Advances and Challenges in Precision imaging

Abstract

Technological innovations in genomics and related fields have facilitated large sequencing efforts, supported new biological discoveries in cancer, and spawned an era of liquid biopsy biomarkers. Despite these advances, precision oncology has practical limitations, partly related to cancer’s biological diversity and spatial/temporal complexity. Advanced imaging technologies are being developed to address some of the current limitations in early detection, treatment selection and planning, drug delivery, and therapeutic response, as well as difficulties posed by drug resistance, drug toxicity, disease monitoring, and metastatic evolution. We discuss key areas of advanced imaging on the horizon for improving cancer outcomes and survival. Finally, we discuss practical challenges to the broader adoption of precision imaging in the clinic and the need for a robust translational infrastructure.

Introduction

Precision or personalized medicine is an increasingly accepted approach to cancer care, in which therapy is planned based on the distinct molecular characteristics of a given tumor. Critical to precision medicine is the concept that the right combination of drugs must be used at the right stage and time in the progression of the disease. Although precision oncology has historically been defined by tailoring treatments to genetic mutations, its clinical translation has proven more complex. Contributing to the complexity are biological factors including the variable nature of the transcriptome, proteome, tumor microenvironment, physiology, lineage plasticity, and, importantly, dynamic temporal changes due to treatment pressures. Further complicating the choice of therapies is that new mutations evolve over time, and some deleterious mutations can also occur in normal tissues. Furthermore, the host also plays a key role in determining treatment response (i.e., different responses to the same dose).

In parallel to the emergence of precision oncology, there have been remarkable advances in imaging diagnostics, radiotheranostics, and image-guided therapy. Indeed, imaging has become indispensable in the entire treatment chain of cancer care, from screening, detection, and staging through treatment selection, planning, efficacy monitoring, image-guided treatment delivery, toxicity monitoring, long-term surveillance, and drug development. Yet, imaging is generally not featured in the design and validation of precision oncology. We would like to point out that to bring personalized medicine to the next level, the field must embrace imaging technologies to map the spatial and temporal composition of tumors and surrounding host tissue (Fig. 1). In other words, to optimize the efficient use of resources and to maximize clinical utility, developments in precision imaging need to be closer aligned with precision medicine roadmaps. This is even more important given the long lead times of technology development compared to those in the more established, well-funded, and reimbursed drug development pipeline. This article discusses the most recent imaging advances that enable precision oncology and summarizes current needs to further the field.

Fig. 1: Precision imaging and its role in precision oncology.

Precision oncology has historically been defined by the use of genetic mutations to identify patient populations who will respond to a given drug/dose combination. Precision oncology has practical limitations, partly related to cancer’s biological diversity and spatial/temporal complexity. Precision imaging technologies are being developed to address some of the current limitations. These emerging imaging methods strengthen precision oncology by providing clinically relevant information not obtainable by other means (brown box).

Advanced precision Imaging: An Overview

Broadly, we define precision imaging as advanced imaging approaches that allow increased anatomic coverage, improved spatial and temporal resolution, multiplexing, high-throughput screening, targeted sampling of tissue, and precise delivery of new therapeutics (Fig. 2). Fig. 1 summarizes some of the reasons why these advances are essential to precision oncology. Total-body imaging, for example, expands surveillance coverage and allows for better delineation of where the disease is located. Hybrid imaging systems improve the spatial information important for tumor heterogeneity assessment and permit the acquisition of multiplexed data at the anatomic, physiologic, and molecular levels, allowing stratification. Emerging new image-guided approaches add to the armamentarium of precision oncology and enable additional treatments, often with better patient acceptance than standard therapies.

Fig. 2: Examples of precision imaging during the treatment history of cancer.

A. Case history of a patient with cancer. The lesion size is plotted as a function of time and for primary tumor, locoregional invasion, and distant metastases. Note the frequent imaging. B. Role of precision imaging during different cancer stages. The grey boxes with checkmarks represent the main applications. Note: low-dose imaging is mostly for lung cancer (CT) or breast cancer screening (mammography). IO imaging: intraoperative imaging; radiopharmaceutical therapy (RPT).

