Positron Emission Tomography (PET)/Computed Tomography (CT) Imaging in Radiation Therapy Treatment Planning: A Review of PET Imaging Tracers and Methods to Incorporate PET/CT

Abstract

Purpose

Positron emission tomography (PET)/computed tomography (CT) has become a critical tool in clinical oncology with an expanding role in guiding radiation treatment planning. As its application and availability grows, it is increasingly important for practicing radiation oncologists to have a comprehensive understanding of how molecular imaging can be incorporated into radiation planning and recognize its potential limitations and pitfalls. The purpose of this article is to review the major approved positron-emitting radiopharmaceuticals clinically being used today along with the methods used for their integration into radiation therapy including methods of image registration, target delineation, and emerging PET-guided protocols such as biologically-guided radiation therapy and PET-adaptive therapy.

Methods and Materials

A review approach was utilized using collective information from a broad review of the existing scientific literature sourced from PubMed search with relevant keywords and input from a multidisciplinary team of experts in medical physics, radiation treatment planning, nuclear medicine, and radiation therapy.

Results

A number of radiotracers imaging various targets and metabolic pathways of cancer are now commercially available. PET/CT data can be incorporated into radiation treatment planning through cognitive fusion, rigid registration, deformable registration, or PET/CT simulation techniques. PET imaging provides a number of benefits to radiation planning including improved identification and delineation of the radiation targets from normal tissue, potential automation of target delineation, reduction of intra- and inter-observer variability, and identification of tumor subvolumes at high risk for treatment failure which may benefit from dose intensification or adaptive protocols. However, PET/CT imaging has a number of technical and biologic limitations that must be understood when guiding radiation treatment.

Conclusion

For PET guided radiation planning to be successful, collaboration between radiation oncologists, nuclear medicine physicians, and medical physics is essential, as well as the development and adherence to strict PET-radiation planning protocols. When performed properly, PET-based radiation planning can reduce treatment volumes, reduce treatment variability, improve patient and target selection, and potentially enhance the therapeutic ratio accessing precision medicine in radiation therapy.

Introduction

Since its development in the late 20th century, positron emission tomography (PET)/computed tomography (CT) has become a nearly indispensable tool in clinical oncology, demonstrating a number of clinical applications from improved tumor staging, treatment planning, response evaluation, and surveillance. While radiolabeled glucose (¹⁸F-fluorodeoxyglucose [FDG]) remains the cornerstone of PET/CT imaging, PET is a rapidly evolving technology with a growing number of radiolabeled probes that are extending its applications and indications. Novel radiopharmaceuticals targeting lipid metabolism, protein synthesis, hypoxia, and a host of other molecular pathways have been developed. Additionally, other probes have been engineered to target specific ligands, antigens, and receptors, such as the recently approved probes targeting prostate-specific membrane antigen (PSMA). These probes are valuable enhancements to precision medicine and are gaining rapid clinical acceptance.

The ability to accurately coregister and integrate PET/CT imaging into radiation therapy planning has increasing utility as molecular imaging evolves and increases in availability. Molecular imaging can be defined as the in vivo visualization, characterization, and quantification of biological processes at the cellular and molecular level by means of remote imaging detectors.¹ Compared with traditional anatomic imaging, molecular imaging using PET/CT often provides superior sensitivity, specificity, and accuracy that can allow for imaging of lesions not apparent on CT or magnetic resonance imaging (MRI) alone, as well as prevent futile radiation of anatomic abnormalities that do not actually contain tumor (eg, FDG PET/CT can reduce futile irradiation of atelectasis and edema).^2,3 This applies to both the delineation of the primary tumor as well as identification of locoregionally involved lymph nodes. Further, the functional information offered through molecular imaging may guide uses of dose escalation to potential tumor subvolumes with increased radioresistance or increased tumor burden. This also creates the possibility of response-adapted therapy in which changes to target volumes or dose are made during a treatment course. Perhaps, the greatest benefit gained through PET/CT is the reduction in interobserver variability in target volume delineation and decreased radiation therapy planning time through semi- or fully automated segmentation.^4,5

To reap these potential benefits, practicing radiation oncologists must understand the PET/CT technologies available to them along with the techniques on how to properly coregister and integrate the structural and functional information they contain. Here, we review the major United States Food and Drug Administration (FDA)–approved positron-emitting radiopharmaceuticals clinically being used today along with the methods used for their integration into radiation therapy planning.

FDA-Approved PET Radiopharmaceuticals

Table 1 describes the major molecular PET agents available in clinical oncology.

Table 1.

