ABSTRACT
In modern radiotherapy, multiple courses of radiation are becoming increasingly common as a treatment regimen to extend progression-free and overall survival in patients with oligometastatic disease. However, normal tissue recovery over time has not been well characterized, and there are few models for clinicians to use when evaluating potential toxicities in subsequent radiation treatments. The lack of standardization when documenting a patient’s radiotherapy history presents a major barrier to conducting large scale studies. To advance our understanding of normal tissue recovery post-radiation, we propose the addition of a new object accompanied by a suite of mathematical models linked to toxicity information in a patient’s medical record. This object leverages the Digital Imaging and Communications in Medicine (DICOM) standard to serve as a centralized data store for radiotherapy planning and treatment, thereby facilitating a better analysis of therapeutic outcomes and tissue response over the course of radiotherapy.
1 Introduction
With evolving treatment paradigms for cancer leading to better prognoses for those with widespread disease, we can now place oligometastatic (i.e. limited in location and number) and widely disseminated metastatic disease on a dynamic spectrum1. In addition to novel systemic agents, clinical trials have shown evidence that stereotactic body radiation therapy (SBRT) can extend progression-free and overall survival in patients with oligometastatic disease,2,3. When planning radiation therapy (RT) courses that may involve treating individual sites in an organ recurring over time (metachronous lesions), upcoming RT courses will be affected by courses of previous radiation. Importantly, this requires striking a balance between the therapeutic effect of a local therapy, such as SBRT, and its potential injury to the organs-at-risk (OARs). For SBRT in particular, dose constraint tables developed by Timmerman, and utilized in cooperative group trials, are widely considered to be the “gold standard” for describing radiation dose limits for different OARs above which the risk of developing tissue complications significantly escalates4.
Another well-studied approach to estimating treatment safety for single courses of radiation therapy is the Lyman-Kutcher-Burman (LKB) model, which calculates toxicity probability as a function of radiation dose received by organ. This model has successfully predicted the toxicity from conventional fractionated RT to multiple OARs but only accurately predicts SBRT tissue toxicity in a limited number of cases5. Generally the LKB model is too conservative in its predictions of the maximum tolerable dose for SBRT. Alternative models exist to predict tissue toxicity; however, these models were developed with relatively limited data, so it remains to be seen how accurate they are6.
In the context of sequential SBRT regimens, one current commonly adapted strategy to assess the safety of administering multiple RT courses involves the calculation of a cumulative dose map. This is done by superimposing prior radiation doses onto the current SBRT plan using image registration techniques. When accumulated dose calculations suggest OARs could potentially receive a total dose exceeding commonly accepted constraints, clinicians must decide whether to alter the RT dose and potentially compromise local tumor control or to proceed treating with an increased risk of toxicity. When considering such choices, some radiation oncologists accept that with sufficient time between radiation courses, some radiation dose can be “forgiven” or “forgotten”7 This comes from prior animal studies demonstrating some normal tissue recovery and a higher tolerance for radiation delivered over two courses separated by time6,8.
With this in mind, we foresee clinical scenarios where treatment with multiple courses of radiation will become increasingly common, especially in light of recent work directed at better understanding the effect of oligometastatic-directed therapy on patient survival9,10. As such, physicians need a way to consistently and reliably account for dose accumulations over multiple courses of radiation, in order to and to reproducibly estimate normal tissue recovery. To achieve this, we propose developing a reference Digital Imaging and Communications in Medicine (DICOM) object that represents a virtual model of the patient, onto which all radiation plans are mapped at the voxel level at the time of delivery. This tool, together with mathematical models of RT dose forgiveness over time, will allow clinicians to factor tissue recovery and response into the planning of multiple RT courses, as well as to collect data on the efficacy and toxicity of specific treatment plans. The models of dose “forgiveness” can be as conservative as simple arithmetic addition of doses across treatments, or as complex as combining patient specific genomic data? with tissue-specific recommendations from the literature. Given the inherent uncertainty of such modeling, the tool is designed to provide structured guidance to help radiation oncologists standardized their approach to understanding the risks and benefits of re-irradiation.