Many of the advances in imaging technologies over the last 5 years have been “under the hood” but have vastly improved our current ability to perform anatomic/physiologic mapping, staging, and response monitoring (Fig. 2). Beyond hardware advances, the major driver has been computational advances in artificial intelligence (AI) that have provided new ways for algorithms to reduce scan time, increase image quality, reduce the patient radiation dose (e.g., in computed tomography, CT), automatically quantify radiographic patterns in images, and offer many other advantages. Today, faster imaging has major benefits, including artifact reduction, higher throughput, better coverage, better patient tolerance, and greater affordability. Better spatial resolution has allowed for seamless reconstructions in different planes, the ability to perform CT and MRI angiography, and better staging. Dose reduction helps with screening and the ability to perform more frequent follow-up imaging and pediatric studies. Targeted molecular imaging opened new frontiers in therapy selection and treatment monitoring. Some of the salient advances are summarized below.

Total-body PET

Over the last five years, long-axial field-of-view positron emission tomography (PET) scanners have emerged as an exciting new tool for improving image quality and reducing scan time¹. This development helps to address a key question in precision oncology: where is the disease located (Fig. 1)? The new-generation scanners (e.g., μEXPLORER, PennPET, Siemens Biograph Vision Quadra) can acquire images of the main body organs simultaneously or image the total body with a single bed position² (Fig. 3). The improved sensitivity and spatial resolution of the scanners mean that a PET study can be acquired in a much shorter time (e.g., 2–4 minutes), and with markedly better image quality, than is possible with standard digital PET-CT scanners³. Therefore, the time a patient spends in the scanner is much reduced, leading to enhanced tolerability. In addition, it is possible to significantly reduce the dose of radiopharmaceutical given for a scan without compromising image quality, hence reducing the radiation dose received by the patient^1,2. The ability to perform whole-body dynamic imaging also has immediate relevance for the evaluation of the kinetics and biodistribution of novel radiopharmaceuticals³. The cost of the new total-body PET-CT scanners is higher than that of standard PET-CT systems. However, given that they enable higher patient throughput and decreased radiation doses while providing incremental diagnostic information, they will likely be implemented widely in clinical care and research in the next few years.

Fig. 3: Total-body PET Imaging.

Examples of studies enabled by total-body PET scanners that image the entire body at once with high detection sensitivity. A. Maximum-intensity projection (MIP) showing total-body parametric image for blood flow using 11C-butanol and quantified in absolute units of mL blood/min/cm³ tissue calculated with kinetic modeling⁷⁸ (adapted, with permission, from⁷⁹). B. MIP of the total-body distribution of [⁸⁹Zr]Zr-crefmirlimab, an antibody fragment that binds to CD8+ T-cells (adapted from⁸⁰). Uptake is observed in the spleen and bone marrow, with exquisite delineation of lymph nodes throughout the body. Image obtained 48 hours post-injection; the injected dose was only 18 MBq to enable repeat imaging.

Hybrid imaging systems

Hybrid molecular imaging refers to the acquisition and integration of information from functional and anatomic imaging, such as with PET-CT, single photon emission computed tomography (SPECT-CT), or PET-MRI (magnetic resonance imaging). Such imaging has two main advantages: i) display of molecular information on top of anatomic maps (the latter of which are often used for image-guided intervention) and ii) improvement of image reconstruction algorithms for PET or SPECT through the use of CT or MRI. The clinical adoption of PET-MRI in precision imaging has been relatively slow, although MRI by itself is used extensively in breast, prostate, liver, and brain cancers. Recent studies have shown that the combined information from PET and MRI is particularly useful in predicting outcomes in lymphoma after CAR-T-cell therapy⁴, identifying intraprostatic lesions⁵, and predicting overall survival in glioma⁶. In addition to having diagnostic advantages, PET-MRI is preferred over PET-CT in pediatric patients and in patients whose diagnostic work-up requires information from both PET and MRI (e.g., patients with advanced-stage cervical cancer or prostate cancer).