Federal Drug Administration–approved molecular agents used in clinical oncology

Abbreviation	Tracer full name	Cellular target	Molecular basis	Clinical application(s)
18F-FDG	Fluorine-18 fluorodeoxyglucose	Glucose metabolism	Increased rates of glycolysis overexpression of GLUT-1 and 3 receptors and increased levels of mitochondrial hexokinase in malignant cells	Tumor detection and staging Target volume delineation of multiple malignancies Monitoring of treatment response
18F-NaF	Sodium fluorine-18 fluoride	Bone metabolism	Increased bone turnover in lytic and blastic bone lesions	Staging, follow-up of prostate cancer Bone metastases
18F-FACBC or 18F-fluciclovine	Fluorine-18 fluciclovine	Amino acid transport	Increased rates of amino acid transport	Biochemically recurrent prostate cancer
11C-CHO	Carbon-11 choline	Lipid metabolism	Neoplastic cells exhibit increased levels of phosphorylcholine	Staging and follow-up of prostate cancers
68Ga-DOTA -TOC -TATE	Gallium-68 DOTA-peptide	Somatostatin receptor	Somatostatin receptors are overexpressed in many tumors	Staging, follow-up, assessment for possible radioisotope therapy for neuroendocrine tumors and meningiomas
64Cu-DOTATE	Copper-64 DOTATATE	Somatostatin receptor	Somatostatin receptors are overexpressed in many tumors	Staging, follow-up, assessment for possible radioisotope therapy for neuroendocrine tumors
18F-FES	Fluorine-18 fluoroestradiol	Estrogen receptor	Estrogen receptors are often expressed in breast cancer	Detection of estrogen receptor–positive lesions as an adjunct to biopsy in patients with recurrent or metastatic breast cancer
68Ga-PSMA-11	Gallium-68 ligand for the prostate-specific membrane antigen	Type II membrane protein	Prostate-specific membrane antigen inhibitor	Prostate cancers staging, follow-up, and 177Lu planning
18F-DCFPyL	Fluorine-18 ligand for the prostate-specific membrane antigen	Type II membrane protein	Enzymatic activity	Prostate cancer staging, follow-up, and biochemical recurrence evaluation

FDG

Over 4000 compounds coupled to positron-emitting radionuclides have been reported in the literature, with only a handful currently being used in clinical practice.⁶ The most common PET radiopharmaceutical used in clinical oncology is the glucose analog FDG. This tracer is routinely used in a number of common cancers including cervical, colorectal, esophageal, lung, lymphoma, and squamous cell cancer of the head and neck. Its role in oncology stems from the observation that most cancer cells produce energy predominantly through the less efficient aerobic glycolysis pathway, rather than the highly efficient citric acid cycle used by most normal cells, and thus demonstrate high levels of glucose uptake (a phenomenon known as the Warburg effect).⁷ Similar to glucose, FDG is transported into cancer cells via glucose transporters^8,9 and then is phosphorylated by overly activated hexokinases⁹ in cancer cells. Unlike normal glucose, phosphorylated-FDG cannot undergo further enzymatic breakdown and instead remains metabolically trapped in the cell.¹⁰ The net result is an increased accumulation of FDG within tumor cells compared with most normal tissues. FDG is inherently a nonspecific tracer and nonmalignant processes have FDG uptake, including inflammation, colonic and gynecologic activity, thymic hyperplasia, brown fat, and high glucose using organs such as the brain and liver.11, 12, 13, 14 Further, not all tumor demonstrates the Warburg effect, and are in fact poorly imaged by FDG. Examples of cancers poorly assessed by FDG PET include prostate cancer, hepatocellular carcinoma, renal cell carcinoma, brain tumors, low-grade lymphomas, low-grade sarcomas, and well-differentiated neuroendocrine tumors.15, 16, 17, 18, 19, 20 FDG is excreted through the kidneys and accumulates in the bladder making evaluation of bladder tumors likewise difficult to assess.

Over the past several decades, our understanding of the various hallmarks and metabolic pathways of cancer has increased allowing for the development of numerous new targeted PET radiotracers.21, 22, 23 These new tracers, which are discussed in the following, serve as alternatives to FDG PET, expanding our diagnostic and targeting capabilities.

Tracers Targeting Bone Metabolism

Sodium ¹⁸F-sodium fluoride (¹⁸F-NaF) is an FDA-approved PET tracer clinically available to target bone metabolism. The uptake of ¹⁸F-NaF is related to blood flow and osteoblastic activity and is therefore nonspecific with increased uptake in osseous sites undergoing rapid remodeling, such as the growth plate in pediatrics, sites of arthritis, healing of broken bone, or in bony neoplastic lesions. Clinically, it demonstrates high sensitivity for bone metastases and can detect neoplastic foci before changes are appreciable on conventional imaing.²⁴ Its diagnostic accuracy is greatest for sclerotic lesions over lytic lesions, and, thus, its primary clinical application is for the diagnosis of bone metastases from prostate cancer, lung cancer, and mixed bony lesions of breast cancer.²⁵ Compared with ^99mTc-bone scan and FDG PET in these patient populations, ¹⁸F-NaF shows improved diagnostic accuracy with nearly 100% sensitivity for bone metastases.²⁶ When further compared with traditional planar bone scintigraphy, ¹⁸F-NaF with its CT component offers superior image quality, anatomic localization, and additional morphologic information that improves specificity.^24,27 However, despite these mild advantages, bone scintigraphy is still far more frequently used today given its cheaper cost, wider availability, and still relatively acceptable sensitivity, specificity, and positive and negative predictive values.

The primary limitations of ¹⁸F-NaF in evaluating metastatic disease are its mechanism of tracer uptake, which is not specific for tumor but for osteoblastic response, and its inadequacy to depict other findings beyond osseous lesions. Coverage and reimbursement must meet specific clinical needs to change management, and as a result, this tracer is infrequently used currently.

Tracers Targeting Amino Acid Transport and Protein Synthesis

Another hallmark of cancer is that amino acid transport and protein synthesis are often dysregulated in a manner that promotes increased replication and survival.^22,23 Molecular imaging can target this altered cellular metabolism using radiolabeled amino acids. These amino acid analogs can be metabolized and/or transported across cellular membranes, similar to endogenous amino acids, and can be helpful in targeting a number of cancer cell types. There are currently 2 FDA-approved PET tracers targeting amino acid transport.

