Application of radiomics in lung immuno‐oncology

Radiomics is a rapidly evolving field of research that extracts and analyzes quantitative features within medical images. Those features are termed as radiomic features that can characterize a tumor in a comprehensive and quantitative manner with regard to its internal structure and heterogeneity. Radiomic features can be used, alone or in combination with demographic, histological, genomic, or proteomic data, for predicting prognosis or treatment response. Immunotherapy, or immune‐oncology, is the study of cancer treatment by taking advantage of the body's immune system to prevent, control, and eliminate cancer. In this review, we first provide a brief introduction to both radiomics and immune‐oncology in lung cancer. Then, we discuss the need for developing immune‐oncology biomarkers, and the advantages of radiomics in identifying biomarkers related to immunotherapy. We also discuss potential areas in and out of tumors, such as the intra‐tumoral hypoxic region and tumor microenvironment, where radiomic markers might be extracted, as well as a potential application of radiomic biomarkers in clinical lung cancer management. Finally, we present radiation and immune modulation in non‐small cell lung cancer, clinical trials and their design to incorporate radiomic biomarkers, and radiomics‐guided precision radiation therapy.


F I G U R E 1
The number of publications on radiomics and radiomics lung extracted from PubMed database (accessed on April 15, 2021). cancers to identify prognostic phenotypes, 2 predict clinical outcomes in early-stage non-small cell lung cancer (NSCLC) patients treated with stereotactic body radiation therapy, 3 and assist in immunotherapy by estimating CD8 cell infiltration into the tumor and predicting clinical outcomes. 4,5 Studies using radiomics in lung cancer have generated radiomic signatures with markers to predict metastasis and survival. 5 The analysis of tumor heterogeneity has suggested that tumor biology can be reflected through tumor shape and intratumor density variation, from which imaging characteristics can be used to predict tumor development and patient survival. 6  Immune checkpoint inhibitors (ICIs) work on the patient's own immune system to enhance the actions of tumor-specific cytotoxic T cells. An independent prognostic marker of survival for patients with NSCLC is the presence of tumor-infiltrating lymphocytes in cancer cells. 9 Pembrolizumab, a monoclonal immunoglobulin antibody against PD-1, was recently approved by the US Food and Drug Administration for monotherapy for patients with stage III NSCLC who are not candidates for surgical resection or definitive chemoradiation, or have metastatic NSCLC with no epidermal growth factor receptor (EGFR) or anaplastic lymphoma kinase genomic tumor aber-rations. This is based on findings from the KEYNOTE-042 trial. 10 Subsequently, the combination of chemotherapy with checkpoint inhibitor was tested. KEYNOTE 189 and 047 reported that the addition of pembrolizumab to chemotherapy resulted in a clear overall survival (OS) and progression-free survival benefit. OS benefit was observed across all patient subgroups, including a cohort without PD-L1 expression. 11,12 Currently, there is a clinical need for identifying biomarkers that predict response and help selection for an appropriate combination of therapies. 13,14

Need for developing immuno-oncology biomarkers
So far, only PD-L1 IHC has been approved as a diagnostic biomarker by the US Food and Drug Administration for patients with advanced NSCLC for PD-L1 status and patient selection for PD-1/PD-L1directed therapy. 15 However, PD-L1 seems to be not the only biomarker to predict response. There are still many uncertainties about the practical value of PD-L1 in the clinical practice of ICI in untreated NSCLC. For example, pembrolizumab is known to be effective in patients with PD-L1 expression >50%. Interestingly, there is a subgroup of patients who benefited from pembrolizumab with chemotherapy with PD-L1 expression <1%. 12 Identifying biomarkers in this subset of patients with low PD-L1 expression, but who could benefit from pembrolizumab, would be of interest.
For patients with higher PD-L1 levels (expression range of 50%-100%), the overall response rate was 44.4%, indicating >50% of patients with high PD-L1 levels would not respond to immune checkpoint inhibition as expected. Responses and clinical outcomes were improved with PD-L1 expression levels ≥90%. 16 Therefore, additional biomarkers for patients with PD-L1 expression levels <90% would be desirable. Furthermore, mechanisms exist beyond tumor PD-L1 staining from a biopsy that can monitor/predict possible immune responses.
Recent data have shown that tumor-infiltrating immune cell PD-L1 expression has a stronger association with treatment response than that of tumor cell PD-L1 expression. 17 The baseline density and location of CD8 + T cells at the margin or the core of tumor, and the pre-existing CD8 T cells that are negatively regulated by PD-1/PD-L1 could determine antitumor activity by PD-1 immune checkpoints. 18 Regulatory agencies and professional societies, such as the Society for Immunotherapy of Cancer, have recognized the need for developing immuno-oncology biomarkers for clinical use. 19 Immune biomarkers will monitor disease progression, decide when to alter/stop costly immunotherapy treatment, and potentially enable a better selection of patients for cancer immunotherapies while avoiding adverse events.