Molecular imaging agents

Over the last decade, many new molecular imaging agents have been tested preclinically⁷ or through first-in-human microdose exploratory investigational new drug (IND) applications or new drug applications (NDAs), and a smaller number have entered the market⁸. The latter include [⁶⁸Ga]Ga-dotatoc (2019), [⁶⁴Cu]Cu-dotatate (2020), [¹⁸F]Fluoroestradiol (2020), and various prostate-specific membrane antigen (PSMA)-targeting PET probes ([⁶⁸Ga]Ga-PSMA-11, [⁶⁸Ga]Ga-gozetotide, [¹⁸F]F-piflufolastat)⁹. Additional examples in the pipeline include agents for fibroblast activation protein (FAP; Fig. 4)¹⁰ and a large number of labeled antibodies for receptor imaging and cell tracking (ImmunoPET)¹¹. The availability of these new probes improves the physiologic, pharmacologic, and molecular profiling of cancers and enables the assessment of their heterogeneity (Fig. 2 and 4). Novel molecular imaging agents could also be co-developed as companion diagnostics for next-generation therapies.

Fig. 4: Whole-body PET imaging with new fibroblast activation protein inhibitor (FAPI) tracer.

Thirty-eight-year-old female patient with a solitary fibrous tumor of the right abdominal wall presenting with lung, peritoneal and bone metastases. A. The left two images represent [¹⁸F]FDG PET maximum projection images before and after 4 cycles of treatment. B. As the patient exhausted all treatment options, a [⁶⁸Ga]Ga-FAPI-46 PET-CT was performed (second from right), displaying high FAP uptake in all [¹⁸F]FDG-avid lesions. After four cycles of ⁹⁰Y-FAPI-46 RLT, restaging revealed partial response according to RECIST criteria. Figure courtesy of Helena Lanzafame, Rainer Hamacher and Wolfgang Fendler (Universitätsmedizin Essen, Germany).

Companion diagnostic imaging

The integration of imaging agents as “companion diagnostics” is another frontier in imaging for precision oncology. The goal is to delineate a patient’s disease landscape, guiding clinicians with unparalleled clarity toward the most effective therapeutic avenues. The convergence of imaging technology with standard-of-care and new therapies presents an unmatched opportunity to improve outcomes. Several examples stand out. For example, [⁸⁹Zr]Zr-DFO-SC16.56 is being used for the non-invasive in vivo imaging of delta-like ligand 3 (DLL3)-expressing malignancies¹², which has been the focus of emerging therapies such as antibody-drug conjugates, T-cell engager molecules, CAR-Ts for small cell lung cancer¹³ and other neuroendocrine neoplasms¹⁴. Another example is the use of 16β−¹⁸F-fluoro-5α-dihydrotestosterone ([¹⁸F]FDHT), a PET tracer for detecting androgen receptor expression in prostate and breast cancers, which is used to study new therapeutic agents^15,16.

Metabolic imaging

The success of [1⁸F]Fluorodeoxyglucose ([¹⁸F]FDG) PET has firmly established metabolic imaging in oncology. However, [¹⁸F]FDG PET cannot reliably distinguish between glucose uptake in normal cells (oxidative metabolism) and uptake in cancer cells (glycolytic metabolism with excess lactate production). Recently, two MR spectroscopy-based tests that probe dynamic changes in tissue metabolism have emerged: hyperpolarized MRI (HP-MRI) and deuterium metabolic imaging (DMI)¹⁷. Hyperpolarization, most commonly of ¹³C-pyruvate, increases the in vivo MR signal and, thereby, HP-MR image contrast. This allows for rapid acquisition of spatial metabolic maps within seconds of infusion. While technically, all downstream metabolites of HP ¹³C-pyruvate can be assessed, the conversion of HP ¹³C-pyruvate to HP ¹³C-lactate is the most relevant HP-MRI parameter to assess cancer metabolism. There are several other HP probes in preclinical research, including 13C urea, 13C fumarate, and ¹³C dehydroascorbate (DHA). The shortcoming of HP-MRI is the loss of hyperpolarization and associated signal decay that occurs within minutes after intravenous injection. DMI enables metabolic imaging for >1 hour after oral administration of deuterated (²H-) glucose, and assessment of the conversion of ²H-glucose to ²H-lactate. However, DMI is limited in the spatial resolution and sensitivity that can be achieved for a target organ. Both HP-MRI and DMI hold clinical potential for in vivo interrogation of tumor biology and early assessment of treatment response. As MRI scanners are far more abundant than PET scanners, HP-MRI and DMI – although presently still experimental– could potentially improve patient management and speed up treatment decisions.