2 Models and Methods
The process of creating a temporally corrected dose accumulation (TCDA) plan begins with selecting the model of tissue recovery to use. If one assumes no recovery and uses arithmetic addition of doses across treatments, subsequent radiation doses are restricted to not exceed the maximal tolerable limits. Given the proven survival benefit of treating oligometastatic lesions with SBRT2,3,11–13, this conservative assumption will often limit our ability to continue treating if multiple courses are required. Furthermore, there exist data suggesting that the assumption of “no tissue recovery” is too conservative. In one study, primates were able to tolerate an extra 37.2 Gy EQD2 (equivalent dose in 2 Gy fractions) of cumulative radiation before toxicity than would be predicted by no recovery over an intervening two years14. Clinical data also indicate surprising resilience of spinal cord tissue to radiation myelopathy upon re-irradiation of tissue.15,16 Even with these studies, a precise way to model the RT dose forgiveness in the spinal cord, let alone other tissues, is not clear. This uncertainty makes rigorous study of radiation toxicities difficult.
Despite the dearth of data, some guidelines do exist for repeat radiation treatment. One review posits that for spinal cord SRS, re-irradiation EQD2 should not exceed 25 Gy and cumulative EQD2 should not exceed 70 Gy8. However, the authors also note the limited case control data available and write that these guidelines are likely conservative. A meta-analysis of patients with recurrent rectal cancer found that re-irradiation using total cumulative doses ranging from 66.4 to 103.3 Gy was well tolerated, with acute and late grade 3-4 toxicity rates of 11.7% and 25.5% respectively17. SBRT re-irradiation for rectal cancer appears especially promising in eligible candidates, with one study showing high response and low toxicity using a 30 Gy dose in 5 fractions18. A broader review examined re-irradiation across multiple tissue types and concluded that re-irradiation is feasible using either a hyperfractionated or stereotactic approach19. Given the high threshold for cumulative dose without toxicity in a majority of patients, these data indicate that dose forgiveness is a function of both the underlying tissue and cancer type (and likely individual patient biology).
As a starting point for our model, simple mathematical decay functions including linear, exponential, logarithmic, and Gaussian decay can be used when considering subsequent radiation. In reality, a more complex non-linear decay function that combines tissue type, genomics, age, and other patient characteristics will be needed to fully characterize dose forgiveness. This function should be integrated with existing clinical data on tissue-specific maximum tolerable doses. For example, the Quantitative Analysis of Normal Tissue Effects in the Clinic (QUANTEC) review was a landmark effort in creating guidelines for dose/volume/outcome relationships20. By combining dose decay models with dose limits, we create a robust model that computes the maximum tolerable dose of a planned RT course as a function of tissue, time, and dosage of prior radiation therapies. However, gathering sufficient data to develop such a model is not feasible without a standardized method of dose comparison between patients in different institutions and over time. This quintessential dilemma in clinical radiation oncology can be overcome by the creation and adoption of a new data standard that stores a patient’s temporal radiation history and a 3D “body map”, generated from a reference treatment planning medical image such as computed tomography (CT) or magnetic resonance imaging (MRI), as a permanent object in the medical record.
Both the treatment planning image and radiation history can be represented using the widely adopted DICOM medical imaging standard through a pre-existing entity within the standard. DICOM was originally created as a way for radiologists and device manufacturers to generate and share data in a common format21. Over time, the standard has seen numerous iterations and has accommodated radiotherapy information since 1997, with the publication of Supplement 11 by Working Group 07. In 2016, Working Group 06 created a second generation of RT objects in Supplement 147, including “RT course” which is a top-level entity that encapsulates both an RT prescription and physically delivered doses for one or more tumors22.
We propose storing each round of RT that a patient receives as a separate RT course. After each fraction of radiation, we update the current course with the delivered dose and apply image registration to map the dose into corresponding voxels (a three dimensional volume) at specific locations in the underlying virtual model of the patient. The parameters of the image registration process will be stored separately as DICOM files and contain references to the treatment planning image and RT doses which were used in the mapping process. This separation provides flexibility in case the reference virtual model or preferred registration algorithm changes in the future. In a hypothetical situation, if a patient requires re-irradiation, the clinician is able to visualize temporally corrected dose accumulations of prior treatments by selecting from one of multiple algorithms which process the stored RT courses. Once these calculations are performed on the reference virtual model, the corrected doses are then mapped back onto the current treatment planning image and imported into a desired RT planning application. In this way, we are able to leverage the sophisticated optimization algorithms that already exist. Figure 1 gives examples of several decay functions that could be used initially.
The choice of re-irradiation will still be at the discretion of clinician, but such a feature would allow for a more quantitatively informed judgement of tissue recovery, and more rigorous study and comparison of plans and outcomes, and subsequent model refinement. Furthermore, our proposed standard is inherently extensible as the underlying RT courses will be stored for analysis by future algorithms. Figure 2 shows what a future user interface tool might look like. Alternatively, this functionality could be incorporated into existing RT planning or contouring software.