Fluorine-18 fluciclovine (¹⁸F-FACBC or ¹⁸F-fluciclovine) is an amino acid analog that leverages upregulated amino acid transport in certain malignancies. Clinically, it has demonstrated benefit primarily in the staging and follow-up of prostate cancer. Compared with prior historical controls in prostate cancer, ¹⁸F-fluciclovine shows higher sensitivity, specificity, accuracy, positive predictive value and negative predictive value in the detection of prostate, prostatectomy bed, and extraprostatic disease.^28,29 In the EMPIRE-1 study, incorporation of ¹⁸F-fluciclovine into postprostatectomy radiation therapy decision making and planning was shown to improve biochemical recurrence-free survival compared with conventional imaging.³⁰ Since its FDA approval, progress has been made in other PET tracer used in prostate cancer, particularly PSMA tracers. Compared with PSMA tracers, ¹⁸F-fluciclovine is less likely to yield positive results in patients with prostate-specific antigen (PSA) levels <1 ng/mL. While PSMA tracers are likely to largely replace ¹⁸F-fluciclovine in prostate cancer, given PSMA's superior sensitivity and specificity, there may still be a role for this tracer. One unique feature of ¹⁸F-fluciclovine is that there is often minimal activity in the excreted urine at the time of scanning, unlike PSMA. This makes this tracer ideal in biochemically recurrent prostate cancer when there are equivocal findings in the prostate bed on conventional imaging or PSMA. Further it can be helpful in the ∼10% of prostate cancers which are PSMA negative. Given these advantages we would recommend consideration of ¹⁸F-fluciclovine in patients with rising PSA and concern, for prostate bed recurrence when conventional or PSMA scans are equivocal or unrevealing.

Another clinically useful synthetic amino acid tracer is fluorine-18 fluorodihydroxyphenylalanine (¹⁸F-DOPA). L-DOPA is the precursor of the neurotransmitters dopamine, norepinephrine, and epinephrine. Once transported across cellular membranes, it is metabolized, stored, and trapped in secretory vesicles of sympathetic cells. Simultaneous administration of carbidopa increases its plasma concentration by slowing its degradation and excretion leading to enhanced image quality. While its FDA approval is for the evaluation of Parkinson's disease, it has potential oncologic uses. As a target of sympathetic cells, it allows for evaluation of neuroendocrine tumors (NETs), especially well-differentiated NETs, like medullary thyroid cancer, pheochromocytoma, and paragangliomas.³¹ It is also useful for clinical brain tumor imaging with a 96% sensitivity for detection of primary and recurrent brain tumors and can help differentiate low and high-grade tumors.32, 33, 34, 35, 36

Tracers targeting lipid metabolism

In order for dividing cells to survive, cell membranes must duplicate at the same rate as cell duplication. Thus, cancer cells use greater amounts of substrates for synthesis of cell membranes than normal cells. Choline is one of the primary precursors to the phospholipids that make up cells membranes and has been used in PET imaging.^{37 11}C-choline (¹¹C-CHO) was introduced in 1998 and has been used to target a variety of tumors such as brain tumors, lung cancer, gastrointestinal and genitourinary cancers, and especially prostate cancer.38, 39, 40 As with all ¹¹C-labeled tracers, its short half-life of 20 minutes limits its use to centers with an on-site cyclotron and presents numerous logistical challenges. For this reason, longer-lasting ¹⁸F analogs have been developed including ¹⁸F-fluoroehtylcholine (¹⁸F-FECH) and ¹⁸F-fluoromethylcholine (¹⁸F-FCH)39, 40, 41, 42; however, these are not yet FDA approved. Although these show improved detection of prostate cancer, they are largely replaced by the new PSMA tracers.

Tracers targeting specific ligands and receptors

Numerous tracers have been developed to target specific ligands and receptors found on cancer cells. These include somatostatin receptor (SSTR)–based agents, hormone receptor-based agents, and PSMA-based agents.

SSTRs are expressed throughout the body and are overexpressed in many malignant tumors. Five different subtypes of SSTRs have been characterized and can be targeted by a group of galium-68 (⁶⁸Ga)– or copper-64 (⁶⁴Cu)–labeled somatostatin analogs. All of these are composed of the chelator 1,4,7,10-tetraazacyclododecane-N,N′,N′′,N′′′-tetraacetic acid (DOTA) and a ligand of a particular SSTR. The 2 FDA-approved ⁶⁸Ga-DOTA-peptides are tyrosine-3-octreotide (⁶⁸Ga-DOTA-TOC) and tyrosine-3-octreotate (⁶⁸Ga-DOTATATE). These tracers are FDA approved for localization of SSTR-positive NETs in adults and pediatric patients. However, they can also be used for imaging breast cancer, medulloblastoma, meningiomas, neuroblastoma, pheochromocytoma, paragangliomas, and primitive neuroectodermal tumors.43, 44, 45, 46, 47

Estrogen receptor– and androgen receptor–based agents have also been developed.⁴⁸ Fluorine-18 fluoroestradiol (¹⁸F-FES) allows in vivo evaluation of estrogen receptor status in breast and gynecologic cancers and is currently FDA approved for the detection of estrogen receptor–positive lesions as an adjunct to biopsy in patients with recurrent or metastatic breast cancer. Like all PET studies, ¹⁸F-FES PET enables evaluation of the entire metastatic tumor burden, unlike tissue sampling. ¹⁸F-FES has been used to predict response to hormonal therapy in breast cancer and to identify responders and nonresponders of hormone therapy. Combination of FDG PET and ¹⁸F-FES is currently used to help guide the timing of hormone therapy and/or the selection of another target therapy or chemotherapy.49, 50, 51 This tracer has potential benefit for identification and targeting of oligoprogressive lesions that are unlikely to benefit from hormone therapy but may benefit from stereotactic body radiation therapy; however, there are no data to support this use currently.