Smarter biopsies

Precision oncology has increased the need for accurate tissue sampling (Figs. 1, 2). Advanced imaging has especially contributed to identifying specific sites for targeted biopsies. Tissue sampling is necessary to better understand the heterogeneity of cancer, measure temporal changes in expression profiles and somatic mutations, analyze immune cell infiltration as a response to treatment, perform spatial biology for biomarker discovery, and supply tissue aliquots to drug trial sponsors as part of their reporting requirements. While core biopsies (16–18G) have many advantages for tissue acquisition, they also have some downsides, including procedural complication risks in certain types of biopsies, long turnaround times, and a substantial fraction of “non-diagnostic specimens” in part because of conventional analyses that require large sample sizes. For all of these reasons, new, “smarter” types of biopsies are being explored to improve information content, reduce complication rates, and improve throughput^18,19.

Fine-needle aspirates (FNA) are performed with 21–25G needles and typically yield cells rather than core tissue. FNA sampling has been firmly established for conventional cytopathology and flow cytometry analysis in the case of lymphoma. More recently, highly multiplexed analyses of scant specimens have been developed by using DNA origami²⁰ and bioorthogonal chemistries²¹ for staining and which do not require the use of harsh chemicals that would destroy cells during cycling. Point-of-care analytical systems are being explored for multiplex sample processing, especially for resource-constrained locations and countries²².

Fine-needle biopsies (FNB) utilize small-gauge (20–22G) specialty needles that are designed to obtain microcores with preserved tissue architecture²³. The microcores are commonly procured by endoscopic ultrasound, but fine-needle biopsies are also being adapted to percutaneous procedures.

Percutaneous fiberoptic sampling is yet another method that allows precision biopsy. A new development in the interventional community has been fiberoptic-guided biopsies (Spyglass) with forceps²⁴. This emerging field is particularly suited for otherwise inaccessible locations, such as peripheral intraductal cholangiocarcinoma. A second and complementary development is the adaptation of this miniaturized cholangioscope to multiplexed fluorescence imaging using probes, an approach similar to fluorescence-guided surgery. These probes can be designed to “highlight” cancer features, which can be either diagnostic by themselves or biopsied for further analysis.

INTEGRATED DIAGNOSTICS AND THE COMPLEMENTARY ROLES OF LIQUID BIOPSIES AND IMAGING

Integrated diagnostics is an emerging field that involves the use of complementary imaging, laboratory biomarkers, pathology, and patient demographic data augmented with information technology. Integrated diagnostics will: i) offer greater diagnostic accuracy than single tests by identifying complementary biomarkers, ii) shorten the time from diagnosis to delivery of molecularly informed therapy, and iii) improve the longitudinal monitoring of outcomes through alternating use of redundant biomarkers²⁵. Including diagnostic imaging in AI-based integrated diagnostic algorithms is essential for developing the strongest possible predictive and prognostic biomarkers to direct precision oncology.

