Developing and validating quantitative imaging biomarkers that reflect underlying histologic and/or genomic features may provide a novel way of noninvasively assessing tumor heterogeneity and predicting drug response or resistance in oncology. The advent of molecularly targeted therapy and anti-angiogenic strategies has heightened the need for sensitive biomarkers for predicting treatment response and outcome, as resistance to chemotherapeutics and targeted therapies is common. Genetic intratumoral heterogeneity may contribute to treatment failure by initiating phenotypic diversity that introduces tumor sampling bias and enables drug resistance to emerge. In patients with primary and metastatic renal cell carcinoma, multiregion genetic analyses showed that spatially heterogeneous somatic mutations and chromosomal imbalances lead to phenotypic intratumoral diversity and that a single tumor biopsy specimen reveals a minority of the genetic aberrations present in an entire tumor [25]. Using phylogenic tree analysis to evaluate the relationships between tumor deposits in patients with ovarian cancer, Cowin et al. [26] found substantial copy number differences between metastatic deposits within individual patients and identified signaling pathways plausibly linked to peritoneal dissemination and establishment of metastatic foci. Significantly greater genomic change was observed in patients who experienced relapse after responding to chemotherapy than in patients who were resistant from the outset, possibly reflecting the requirement for selection of a subpopulation of resistant cells in cases initially sensitive to treatment [26].

Morphological heterogeneity between and within tumors is readily apparent in clinical imaging; subjective expressions of these differences, such as spiculated, enhancing, and necrotic, are common descriptors in the imaging lexicon. In the past several years, imaging research has focused on quantifying these image features in an effort to understand their biological and clinical implications. More recently, in an approach referred to as “radiogenomics,” quantitative shape, border, and texture analysis features (radiomics) obtained from CT, MRI, and PET have been correlated with genomic data and outcome data with the goal of developing robust biomarkers. Following extraction of more than 100 CT imaging features, Segal et al. [27] found that a subset of 14 features predicted 80 % of the gene expression pattern in hepatocellular carcinoma. A similar extraction of features from MRI of glioblastoma was able to predict protein expression patterns [28]. Radiomic features such as texture can predict response to therapy in renal cancer [29] and overall survival in primary [30] and metastatic colon cancer [31]. Analysis of heterogeneity on DCE-MRI and DW-MRI has been shown to improve diagnosis and assessment of prognosis for several tumor types [32–35]. Furthermore, measures of metabolic heterogeneity at baseline PET were able to predict response to chemoradiotherapy in patients with esophageal cancer [36].

Although radiomics analyses have shown a high degree of prognostic power, they are not spatially explicit. Radiomic features are generated over the entire tumor, using the assumption that tumors are heterogeneous but well mixed. However, it is readily apparent from DCE-MRI that perfusion can vary markedly within spatially distant tumor subregions that may represent distinct phenotypic habitats. Therefore, image-guided multiregional tumor analysis may be required to fully characterize tumor heterogeneity.

Optical imaging techniques have been developed to guide minimally invasive procedures (e.g., involving endoscopy or robotic techniques) as well as conventional surgical procedures. The use of fluorescence during routine endoscopic screening examinations in the gastrointestinal, bronchial, and urinary tracts is proving to be invaluable for detecting occult dysplastic lesions [37]. Furthermore, intraoperative optical imaging using fluorescence has been applied for tumor margin delineation and to identify malignant nodes. For example, optical imaging after the injection of indocyanine green (ICG) has been used to delineate tumors in patients with brain cancers as well as to map sentinel lymph nodes in patients with breast cancer [38, 39]. Recent research suggests that in ovarian cancer, tumor-specific intraoperative fluorescence imaging using a folate receptor-targeted fluorescent agent may improve intraoperative staging and allow more radical cytoreductive surgery [40].