There is a lot of interest in targeting PSMA in oncology. PSMA is a transmembrane protein that is selectively overexpressed in 90% to 100% of prostate cancer lesions. Its overexpression is correlated with increased tumor grade, stage, biochemical recurrence, and androgen-independence, making it a useful target in the evaluation and treatment planning of prostate cancer. There are now 2 FDA-approved PSMA tracers: 68-Ga-gozetotide (⁶⁸Ga-PSMA-11) and ¹⁸F-piflufolastat (¹⁸F-DCFPyL).⁵² These are functionally similar; however, the ¹⁸F ligand has a number of potential advantages including, improved spatial resolution (due to shorter positron range in tissues of ¹⁸F compared with ⁶⁸Ga), more accurate quantitation, higher tumor-to-background ratios leading to better detection of lower-grade or smaller size prostate cancers and better stability allowing for higher production capacity and use at more centers.

These PSMA tracers are FDA approved for use in patients with suspected metastases who are candidates for initial definitive therapy as well as patients with suspected recurrence based on elevated PSA level. The former indication is unique to these tracers compared with fluciclovine, which is approved only for detection of disease with biochemical recurrence. ⁶⁸Ga-PSMA is also approved for selection of patients with metastatic prostate cancer for whom lutetium (¹⁷⁷Lu) PSMA–directed therapy is indicated. PSMA tracers have been shown to be more accurate than conventional imaging with CT and bone scanning in patients with high-risk prostate cancer, and there is mounting evidence that PSMA is likewise superior to ¹¹C-CHO and ¹⁸F-fluciclovine.^53,54 There is growing evidence of its extremely high sensitivity in the setting of biochemical recurrence, even in patients with low PSA levels (<1 ng/mL).⁵⁵ It is effective for imaging lesions in the prostate itself, lymph nodes, soft tissues, and bone, making it the most versatile of the prostate specific tracers. Although, as noted previously, ¹⁸F-fluciclovine may better perform in the prostate bed. With its high sensitivity, PSMA imaging is said to be redefining our believed patterns of spread for prostate cancer.⁵⁶

Users must understand that PSMA is not specific to prostate cancer and is physiologically expressed in the salivary glands, proximal renal tubules, epididymis, liver, spleen, small bowel, osteoblastic activity and astrocytes and also can be expressed in bladder, pancreas, lung, and kidney cancers (Fig. 1).⁵⁶ These tracers also demonstrate increased uptake in ganglia, which may present diagnostic challenges.⁵⁷ It is also renally excreted and accumulates in the bladder which can impede assessment/targeting of nodes near the ureter or the prostate bed. Knowledge of its physiological distribution and limitation is essential to safe and efficacious radiation therapy planning in prostate cancer. We recommend collaboration with nuclear medicine physicians when planning from PSMA scans.

Figure 1.

Anterior maximum-intensity projection of a ⁶⁸Ga-PSMA-11 PET/CT scan in patient with history of prostate cancer, treated initially with radical prostatectomy, now with biochemical recurrence with scan showing disease in a right internal iliac pelvic lymph node (yellow arrow). Note the normal physiological distribution of ⁶⁸Ga-PSMA-11 in the salivary glands, liver, spleen, kidney, small bowel, and retention in the bladder (blue arrows). Understanding of normal tracer distribution is essential in PET-guided radiation therapy planning. This image was originally published in Calais et al¹⁴¹ and was modified for this publication. Abbreviations: CT = computed tomography; ⁶⁸Ga-PSMA-11 = gallium-68 ligand for the prostate-specific membrane antigen; PET = positron emission tomography.

Additional agents being explored

A number of PET tracers are still undergoing active investigation. For example, there are tracers being developed to target DNA synthesis pathways,58, 59, 60, 61, 62, 63, 64 angiogenesis,65, 66, 67, 68 apoptosis,⁶⁹ hypoxia, and even immuno-PET tracers with monoclonal antibodies targeting immune checkpoints or other immune system pathways.^70,71 Many of these tracers could guide dose painting and response adaptive treatment protocols in the future. The ability to target hypoxic regions of a tumor is especially appealing to an oncologist as hypoxia is known to reduce chemotherapeutic and radiotherapeutic efficacy of treatment. A number of such agents have been developed including fluorine-18 fluoromisonidazole (¹⁸F-FMISO), fluorine-18 fluoroazomycinarabinoside (¹⁸F-FAZA), and copper-64 diacetylmethylthiosemicarbazone (⁶⁴Cu-ATSM).⁷² Hypoxia imaging with these tracers could be used to optimize radiation therapy planning by identifying resistant regions that may benefit from altered dose/fractionation.⁷³ These agents have shown promise in a number of solid tumors, particularly head and neck and lung carcinomas.73, 74, 75, 76, 77 Another noteworthy agent is gallium-68 fibroblast activation protein inhibitor (⁶⁸Ga-FAPI), which shows promise as pan-cancer agent and is being explored as an alternative to FDG PET.78, 79, 80, 81, 82 We do not recommend routine use of these tracers for radiation therapy treatment planning outside of centers with specific expertise until further data are acquired. Additionally, some groups are exploring the benefit of multitracer imaging techniques to guide radiotherapy⁸³; however, we likewise do not recommend multitracer use until further trials are performed.