Liquid biopsy²⁶ refers to the sampling of body fluids such as peripheral blood for analysis of analytes that include cell-free DNA (cfDNA), circulating tumor DNA (ctDNA), circulating tumor cells (CTC), extracellular vesicles and proteins. Similarly to imaging, the field of ctDNA analysis is changing rapidly. Recent introductions of fragmentation analyses²⁷, mutational signatures²⁸ and analysis of repeat elements²⁹ have made ctDNA analysis much more accurate including for early stage disease. It is generally believed that liquid biopsy complements advanced imaging and allows up or downstaging of intermediary imaging findings. However, future studies are required to compare methodologies head to head for specific cancers and clinical indications. Thus, AI-informed integrated diagnostics will likely play a growing role in the future. Applications for liquid biopsies in oncology, and their relationships to imaging, are described below.

Early detection.

Several large-scale clinical trials, such as those using the multi-cancer detection Galleri^® test (ISRCTN91431511; based on cancer-specific DNA methylation patterns), reported high specificity (99–100%) but only moderate sensitivity^30,31. Sensitivity depends heavily on the tumor stage – not all patients with localized early tumors will have sufficient quantities of circulating tumor DNA³². In one recent study, ctDNA of KRAS mutation alone could only diagnose 25% of stage 1 pancreatic cancer³³. Similar results were also observed in a large cohort of gynecological, lung, or GI tract cancers (sensitivity for detecting stage 1 cancer: 24.2%)³⁴. For certain cancers such as lung cancers, imaging tests can be superior to liquid biopsies. For example, a recent study on lung cancer screening showed the sensitivity of CT to be >80%, with a negative predictive value (NPV) of 97.7–100%, but a positive predictive value (PPV) of only 3.3–43.5%³⁵.

Finding the primary.

In patients with metastatic disease at initial presentation, liquid protein biomarkers may be helpful in identifying the primary tumor, which can then be confirmed with anatomic or molecular imaging followed by image-guided biopsies.

Treatment monitoring.

Liquid biopsy and imaging biomarkers may complement in outcome prediction and prognostication, given that both types of biomarkers by themselves have demonstrated potential in different malignancies and treatment scenarios. To date, data on their integrated use are scarce; however, a recent study in follicular lymphoma reported 88% sensitivity and 100% specificity for 2-year progression-free survival for the combination of [¹⁸F]FDG-PET and ctDNA³⁶. Some emerging studies indicate that liquid biopsies may be more sensitive than imaging in certain mutation-positive patients and will likely lead to the generation of new guidelines.

Gauging minimal residual disease (MRD) in solid tumors.

MRD is closely associated with disease recurrence, and ctDNA has been explored as a new biomarker. Emerging data suggests that post-operative ctDNA can be a strong prognostic marker of regression-free survival³⁷. It remains to be studied how imaging and liquid biopsy will improve MRD analysis.

Image-guided therapeutics

Radiopharmaceutical therapy (RPT, “radiotheranostics”). RPT refers to the combined imaging and delivery of precision radiotherapeutics, where “what you treat is what you see” (Figs. 4 and 5). The growth of RPT in oncology has been exponential^38,39 in part due to its clinical success, marked by improved outcomes, low off-target toxicities, and better quality-of-life data compared to alternative therapies. While different forms of RPT have been explored experimentally and clinically^38–41, clinical practice has seen a number of mainstream (FDA approved) applications. These have included i) systemic administration of small molecules (e.g., [²²³Ra]Ra-dichloride⁴², [¹³¹I]I-sodium iodide⁴³, radioligands (e.g., [¹⁷⁷Lu]Lu-PSMA-617⁴⁴, [¹⁷⁷Lu]Lu-DOTATATE⁴⁵) or antibodies (e.g., [¹³¹I]I-tositumomab⁴⁶); and ii) selective transarterial tumor embolization with ⁹⁰Y and ¹⁶⁶Ho microspheres.

Fig. 5: Therapeutic approaches involving radiotheranostics.