Nanoparticle-based probes that contain more than one type of targeting component offer possibilities for applying multiple molecular imaging modalities in quick succession to improve gross tumor resection as well as subsequent fine margin resection. For instance, in studies using animal models, one probe tested allowed gliosarcomas to be identified by both preoperative MRI and intraoperative optical imaging [41], while another allowed visualization of glioma by MRI, photoacoustic imaging, and Raman imaging [42].

Molecular imaging can play an important role in radiotherapy planning. Higher radiation doses can be selectively applied in areas of tumor that show high metabolic activity and are therefore likely to represent especially aggressive disease. For example, FDG-PET/CT-based radiation planning for lymph nodes has been found to change the radiotherapy field in 21–25 % of patients with lung cancer [43–45]. In a study of 41 patients with head and neck cancer, the radiation boost dose was markedly increased and targeted at the tumors with the highest FDG avidity [46]. The study indicated that a boost dose of 3.0 Gy/fraction to ≤10 cm³ could be safely applied to the FDG-avid regions during the first 2 weeks of conventionally fractionated IMRT. A majority of local relapses occurred within the FDG-avid regions that had received elevated doses, suggesting that FDG avidity is a marker of the likelihood of relapse [46]. Randomized controlled trials comparing conventional and PET-based radiotherapy of locally advanced non-small cell lung cancer are ongoing (PETPLAN, NCT00697333; RTOG-1106, NCT01507428). These trials will investigate whether local control rates and side effects are improved by PET-based radiation treatment planning.

Fluoro-l-thymidine (FLT)-PET/CT can noninvasively measure tumor cell proliferation and can detect early radiotherapy response. For example, in head and neck cancer patients, rapidly decreasing FLT-PET SUVs have been observed during the course of radiotherapy. Conceivably, the radiation dose could be boosted in selected tumor areas with high proliferative activity identified by FLT-PET [47].

In organs such as the brain, where the accuracy of FDG-PET/CT is limited, MRS or PET with radiolabeled amino acids can provide an alternative means for guiding radiotherapy based on molecular information. The presence of metabolic abnormalities on MRS consistent with tumor outside of the target volume of Gamma knife radiosurgery is associated with significantly shorter overall survival [48]. Tumor volumes defined by PET with the radiolabeled amino acid analog fluoroethyltyrosine (FET) are markedly different from those defined by contrast enhancement or FLAIR signal abnormalities on MRI (Fig. 3.3) [49, 50]. A randomized controlled trial comparing amino-acid PET-based radiation treatment planning with MRI-based radiation treatment planning is ongoing (GLIAA, NCT01252459). This trial will determine whether progression-free survival is improved by PET-based radiotherapy.

Various techniques are available for imaging tumor hypoxia, which increases resistance to radiotherapy and chemotherapy [51, 52]. These techniques include PET with tracers such as [¹⁸F]fluoro-misonidazole (FMISO), [¹⁸F]fluoroazomycin-arabinofuranoside, or [^60/64Cu]copper(II)-diacetyl-bis(N4-methylthiosemicarbazone (ATSM), as well as blood oxygen level-dependent (BOLD) MR imaging, which detects hypoxia based on an increase in the transverse relaxation rate (R2*) of water caused by the paramagnetic effect of endogenous deoxyhemoglobin [53, 54]. Studies are ongoing that evaluate whether local control rates can be improved by applying higher radiation doses to hypoxic tumor subvolumes [55]. However, it remains an open question whether the size and location of hypoxic volumes remains stable in untreated tumors and during radiotherapy.

Imaging of hypoxia has also shown promise for guiding the use of chemotherapeutic drugs that specifically act on hypoxic tumors (Fig. 3.4): The hypoxia-selective drug tirapazamine was not beneficial in unselected head and neck cancer patients but only in the subgroup of patients with FMISO-PET-positive tumors [56, 57].