Integration of Molecular Imaging to Radiation Treatment Planning

Registration

Radiation therapy planning is highly dependent on imaging to accurately delineate the treatment target(s) and to confidently spare normal tissues from unnecessary radiation. Usually this was done using basic anatomic imaging like CT and MRI, but now the registration and fusion of PET/CT offers valuable functional information to further guide treatment planning. Registration refers to the spatial alignment of 2 similar images to each other and fusion refers to the combined display of the spatially aligned data. The registration and fusion can be achieved through a number of techniques, which we will discuss here.

Table 2 is an overview of registration methods.

Table 2.

Comparison of registration methods

Registration method	Pros	Cons
Visual/cognitive fusion	• Free • No added software needed • Images do not need to be in DICOM format	• Substantial interobserver variability • Decreased precision
Rigid registration	• Fast • Software is inexpensive and widely available • Automated and manual options • Software is generally easier to use	• Difficult when scans are acquired in different positions • Errors if changes in patient weight or organ filling • Difficult to account for motion • Difficult over large volumes of interest • Difficult when there are multiple foci of interest for registration
Deformable registration	• Increased flexibility • Can compensate for some differences in treatment position, changes in patient weight, and organ filling • Can handle multiple foci of interest • Less volume dependent • Can account for most motion	• Complex, requires expertise, and manual review of accuracy • Purchase of software often required • Less widely available than rigid registration software • Struggles with absence of structures between scans • Potentially large error in areas of low image contrast with only soft tissue • Studies show marginal value relative to rigid methods unless there are significant anatomic differences
PET/CT simulation	• Most precise and reproducible • Limits issues with position, weight changes, and motion as scans are acquired at same time • Less time consuming as registration is automated • Can limit number of hospital trips for patients	• Costly to purchase, staff, and maintain • For some scanners bore size can be limiting for large patients or immobilization devices • Clinical benefit over standard methods not necessarily demonstrated

Abbreviations: CT = computed tomography; DICOM = Digital Imaging and Communications in Medicine; PET = positron emission tomography.

Cognitive fusion

Perhaps the most basic way to incorporate PET information into treatment planning is by using a “visual or cognitive fusion” of the 2 images. In this technique, a diagnostic PET scan is displayed next to the radiation therapy planning CT, and the provider simply compares the 2 images to inform their contouring of volumes and makes cognitive adjustments to account for any anatomic or spatial differences between scans. No actual registration is performed, and the PET images do not need to be in DICOM (Digital Imaging and Communications in Medicine) format. This lends to intra- and interobserver variability and decreased precision.⁸⁴ However, it is low cost and is often necessary when scans are drastically different from one another and the more formal registration methods are not practical.

PET/CT simulation

At the opposite end of the spectrum from cognitive fusion is a single-scan PET/CT simulator approach in which the radiation therapy planning CT is acquired together with the PET scan with the patient in the treatment position on a flat couch, and the fusion process is executed automatically based on the shared coordinate system between the CT and PET images. This can limit the number of visits for patients while also decreasing the amount of variation between the PET and the radiation therapy planning CT. The flat couch used on PET/CT simulators is a major advantage to this technique as radiation therapy is traditionally planned and delivered on a flat couch, whereas most diagnostic PET/CT scanners use a nonflat couch and thus introduce inherent limitations when trying to register diagnostic PET/CT scans to radiation therapy planning CTs via rigid and even deformable methods. Additionally, on a dedicated PET/CT simulator there is opportunity for implementation of motion management techniques such as respiratory gating to be implemented during PET acquisition, allowing for improved targeting and easier registration in areas of motion. Unfortunately, the cost of purchasing, staffing, and maintaining such a device can be prohibitive for many practices. Lastly, use of a dedicated PET/CT simulator does not always lead to a clear clinical difference.⁸⁵

Rigid registration

For cases where the PET scan and radiation therapy planning CT scan are acquired on different scanners, rigid or deformable image registration techniques are often used. Both techniques require dedicated software though rigid registration software is more widely available, easier to use, but also more limited in ability. Rigid registration involves transformations that maintain the distance between all points of an image.⁸⁶ This includes translations as well as rotations in all directions. Most RT planning software have automatic and manual rigid registration modes and this technique works well for most fixed structures like the skull. The largest limitation of rigid registration stems from the fact that distances between corresponding anatomic points are fixed and, therefore, any differences in anatomic position between scans are inherently maintained in rigid fusion. This means some degree of mismatch will occur if scans are done in different positions, if there are changes in patient weight, or if there is soft tissue movement such as breathing, peristalsis, bladder filling, heartbeat, or other involuntary motion. This further translates to greater mismatch and complexity as the number and distance between regions of interest increases. For example, when using a rigid technique, it is easier to register to a single PET-avid node than it is to register to multiple avid nodes spaced distantly apart. In such, cases the provider must make tradeoffs to optimize the single rigid registration to reduce mismatch in the area of greatest clinical significance and account for the added uncertainty in their radiation therapy planning. Another option often used by the authors, is to make multiple rigid registrations, each aligning to a different area of clinical focus, for example making one registration at the upper neck to define volumes in that anatomic region and then another at the lower neck for similar purposes. This takes longer, can only be used for target delineation, cannot be used for cumulative dose calculations, and can introduce other possibilities for error that should be accounted for in planning.