Therapeutic effects on cancer cells caused by DNA damage induced by either α-, β- or auger-emitting radionuclides can be enhanced via combination with drugs that either cause direct damage to DNA (such as chemotherapies) or inhibit DNA damage repair directly (such as PARP inhibitors) or through modulation of the associated signaling pathways (e.g., with novel androgen-deprivation therapies). Radiotheranostics can also target the tumor microenvironment (fibroblast activation protein) and kill stromal cells, which can indirectly lead to tumor regression. Bystander effects, owing to the use of β-emitters, on the DNA of cancer cells that do not express radiotheranostic target proteins can nonetheless lead to tumor cell death. Targeted radionuclide therapies might also induce antigen presentation following cancer cell death and, when combined with immune checkpoint inhibitors, lead to enhanced antitumor activity. DDR: DNA damage response³⁸.

As the availability of isotopes for diagnostic and therapeutic purposes increases and the number of delivery platforms becomes more sophisticated, truly personalized treatment paradigms can be realized³⁹. Further, many RPT approaches are being deployed in earlier stages of the disease, and ongoing trials are underway in which they are begun after diagnosis. One example is the NETTER-2 trial establishing [¹⁷⁷Lu]Lu-DOTATATE as first line treatment in newly diagnosed patients with advanced grade 2 and grade 3, well-differentiated gastroenteropancreatic neuroendocrine tumors⁴⁷. Another example is in prostate cancer, where treatments are transitioning from usual late-stage metastatic/advanced disease therapy to earlier-line treatment settings such as metastatic hormone-sensitive disease (PSMAddition) and even prior to prostatectomy (LuTectomy)⁴⁸. There is also excitement about the concept of using RPT in combination with established systemic therapies, including targeted therapies and immunotherapy (e.g., NCT04343885, NCT05146973, NCT03874884, NCT05109728, and NCT03658447), as well as for accessing previously unexplored targets that were not suitable for traditional systemic therapies (Fig. 4).

The expansion of the field is not without its challenges⁴⁹, including isotope production and supply^50,51, workforce expertise, and regulations^39,52–54; however, efforts to address these issues have been initiated by governments, industry and major professional organizations, as well as in a current Lancet Oncology Commission which is exploring the global availability of theranostics and recommendations for improving patient access.

Transarterial radioembolization (TARE, also known as selective internal radiation therapy, SIRT) with Yttrium-90 (⁹⁰Y) microspheres is a liver-directed therapy for primary and metastatic disease, for which indications have blossomed following FDA approval and efficacy trials⁵⁵. In TARE, hepatic artery branches are accessed with microcatheters, injecting ⁹⁰Y glass or polymer microspheres. These microspheres lodge in tumor-feeding arteries, and ⁹⁰Y induces locoregional cellular damage. The current indications for TARE include primary treatment of hepatocellular carcinoma (HCC) in nonsurgical candidates, bridging to transplant in HCC, primary treatment of isolated oligometastatic liver lesions (radiation segmentectomy), radiation lobectomy to induce hypertrophy before resection, and palliation or delay of progression for advanced tumor burdens⁵⁶. To improve dosimetry, planning angiography with the administration of ^99mTc macroaggregated albumin is being performed to exclude patients with lung shunting or aberrant abdominal supplies. Cone-beam CT with intraarterial contrast material injection is often performed to exclude non-target perfusion and measure the perfused treatment volume. Bremsstrahlung SPECT/CT or ⁹⁰Y PET can also be performed after TARE for dosimetry purposes and patient-centric future planning. More recently, ¹⁶⁶Ho made its way into the clinic, providing additional therapeutic options complementary to peptide receptor radionuclide therapy (HEPAR Plus trial⁵⁷.

Image-guided percutaneous ablation methods have transformed the approaches to many primary and metastatic tumors, especially in non-surgical patients. The tumor ablation methods include microwave, radiofrequency, and cryoablation (among others, such as electroporation, histotripsy, and chemical ablation)⁵⁸. While the different methods have unique advantages and disadvantages, they generally rely on image-guided precision placement of applicators into cancerous lesions in the liver, kidneys, lung, bone, prostate, thyroid, and soft tissue. Following minutes of precision treatments, patients can be discharged the same day. There are usually short recovery periods, less bleeding, and more preservation of organ parenchyma, thus expanding future treatment options.