Advantages of radiomics in identifying immunotherapy-relevant biomarkers
Several tumor-derived biomarkers have been reported, such as PD-L1 expression on tumor cells (and immune cells), tumor mutational load/burden, DNA mismatch repair genes and their products, and multigene signatures (tumor microenvironment). 20 These tumorbased biomarkers are inherently subject to sampling bias. Nextgeneration sequencing and bioinformatics have shown that there is heterogeneity intratumorally, which suggests that the tumor is not homogenous, but a complex system. 21 Multiple site biopsies have revealed that there is spatial composition and different evolution of different subclones in a given tumor. [22][23][24][25] Hence, a single biopsy or acquiring material for biomarker analysis will lead to omitting intratumor heterogeneity and other subclones, which generate either limited information or lead to rapid resistance when resistance mutations were not detected in the sampling process. 26,27 Alternatively, radiomics could complement these biomarkers and, thus, lead to a more precise, multimodal prediction. As aforementioned, the study of spatial intratumor heterogeneity is an inherent nature of radiomics, which can see the whole tumor in a whole elephant approach and avoid the sample limitation from a simple tumor biopsy or limited sampling from the whole tumor. Radiomics cannot detect the biological ancestry of cells in a given tumor yet, but can detect areas that harbor different clonal populations intratumorally. It also has the potential to be combined with liquid biopsy to extract valuable information in determining tumor aggressiveness. 28 Application of radiomics in evaluating tumor immune cell infiltration and predicting response to anti-PD-1 therapy has been promising: a radiomic signature that measures the intensity of tumor-infiltrating CD8 cells was identified. This CD8 cell signature was validated with pathologists quantifying tumor-infiltrating CD8 lymphocytes from tissue slides from biopsy samples corresponding to the primary tumor and matching computed tomography (CT) scans from The Cancer Genome Atlas. After confirmation of validity, this signature was used to evaluate patients with advanced solid tumors treated in phase I trials of PD-1/PD-L1 monotherapy, which showed that a higher score in this signature is related to longer OS. 4 Recently, Gilles et al. used the multiparametric radiomics signature from features extracted from positron emission tomography (PET)/CT images to predict the response to immunotherapy. Their findings suggested that tumors that contain radiomics signatures associated with more heterogeneous tumors, more convexity and lower mean standard uptake value and Hounsfield unit had a larger probability of responding to immunotherapy. Also, the predictive result with an area under the curve of 0.82 showed promise of the radiomics signature being used as a predictive biomarker to guide immunotherapy. 29

Potential application of radiomic biomarkers in lung cancer management
Radiomics-identified biomarkers can potentially be used to guide treatment decisions, predict prognosis, and/or monitor treatment response.
With prognostic and predictive immune biomarkers, clinicians can assess the responses throughout treatment, predict patient benefits from therapeutic agents, monitor their response, and enable personalized treatment plans.
For example, the association of CT features and genetic markers, such as the mutational status of the KRAS mutation and EGFR mutation in NSCLC, has been studied. 30 There are some potential features (shape, pleural retraction, internal air bronchogram) that could suggest a correlation with genetic changes. Several other groups conducted CT-based studies to predict the EGFR mutation in lung adenocarcinoma. [31][32][33] One recent prediction model was built on radiomics features, such as shape and size, and textual and wavelet features. The AUC value was 0.802. However, the sensitivity, specificity, and accuracy were still not satisfactory for clinical use. 33 Radiomics is non-invasive, and can be performed readily in a rapid manner to predict or monitor treatment response. Radiomics analysis has shown that abnormal texture presented on baseline 18Ffluorodeoxyglucose PET, such as coarseness, contrast, and busyness, are associated with non-response to chemoradiotherapy and with poorer prognosis. A prognostic radiomic signature that correlated to intratumor heterogeneity is associated with gene-expression patterns and proliferation of tumors.