Deformable registration

Sophisticated deformable registration methods were developed to try to overcome the limitations of rigid registrations. Deformable image registration allows for the independent displacement of every voxel in the source data set to match the target data set. This translates to nonlinear spatial variations where the number of degrees of freedom can be as large as 3 times the number of voxels.⁸⁶ These capabilities reduce overall geometric differences between 2 images sets by estimating spatial relationships between the volume and/or intensity elements of corresponding structures.86, 87, 88 For radiation treatment planning, deformable registration is first performed between the CT of the diagnostic PET/CT (source) to the planning CT (target). The PET image is then codeformed to the planning CT with the same deformation map. There are automatic and manual deformable registration modes now commercially available for nearly all treatment planning systems; however, these often require specific license purchases to use. Given the inherent complexity of these registrations, the authors recommend these be done by those with expertise and under the oversight of a dedicated image registration quality assurance committee.⁸⁶ Even with expertise and the sophisticated deformable registration algorithms available, limitations still exist. Many deformable models still struggle when registering 2 images with one structure present in one and not present in the other such as scans with and without a vaginal applicator, mouth block, hardware, or surgically resected organ. Further, some algorithms will struggle in areas with very low tissue contrast. Lastly, while deformable techniques have demonstrated potential for improving clinical outcomes,⁸⁹ there is marginal value with the current commercially available software unless there are significant anatomic differences between PET/CT and planning CT images.^90,91 Overall, a dedicated PET/CT simulator may be most ideal for treatment planning but is not necessarily economically justifiable for many institutions. In absence of such technology, rigid registration techniques may be most practical and likely provide adequate precision and accuracy for majority of cases. In cases where there are significant anatomic differences between the PET/CT and the planning CT, we recommended application of deformable registration techniques with the aid of those with technical expertise in image registration.

Target volume delineation

The primary goal of curative radiation therapy is to deliver tumoricidal doses of radiation to entire tumor volumes and any site of microscopic disease extension while minimizing dose to normal tissues. Modern radiation therapy techniques such as intensity modulated radiation therapy, charged particle radiation therapy (ie, proton therapy), and image-guided radiation therapy allow for unprecedented targeting capabilities, which permit dose escalation with relative sparing of the normal tissues. To optimally exploit these gains in conformality, we must be able to identify the precise location, extent, and character of the tumors being treated. Historically, we relied on anatomic data from CT and MRI scans, but now with rising availability of PET/CT and its growing arsenal of tracers, more complex functional and biological data can be assimilated into our treatment planning to improve our target volume segmentations.92, 93, 94, 95, 96 There is no current consensus agreement in how PET data should be used to create target volumes for radiation therapy planning. We will review some of the most common techniques and concepts.

Visual assessment

Manual segmentation (ie, contouring) using visual assessment is the most common method of incorporating PET/CT data in target volume delineation. This method relies on user experience and an understanding of each radiotracer's unique limitations and mechanisms of uptake. While using PET-imaging reduces interobserver variability in target volume delineation compared with using standard anatomic imaging alone,97, 98, 99 manual segmentation using visual assessment is still very operator dependent and has its own interobserver variability.^100,101 This is because the margins of PET-detected lesions are often unclear and can be influenced by many factors like the windowing and color scale of the display, the contrast between the lesion and background uptake, motion blur, and artifacts such as signal spill over due to the limited spatial resolution of PET (Fig. 2).² For this reason, we recommend that a detailed PET segmentation protocol by disease type and tracer be followed when using visual assessment techniques. Each institution's protocol should also include appropriate windowing and display settings, should specify that all scans be performed on the same scanner and in the same mode of imaging, and should also incorporate input from a nuclear medicine physician. Adherence to well thought out protocols will promote reproducibility in target delineation.

Figure 2.

Example of PET-guided target delineation differences between a visual assessment–based contour versus threshold-based autosegmentation technique for patient with prostate cancer with bilateral ilia lesions on fluciclovine scan. A rigid image fusion was performed between the simulation CT and the fluciclovine PET/CT. This is an axial slice of the pelvis showing the fused images. The physician-determined gross target volume is seen in red (planning target volume is cyan) and was based on visual assessment of avidity. The autosegmented gross target volume is seen in yellow (planning target volume is dark blue) and was generated using a percentage-based thresholding technique with contour edge determined by 50% intensity of the maximum standardized uptake value. PET-guided volumes in general are more reproducible between physicians; however, use of technique can lead to meaningful difference in treatment volume. Abbreviations: CT = computed tomography; PET = positron emission tomography.