Image-guided precision drug delivery allows patient-specific administration of next-generation therapeutics (e.g., viral therapeutics, cell therapeutics, immune modulators) through catheters and needle-based approaches. This allows for high local drug concentrations while minimizing systemic therapies. Precision drug delivery may be of particular value to early drug developers in efforts to drive new applications.

Intraoperative Imaging

Intraoperative imaging with ultrasound, fluoroscopy, or MRI has long been used for conventional surgery (Fig. 6). In recent years, however, there has been a shift to precision surgery, where resections are individualized to patients. This potentially allows for improved resection accuracy, sparing essential structures, and faster convalescence. These trends and public demand have driven the growth of minimally invasive surgery, especially robotic surgery and image-guided surgery technologies. Most applications have been in prostate, breast, colorectal, lung cancer, and glioma resections. For example, the main challenge in radical prostatectomy is complete cancer excision with the preservation of continence and erectile function. Positive margins still occur in up to 35% of cases. In breast-conserving surgery, the primary goal is to prevent local recurrence with acceptable cosmetic outcomes. Rates of residual cancer following initially negative lumpectomy margins have been shown to exceed 40% in some studies⁵⁹.

Fig. 6: Fiberoptic and intraoperative imaging.

Fluorescence-based cellular and molecular imaging is an up-and-coming modality, primarily employed for endoscopic and intraoperative imaging. A: results from fluorescent guided surgical detection of residual tumor in the resection cavity using the recently FDA-approved Lumicell system^63,64. B: the use of bioorthogonal chemistry has enabled, in vivo, cyclic imaging for up to 12 different targets⁸¹.

Fluorescence-guided surgery (FGS). Near-infrared fluorescence imaging uses a combination of injectable fluorescent imaging agents with specialized detection systems to visualize cancers and their margins more accurately. The emergence of the field dates back nearly 25 years⁶⁰, and a good many different agents and imaging systems have been developed since then^61,62. Despite these efforts, it has been challenging to commercialize systems and then test their efficacy prospectively in large-scale studies. A recently FDA-approved system is Lumicell’s activatable fluorescent imaging agent (pegulicianine)⁶³ combined with a handheld device⁶⁴. In one recent prospective trial involving 406 patients (NCT03686215), the margin status was assessed with or without pegulicianine fluorescence-guided surgery (PFGS). In 27 of 357 patients undergoing surgery, PFGS for surgical margins removed tumors left behind after standard lumpectomy. PFGS prevented second surgeries in 9 of 62 patients with positive margins⁶⁵

Alternative methods are being explored to enable intraoperative imaging. Early proof-of-principle studies primarily validate the emerging technologies in resection specimens, but the overall future goal is imaging of the resection cavity. The latest modality to be evaluated for assessment of margins is PET-CT⁶⁶. Additional methods include Cherenkov, Raman and photoacoustic imaging⁶⁷.

Artificial intelligence in Precision Imaging

Discussions on artificial intelligence (AI) have become central to precision medicine⁶⁸. For the last several years, AI has been proposed to increase sensitivity in disease detection, enhance measurement reproducibility, reliably extract quantitative disease markers, recognize new patterns encoded within complex data, and serve as an inference engine after empirical “training” using real-world, large-scale data. Since its earliest days, AI has been integral to the imaging sciences^69,70.

Most AI-driven imaging workflows focus on the back end of care management, including tumor detection, segmentation⁷¹, and staging, serial monitoring of tumor metrics, defining prognosis, radiation treatment planning⁷², and determination of tumor heterogeneity⁷³. Vendors have rapidly adopted AI at the front end of imaging care to improve image quality, accelerate imaging times, calculate dosimetric profiles for theranostics, and predict whether a patient’s tumor has a sufficient amount of target to undergo targeted therapy.