As with most electronic medical recording, while this data format can function as a clinical tool, it is also effective for research since it records treatment plans and predictions. Specifically, this new object in the medical record will store the algorithm used, radiation dosage given, tissue type irradiated, and any toxicity experienced. This is the exact data needed to create more accurate models and guidelines that optimize re-irradiation toxicity.
3 Patient Case
Consider a sample and hypothetical patient, Mr. X, a man with oligometastatic soft tissue carcinoma. Information below, including his history and MRN in the figure are not related to any patient. Mr. X had no significant past medical history and was in his usual state of health until being diagnosed with high grade myoepithelial carcinoma in the left distal thigh. His treatment entailed a large resection with neoadjuvant and adjuvant radiation therapy using a combination of external beam and brachytherapy, which successfully controlled the primary lesion. Nine months later however, he presented with multiple metachronous lesions in both lungs, which arose despite treatment with adjuvant systemic chemotherapy. These lesions were not apparent in his initial course and were treated with 10 sessions of SBRT over approximately two years. The SBRT was successful in controlling the targeted nodules; however, estimating the extent of tissue injury between treatments was challenging. In this scenario, a reference object that maps all of Mr. X’s therapies geospatially onto a reference 3D model might have allowed for easier appraisal of the safety of his treatment plan.
We acknowledge that the lungs would likely not be the first site for application of TCDA due to their parallel nature (i.e. they have many subunits and thus can function well even with prior radiation toxicities). In parallel organs, volume affected by toxicities is more important than prior dose experienced. In contrast, serial organs such as the spinal cord experience a proportional increase in toxicities with accumulated dose. The use of the lung in this proof of concept is primarily to demonstrate the variation in cumulative dose with and without adjustment. Further discussion of parallel vs serial is below in the future directions section.
Figure 3. shows a proof of concept model using the summation of the first four of Mr. X’s SBRT treatment sessions. We only use a subset of the full 10 sessions to simplify computation of the dose adjustment and for ease of visualization in the unadjusted condition. Figure 4 further illustrates the impact of temporal correction by showing dose accumulation after each treatment. These SBRT sessions took place over the course of two years. Under the assumption that Mr. X’s lung tissue experienced recovery, let us now consider how we might plan for dose escalation in future SBRT sessions. Using the previously described primate SRS dose toxicity data from Jones et al.6, we apply a model of temporal dose adjustment. The primate CNS model was chosen from the aforementioned theoretical models as a balance between optimistic and conservative tissue recovery rates.
In this model, we first calculate what percentage of the maximum tolerable dose can be given after a recovery period using the equation:
where Pi is the initial dose, D, as a percentage of the maximum tolerable dose . r is an experimentally derived constant that varies based on the time t since the initial dose: Because 100 − Padd will equal the temporally adjusted Pi, we can use the following equation to determine the numerical dose: Thus, Dcorr is the temporally corrected adjusted dose at each voxel which we use when evaluating our plan against established dose limits. For the purposes of this case, we use 72 Gy as the maximum cumulative tolerable dose of SBRT based on maximum tolerated estimates from dose escalation trials in non small cell lung cancer (NSCLC)23,24.
Figures 3 and 4 demonstrate how having a quantitative method to adjust dose accumulation through time could help guide clinical judgement. With the increasing arsenal of chemo-, targeted25, and immunotherapies26, advances in radiation therapy, and development of new surgical techniques, physicians will be better equipped to provide individuals with personalized therapy regimens. Such complexity requires good bookkeeping in order to ensure accurate communication between care providers, especially if they extend across different institutions. The proposed storage of radiation treatment associated data in the aforementioned DICOM object will allow for such synchronization. This approach leverages the existing standards published by DICOM to record chemotherapy history while allowing for future capabilities, such as indicating sites of surgical resections. Rather than relying on searches through archives of manually written patient notes from multiple sources, the data object will consolidate pertinent information into a standardized format to allow for understanding of a patient’s oncologic treatment history from a quick visual appraisal.
4 Areas for Further Work
We believe that the proposed standardized DICOM object and accompanying mathematical models will serve as a starting point to facilitate research on dose forgiveness, both intra- and inter-institutionally. However, further research is needed to address limitations in the characterization of organ functional capacity and dose mapping.