Automated techniques

Automated and semiautomated methods have been developed to reduce variability and save time in PET-based target volume delineations. The most common of these methods uses thresholds (ie, cutoff values) of the standardized uptake value (SUV) to guide contour edges. For example, an absolute SUV threshold of 2.5 for FDG in non-small cell lung cancer (NSCLC) has been used where all sites with >2.5 SUV are included in the contour.^100,102 Thresholds can also be percentage based, where the edge of the target is set using a fixed percent intensity level relative to the peak or maximum SUV of specified volume of interest (40%-50% of SUV_max for FDG and 60%-70% for SUV_max ¹¹C-CHO are most commonly used) (Fig. 2). Fixed threshold methods tend to underestimate small tumors, and percentage-based models struggle with inhomogeneous tracer uptake and background noise.¹⁰⁰ Threshold methods may also fail when the target is adjacent to an organ such as bladder that has high tracer accumulation. For this reason, more advanced techniques have been developed such as adaptive/iterative, gradient-based, statistical, and machine learning methods. These methods can account for tumor size and background noise and provide subpixel accuracy; however, they are far more complex and require scanner-specific calibration or time-consuming machine training. To date, these advanced techniques appear more robust than simple threshold techniques, with perhaps statistical and machine learning methods showing the greatest potential.103, 104, 105, 106, 107 Unfortunately, most of these algorithms are not yet implemented in commercially available software.

These automated techniques are helpful and reduce variability in PET-derived target volumes but have limitations which users must understand. An ideal PET automated segmentation method would be able to account for the all the physical and technical sources of bias and uncertainty while also accounting for anatomic, physiological, and other clinical information not present in a PET image.¹⁰⁸ Fixed thresholding techniques do not meet this idealistic standard and even the most advanced techniques (gradient based, statistical, and machine learning) are only able to account for the physical and technical components. Currently, no existing method of PET autosegmentation can account for the clinical information not contained in the PET image data itself. This means active physicians contribution remains essential in target volume segmentation and the authors recommend a semiautomatic (ie, mixed manual and automatic) approach if PET autosegmentation methods are going to be used. In this approach, there would be pre and post PET autosegmentation input from physician teams to first guide the software away from physiological uptake and artefact followed by review and editing of the final generated target volumes considering the full clinical context. We recommend this be done with input from a nuclear medicine specialist.

Biologic Target Volumes, Dose Painting, and Biology-Guided Radiation Therapy

In addition to gross tumor segmentation, PET/CT imaging allows for the identification of biological subvolumes of tumor with specific features on molecular imaging suggestive of radioresistance (eg, hypoxia, proliferation) that could be targeted with nonuniform doses of radiation. These PET-guided biological subvolumes have been termed biological target volumes (BTVs) and have been paired with the concept of dose painting in which higher doses of radiation are given to high-risk areas while lower doses are administered to low-risk areas based on functional data acquired from PET imaging.⁹⁶

There are 2 classical approaches to dose painting.¹⁰⁹ The first is dose painting by contours in which the BTV is treated to a specified dose level while keeping the mean dose to the remaining target constant. The second is dose painting by numbers where a dose prescription varies on a voxel-to-voxel basis based on quantitative PET data to produce a desire result, for example similar levels of expected cell kill within a biologically heterogeneous volume (coined kill painting). A number of studies have shown feasibility of these techniques with encouraging results such as reduced target volumes in head and neck cancer and NSCLC populations^103,109,110; however, randomized data are still needed to show if this leads to improved local control or reduced toxicity, and several such trials are now underway in NSCLC populations.¹¹¹

Many commercial treatment planning systems can accommodate dose painting techniques and incorporate specific tools for BTV segmentation and plan evaluation. Users should understand that while coverage can be evaluated by standard dose-volume histograms when using dose painting by contours, such methods do not work for evaluating dose painting by numbers as the dose prescription is heterogeneous. Instead, specific tools are used to evaluate plan conformity for dose painting by numbers such as, dose difference histograms, quality-volume histograms, and quality factor. Overall dose painting methods are limited by factors like voxel size, tracer type, and uncertainty in dose calculations with small treatment fields.¹¹² Further, consideration must be given to the conformality techniques and image guidance used when boosting such small volumes, especially if there are significant dose gradients, to ensure boosted areas receive the intended dose.

Along with BTVs and dose painting, there is an emerging radiation modality termed biology-guided radiation therapy (BgRT) or emission-guided radiation therapy. This treatment modality combines a PET-CT and a 6 MV linear accelerator into a single machine capable of real-time delivery of beamlets of radiation from the linear accelerator according to the outgoing number of PET emissions with only subsecond latency.113, 114, 115, 116, 117 These machines essentially transform PET-avid tumors into their own fiducials and allow for real-time tumor tracking and treatment with a high accuracy. For mobile tumors, this technology has the potential to reduce the need for large margins traditionally used to account for intrafraction motion as well as help overcome some of the limitations of motion experienced with more traditional registration-based PET-guided treatments. BgRT systems were just recently FDA approved and thus commercially available with initial clearance for delivery of BgRT using FDG PET for targeting primary lung and bone tumors, and lung and bone metastases arising from other primary cancers. We hope approval for BgRT to other disease sites and use with other PET/CT tracers such a PSMA will be forthcoming pending future trials.