Despite the tremendous promise noted above, the use of AI in precision imaging is not without certain bottlenecks and controversies. Among the chief bottlenecks is the lack of adequate amounts of data to train convolutional neural networks (CNNs). To decrease the need for manpower and tedious annotation of images in preparation for training, we would need to focus on AI unsupervised methods. However, that is not always possible, as available datasets are either too small or cannot provide reliable images due to non-standardized acquisition parameters⁷⁴. That has led to the development of methods for creating synthetic images⁷⁵, for example, through generative adversarial networks⁷⁶, transfer learning and collaborative/federated learning⁷⁷. Trustworthiness has become a mounting challenge to the use of AI, even with large-scale datasets. Algorithms can be biased toward one racial group or gender, produce unpredictably erroneous results, and make predictions that have no basis in physical reality. Most of the data presented in the literature to date are from retrospective studies. Prospective studies comparing AI performance to more traditional decision-making processes are needed to define its true potential.

Additional Challenges and Future Developments

While the advances in imaging technologies over the last decade have been nothing short of astounding and hold exceptional promise, several opportunities present themselves to facilitate more rapid progress in precision imaging for oncology:

• The co-development of companion imaging diagnostics will ensure the success of next-generation patient-specific therapies. Unfortunately, developing and validating imaging probes is often an afterthought and still underfunded. Unlike therapeutic development, imaging agent development is mostly investigator-driven, relying on lengthy funding cycles and often delaying clinical introduction. With the recent exceptions for [⁶⁸Ga]Ga-PSMA-11 (Illuccix^®, ⁶⁸Ga gozetotide; 8 years to FDA approval in 2020), [¹⁸F]F-PSMA (Pylarify^®, ¹⁸F piflufolastat; 11 years to FDA approval in 2022) or [¹⁷⁷Lu]Lu-PSMA (Pluvicto^®, ¹⁷⁷Lu vipivotide tetraxetan; 7 years to FDA approval in 2022⁴⁴) it is estimated that the average length of imaging approval is still ~13 years, too long to be practical for precision oncology needs. The development and validation of companion imaging diagnostics should occur in the preclinical stage. Many first-in-human imaging studies could be done rapidly as phase I trials to demonstrate safety and proof-of-concept. Currently, phase III trials are expensive and burdensome if industry is not involved, and regulatory approvals are often held to the same standard as for therapeutic drugs. Therefore, regulatory bodies should develop imaging-agent/modality-specific pathways so that approval can be fast-tracked and phase III trials will not require the large-scale cohorts associated with therapeutic drug trial design.

• For precision oncology and imaging to be synergistic, there has to be continued dialogue and collaboration between different specialties. Except at specialized cancer centers, oncology and imaging/diagnostic departments often continue to operate differently for historical and financial reasons. While a primary oncologist typically follows a single patient longitudinally, different imaging physicians will perform and interpret scans during a patient’s journey (Fig. 1). Naturally, this can lead to discrepancies and inter-observer variability, complicating longitudinal assessments. It is hoped that AI and multidisciplinary conferences will present venues for minimizing discrepancies in longitudinal image analysis.

• While imaging centers have been at the forefront of AI developments, tighter integration with digital pathology, molecular diagnostics, and AI in clinical oncology seems logical and well justified. Together, we must find ways to manage the growing IT costs and also invest in the future of integrated diagnostics.

• Capital costs of imaging systems, infrastructure, and new therapies are considerable and have to be harmonized across academic medical centers. Confounding the issue is that reimbursements are decreasing for imaging, just as they are for almost all specialties. Despite the above challenges, there continues to be enormous enthusiasm for future developments and tighter integration of advanced imaging in clinical trials and care.

Summary

Recent advances in medicine and biotechnology have enabled more personalized cancer therapy approaches. However, all too frequently, only limited biomarker analyses are performed on tumors, often before treatment. Because of the dynamic nature of tumor evolution, combating therapy resistance is a game of catch-up requiring sophisticated tools to measure evolution. Here, we have attempted to summarize the extraordinary breadth of progress in developing diagnostic and therapeutic imaging approaches and how these can aid in precision oncology. Advanced imaging and image-guided treatments will continue to play essential roles in precision oncology and will improve cancer outcomes and survival.