While the distinction between parallel and serial organs can be made in the software, the algorithm can be improved so that the effect of re-irradiation can be more accurately measured. As mentioned above, dose accumulation calculations to approximate dose toxicities from previous radiation are more useful for serial organs than parallel organs. For serial organs, prior dose maps will be processed with forgiveness algorithms. In the case of parallel organs, the software will calculate the prior volume exposed to radiation through dose map analysis. There are no studies that support dose adjustment based on the previously radiated volume, but prior radiated volume will be shown in the software to clinicians to help gauge the potential reserve capacity of the organ. But understanding a patient’s functional reserve can be challenging and often rely not just on objective measures (like PFTs) but also subjective measures like symptoms on exertion. This information can be found in the EMR but the location changes from patient to patient. Additionally, functional reserve metrics change from organ to organ; measures to gauge functional lung will be different from measures for functional liver or kidney. In the future, conceivably even this functional information will be included in the DICOM so that subsequent radiation courses can take into account patient functional status for parallel organs.
There are also several technical challenges that must be solved to enable the new standard. The most important is determining how to reliably map RT doses onto the reference model such that the voxels at each distinct time point correspond to the same underlying tissue, a process known as dose warping. This is currently accomplished through the use of deformable image registration (DIR), a mathematical operation that creates a transformation which maximizes similarity between two images. In general, we choose one image as the reference point and map other images onto it.
While we have made excellent progress in DIR, there remain significant limitations to DIR techniques: transformations are most accurate in regions with high contrast and unique landmarks. These features serve as reference points to ensure that the transformation is accurately warping the entirety of the source image. Low contrast or homogeneous regions like the lung and prostate can result in many solutions that yield the same similarity score with no clear ground-truth. Furthermore, uncertainty in dose warping sharply increases in situations where this is not a one to one mapping between voxels in the reference and source image27,28. Recent advancements in using artificial intelligence algorithms to perform image registration showed some promising results but a robust method has not yet been developed. Organs routinely appear to shrink and expand between images, likely due to daily variation, weight loss/gain, or tumor response to radiation. Major anatomical changes, such as a radical resection, provide even more of a challenge. Even within one RT session, we cannot assume that anatomy is static. For example, there exist several techniques to compensate for the effects of respiratory motion on lung tumor volume, including free-breathing, tracking, and gating?.
Another important challenge to solve is ensuring that there is interoperability between radiation planning software packages, including those from MIM Software, RaySearch Laboratories, and Varian Medical Systems. Clinicians at a hospital that uses MIM Maestro© should be able to open a new patient’s medical record and seamlessly import a reference DICOM file from the previous treatment team generated with RayStation©. Verifying interoperability may come in the form of a test suite, created by a consortium of RT software companies, which is run before the release of every new product version. The details of the test suite are beyond the scope of this proposal but would draw inspiration from similar efforts such as the IHE-RO initiative29. Alternatively, information about data formatting could be made part of the standard, thereby ensuring that data adheres to a format in a manufacturer-agnostic way.
Finally, we anticipate that each patient will have unique tissue recovery parameters. In order to further personalize radiation courses, work should be conducted on studying how patient genomics can be used to adjust the models detailed in prior sections, much like how genomic-adjusted radiation dose (GARD) has been shown to predict tumor response to radiotherapy30,31, and allow for formal optimization combining normal tissue and tumor control models.32
While these challenges represent important limitations in our proposed standard, we believe there are compelling immediate potential improvements in both clinical care and long term re-irradiation dose modeling from adoption of our proposed model.
5 Conclusion
Integrating treatment information into a single DICOM file offers substantial advantages that enhance both clinical practice and research. Initially, this approach will facilitate research on dose forgiveness by leveraging patient outcomes from repeat radiation courses. In the future, even basic side-effect profiles can be included so clinicians can better understand dose tolerance and tissue recovery between treatments. This analysis, while already possible with direct EMR access, becomes more versatile when incorporated into DICOM, enabling use across various software applications like treatment planning and contouring aids. Additionally, embedding this data within DICOM files promotes inter-institutional research and analysis. When patients receive treatments at different institutions, technological constraints often limit access to comprehensive treatment details. Including this information in DICOM files eliminates the need for separate record transfers, ensuring continuity and completeness of data. Moreover, recording treatment side-effects within DICOM files supports intra-institutional tracking of treatment tolerance and dose forgiveness, thereby fostering collaborative research efforts and enabling the pooling of patient data for more precise analyses.
Data Availability
Data will be provided upon request.
Footnotes
↵* scottj10{at}ccf.org