Adaptive Radiation Therapy

Another way to incorporate PET data into radiation therapy planning involves the use of adaptive radiation therapy. With this technique, radiation plans are adjusted according to initial response to radiation therapy as gauged by a PET scan performed within the first several weeks of treatment. This concept arose from the fact that functional and molecular changes can be observed early during radiotherapy.¹¹⁸ Depending on PET response, treatments plans could be modified to increase dose to resistant regions, lowered to better spare organs at risk, or justify discontinuing treatment for nonresponders who could benefit from a different treatment. Early FDG PET studies show feasibility and potential benefit in esophageal, lung, and lymphoma patients.119, 120, 121, 122 Adaptive radiation therapy's biggest limitation is its clinical practicality as it requires the repetition of multiple complex and time-consuming tasks including imaging acquisition, registration, sophisticated segmentation, radiation therapy planning, cumulative dose assessment, and the ability to account for previously delivered dose.^123,124 We do not recommend implementing such techniques without strong technical, computational, and logistical support or outside the context of a clinical trial.

Additional Considerations

PET imaging can be used to improve radiation treatment planning in many ways, however there are some general limitations that should be understood before integrating it into practice.

A general limitation of PET imaging is its spatial resolution (∼4.5 mm in modern scanners). This resolution has implications for treatment planning, volume delineation, and treatment response assessment as it can lead to artifacts and possible over or under contouring of small lesions (<10 mm) due to signal spill over or partial volume effects. Another complicating factor is that SUV measurement is semiquantitative and can be influenced by a host of biological, physical, and technological factors.125, 126, 127 For example, scanner variability, differences in scan acquisition and reconstruction parameters, injection measurements, and calibration errors can all introduce significant SUV differences. Use of contrast can artificially elevate SUV readings and hardware artifacts from CT-based attenuation correction can likewise lead to overestimation of the SUV.¹²⁸ The biologic factors affecting SUV calculations are often tracer dependent and can be influenced by tracer uptake time, body size calculation, and other patient related factors. Depending on its application and the tracer used, SUV may not represent a reliable measure in all circumstances. Adherence to strict acquisition and image analysis protocols can help mitigate the errors introduced in SUV calculations to radiation therapy planning.

Motion, particularly respiratory motion, represents another challenge to PET/CT in radiation therapy planning. Respiratory motion can cause misalignment of PET and CT planning scans particularly for lung or upper gastrointestinal cancers with misalignment worse at the lung bases. This motion can also affect target volumes and SUV readings. PET imaging is typically performed during free respiration over many respiratory cycles given the time necessary to obtain adequate imaging statistics. The composite image will lead to an increased size of the tumor while also causing a decrease SUV measurement (up to 24% decrease) over its range of movement.^129,130 When using thresholding techniques or creating a BTV for dose painting, this decrease must be accounted for by using lower threshold values. If planning using a visual method, users must remember that the intensity of uptake will seem less intense at the extreme ends of the tumor movement. A number of techniques exist to limit motion or account for motion-induced artifacts during PET imaging such as respiratory gating during PET acquisition, 4-dimensional-gated PET acquisition, and deformable image registration. If used, it is most important that these techniques are then carried over when actually delivering the radiation. If a BTV is segmented on a respiratory-gated PET/CT but treated with free breathing, there could be target miss. Conversely if the BTV is segmented on an ungated PET/CT but treated with gating, there could be overtreatment of normal lung. In the future, some of these limitations may be overcome by novel total-body PET scanners with extended axial field of views and thus improved sensitivity, signal-to-noise ratio, and rapid scanning time capabilities, making single-breath hold PET imaging feasible; however, these are still only used in the research setting and not widely available.131, 132, 133 Given the aforementioned limitations, we recommend dedicated quality assurance programs be put in place for motion management using PET/CT in radiation therapy planning if PET-guided radiation therapy planning is performed.

Other things to consider when incorporating PET/CT imaging into your radiation therapy planning process are the high costs associated with PET tracers, the potential for treatment delays in areas with limited PET/CT access, and the effect that timing between diagnostic PET/CT and start of radiation therapy can have on outcomes. These delays have been shown to negatively affect treatment outcomes by allowing time for disease progression and increases in target volumes.134, 135, 136, 137, 138, 139, 140 If PET/CT-guided techniques are going to be used, PET/CT should be acquired as close as possible to time of radiation therapy (ideally <3 weeks) for optimal registration, target delineation, and planning.¹³⁴ The disease process and trajectory of progression should always be considered when deciding if new imaging is needed. After 4 to 8 weeks of delay between the diagnostic and radiation planning, complete restaging including PET/CT should be considered, given risk of progression and changes in target volumes, especially in populations at high risk of interim progression such as locally advanced NSCLC.^136,138 Additionally, when using highly conformal radiation techniques, such as intensity modulated radiation therapy, contemporaneous PET/CT with radiation therapy planning is particularly important as these techniques are less forgiving for target delineation errors and lead to increase chance of miss.

Conclusion

As the number of FDA-approved PET radiotracers expands and indications for these agents increase, the use of PET/CT in radiation oncology will undoubtedly grow. Practicing radiation oncologists should have an understanding of PET imaging, including its limitations and pitfalls. For such innovative protocols to be successful, collaboration between radiation oncologists, nuclear medicine physicians, and medical physics is essential, as well as the development and adherence to strict PET radiation therapy planning protocols. When performed properly, PET-based radiation therapy planning can reduce treatment volumes, enhance the therapeutic ratio, reduce treatment variability, improve patient and target selection, and open the door to precision medicine in radiation therapy.