new metric can also be used to predict the quality of the plan in time. Further evaluation warrants the utility of such metric for routine clinical use and automated workflow.

[0073] It should be noted that aspects of the system and method, its control, and calculations can be executed with a program(s) fixed on one or more non-transitory computer readable medium. The non-transitory computer readable medium can be loaded onto a computing device, microprocessor, servo, server, actuator, device processor, smartphone, tablet, phablet, a Control Box, or any other suitable device known to or conceivable by one of skill in the art. Additionally, the non-transitory computer readable medium could be incorporated into thecontrol unit for the radiation delivery device, or other control units within the treatment room. Any other suitable configuration known to or conceivable to one of skill in the art could also be employed for the execution of the present invention.

[0074] It should also be noted that herein the steps of the method described can be carried out using a computer, non-transitory computer readable medium, or alternately a computing device, microprocessor, or other computer type device independent of or incorporated with the radiation detection device. The computing device for executing the present invention can be a completely unique computer designed especially for the implementation of this method. Indeed, any suitable method of analysis known to or conceivable by one of skill in the art could be used. It should also be noted that while specific equations are detailed herein, variations on these equations can also be derived, and this application includes any such equation known to or conceivable by one of skill in the art.

[0075] A non-transitory computer readable medium is understood to mean any article of manufacture that can be read by a computer. Such non-transitory computer readable media includes, but is not limited to, magnetic media, such as a floppy disk, flexible disk, hard disk, reel-to-reel tape, cartridge tape, cassette tape or cards, optical media such as CD-ROM, writable compact disc, magneto-optical media in disc, tape or card form, and paper media, such as punched cards and paper tape.

[0076] It should be noted that the software associated with the present invention is programmed onto a non-transitory computer readable medium that can be read and executed by any of the computing devices mentioned in this application. The non-transitory computer readable medium can take any suitable form known to one of skill in the art. The non-transitory computer readable medium is understood to be any article of manufacture readableby a computer. Such non-transitory computer readable media includes, but is not limited to, magnetic media, such as floppy disk, flexible disk, hard disk, reel-to-reel tape, cartridge tape, cassette tapes or cards, optical media such as CD-ROM, DVD, Blu-ray, writable compact discs, magneto-optical media in disc, tape, or card form, and paper media such as punch cards or paper tape. Alternately, the program for executing the method and algorithms of the present invention can reside on a remote server or other networked device. Any databases associated with the present invention can be housed on a central computing device, server(s), in cloud storage, or any other suitable means known to or conceivable by one of skill in the art. All of the information associated with the application is transmitted either wired or wirelessly over a network, via the internet, cellular telephone network, RFID, or any other suitable data transmission means known to or conceivable by one of skill in the art.

[0077] The many features and advantages of the invention are apparent from the detailed specification, and thus, it is intended by the appended claims to cover all such features and advantages of the invention which fall within the true spirit and scope of the invention.Further, since numerous modifications and variations will readily occur to those skilled in the art, it is not desired to limit the invention to the exact construction and operation illustrated and described, and accordingly, all suitable modifications and equivalents may be resorted to, falling within the scope of the invention.

Claims

What is claimed is:

1. A system for radiation planning comprising:a processing device, wherein the processing device is programmed for:assessing tumor geometry;assessing patient anatomy;determining a 3D dose distribution based on the tumor geometry and patient anatomy;determining a personalized radiotherapy plan using predicted plan properties and the 3D dose distribution.

2. The system of claim 1 wherein the tumor is identified as lung cancer.

3. The system of claim 1 further comprising assessing a treatment stage for the patient.

4. The system of claim 1 further comprising a database of class solution techniques and predictive dose mapping models.

5. The system of claim 1 further comprising generating predictive dose mapping models for different types and stages of tumor.

6. The system of claim 1 further comprising receiving user input related to the treatment.

7. The system of claim 1 further comprising determining a beam arrangement for the treatment.

8. The system of claim 7 further comprising predicting a 3D-dose distribution corresponding to the beam arrangement.

9. The system of claim 1 further comprising a planning dataset.

10. The system of claim 9 wherein the planning dataset is directed to a treatmentmodality.

11. The system of claim 9 wherein the planning dataset is directed to a type and stage of cancer.

12. A method for radiation planning comprising:assessing tumor geometry, with a processing device;assessing patient anatomy, with the processing device;determining a 3D dose distribution based on the tumor geometry and patient anatomy, with the processing device;determining a personalized radiotherapy plan using predicted plan properties and the 3D dose distribution, with the processing device.

13. The method of claim 12 further comprising assessing a treatment stage for the patient.

14. The method of claim 12 further comprising accessing a database of class solution techniques and predictive dose mapping models.

15. The method of claim 12 further comprising generating predictive dose mapping models for different types and stages of tumor.

16. The method of claim 12 further comprising receiving user input related to the treatment.

17. The method of claim 12 further comprising determining a beam arrangement for the treatment.

18. The method of claim 17 further comprising predicting a 3D-dose distribution corresponding to the beam arrangement.

19. The method of claim 12 further comprising generating a planning dataset.

20. The method of claim 19 wherein the planning dataset is directed to a treatmentmodality.

Images