It is essential to be able to detect and localize residual and recurrent cancers so that timely and appropriate salvage therapy can be administered. However, posttreatment changes (e.g., loss of normal anatomy or the formation of scar tissue) often hamper the identification of such cancers by conventional cross-sectional imaging. By adding metabolic to morphologic information, molecular imaging can help to distinguish recurrent or residual disease. The use of FDG-PET for further evaluation of patients with clinically suspected tumor recurrence is well established in many malignancies including head and neck squamous cell carcinomas, lung cancer, rectal cancer, and ovarian cancer. In these patients PET/CT can guide local therapeutic strategies such as resection of solitary liver metastases [49, 58–63]. However, the role FDG-PET for surveillance in asymptomatic patients needs further study. In patients with a low risk of recurrence, the positive predictive value of FDG-PET may be too low to be clinically useful. Conversely, it has not been proven in randomized trials that early detection of asymptomatic recurrences by FDG-PET improves outcome.

MRS can also be a useful supplement to anatomic imaging in the search for recurrence. Studies have demonstrated its ability to help differentiate locally recurrent prostate cancer from benign and necrotic tissue after radiation treatment (Fig. 3.5) [65], hormone deprivation therapy [66], and cryosurgery [67]. DW-MRI and DCE-MRI are also very useful in evaluating local recurrence of prostate cancer and have been incorporated in routine clinical practice (Fig. 3.6) [68, 69].

For detection of lymph node and bone metastases, PET/CT with radiolabeled choline ([¹¹C]choline) or choline analogs ([¹⁸F]choline or [¹⁸F]fluoroethylcholine) (Fig. 3.7) has been found to yield promising results in patients with recurrent prostate cancer [70–72]. Choline PET/CT can detect metastatic prostate cancer at low PSA levels and may guide local therapies such as secondary lymph node dissection [70, 71].

The use of FDG-PET to monitor tumor response to therapy is well established in Hodgkin’s lymphoma (Fig. 3.8) and aggressive non-Hodgkin’s lymphomas. In these diseases, response status is now predominantly dependent on the FDG-PET findings. PET is also increasingly used in lymphomas and many solid tumors to determine tumor response early in the course of therapy.

PET-guided response-adapted therapy has been most extensively studied in Hodgkin’s lymphoma and diffuse large B-cell lymphoma. Several studies have shown that tumor response after two to three cycles of chemotherapy is strongly predictive of progression-free survival [73]. Thus, treatment may be intensified in patients without a metabolic response on PET or de-escalated in patients with a favorable response on PET.

Response-adapted therapies are also being explored in solid tumors. In esophageal cancer quantitative changes in tumor FDG uptake after 2 weeks of preoperative chemotherapy were shown to be predictive of histopathologic response and progression-free and overall survival. Based on these data, a phase II study evaluated response-adapted therapy in patients with locally advanced distal esophageal cancer [74]. In patients for whom PET failed to show a metabolic response, neoadjuvant chemotherapy was stopped after 2 weeks of therapy and immediate surgical resection was performed. In contrast, patients with a metabolic response on PET received the full 3 months of preoperative chemotherapy. The results of the study indicated that response-adapted therapy is feasible and enriches the histopathologic response rate in the group of patients classified as metabolic responders. These patients demonstrated excellent progression-free and overall survival. In the group of metabolic nonresponders, progression-free survival and overall survival were not inferior to historic controls who were treated with 3 months of chemotherapy irrespective of their response on PET. Ongoing studies are evaluating whether outcome of metabolic nonresponders can be improved by changing chemotherapy early in the course of treatment (CALGB-80803, NCT01333033).

A novel approach to study tumor metabolism during therapy is MRI with hyperpolarized, ¹³C-labeled, metabolic substrates. For example, animal studies have suggested that by detecting changes in lactate levels, such metabolic imaging may be useful for monitoring tumor response to therapy for lymphoma and brain tumors [22, 75]. Hyperpolarized [1-¹³C]pyruvate has also been found to distinguish early-stage prostate cancer from late-stage prostate cancer based on lactate levels [22]. The first clinical trial of MRI with hyperpolarized [1-¹³C]pyruvate was recently conducted in patients with prostate cancer.