2.4
Potential areas in and out of tumor that may provide radiomic biomarkers

Intratumor hypoxic regions
Hypoxia is a common scenario among solid tumors. Indeed, it enables many events in the tumor microenvironment that lead to the expansion of aggressive clones that adapt to hypoxia. Hypoxia in tumors is related to poor clinical prognosis, 34 Hypoxia is known to contribute to the upregulation of the expression of PD-L1 and establish an immunosuppressive tumor microenvironment. 42 Also, hypoxia harms CD8 + T cells in vitro and global antitumor immune response in vivo. 40 The efficacy of the PD-1 blockage is potentiated by the metformin-induced reduction of tumor hypoxia in preclinical mice models. 39 Furthermore, metformin also enhances the PD-1 blockade through a reduction of tumor hypoxia.
Patient outcomes are improved with metformin in combination with ICIs without reaching significance, which hints that reversing hypoxia can contribute to better immunotherapy outcome. 43 Radiation studies that targeted hypoxic regions of tumors contribute to improved tumor abscopal response without the use of ICIs.
Fluoromisonidazole PET can detect tumor hypoxia, but it is not widely available. 44 Radiomics methods have been shown to be able to identify hypoxic regions in tumors. It has been shown that imaging features extracted from CT and 18F-fluorodeoxyglucose PET imaging can be correlated with the magnitude of hypoxia in tumors as detected by fluoromisonidazole PET. 45 The collection of radiomics signatures from these hypoxic regions might provide biomarkers for immune resistance and be used to select tumors that are not likely to respond to immunotherapy without the need of specialized imaging devices, and can be applied to general CT images.

Tumor microenvironment
The microenvironment surrounding the tumor mass is a dynamic field of tumor growth, and ongoing interaction of malignant cell expansion and cells that work against this progress. 46

Radiation and immune modulation in NSCLC
Radiation therapy is an integral part of lung cancer management.
Stereotactic body radiation therapy is used for early-stage lung cancer, 52 whereas chemoradiation is reserved for locally advanced NSCLC. 53 The application of radiation therapy is fairly limited in stage IV lung cancers, mainly for palliation purposes. Approximately 40% of patients with NSCLC are diagnosed with stage IV lung cancer on diagnosis. 54 These patients are frequently treated with PD-1 checkpoint inhibitors, 55 and radiation has been suggested to play a synergistic effect with anti-PD-1 therapy. 56 Thus, radiation doses have the potential to be better utilized as a trigger for immune priming.
Radiomics not only defines tumors in a biological manner, but also provides a map for high-risk regions and intratumor regions that can potentially modulate immune responses, through which it can guide function-based radiation planning and delivery.
The biology of radiation on the human body is not yet fully understood. Radiation therapy is generally believed to exert its therapeutic effects exclusively locally within the irradiated field to cause direct and indirect cellular DNA damage. 57 A great deal of radiobiology research was dedicated to DNA damage. Distant effects of radiation were found in the 1950s, and termed the abscopal effect by Mole from the Latin words "ab" and "scopus." 58 The abscopal effect refers to the regression of metastatic cancer outside of the irradiated field, which suggests that the immune system is modulated to combat cancer in the whole body as a result of the local therapy.
Radiation therapy was initially considered immunosuppressive, owing to the sensitivity of lymphocytes to radiation and the potential of killing tumor-infiltrating lymphocytes by radiation therapy. 59 Radiation is associated with the depletion of circulating lymphocytes 60 and chronic toxicities. 61  However, radiation therapy has recently been shown to enhance various components of the immune system, including but not limited to: generating neoantigen from tumor after radiation to prime T cells, 63 facilitating immunogenic cell death-induced antigen release and proinflammatory signals, 64 activation of cytokine cascades to activate innate immune response, 65 and enhancing T-cell homing, traffic to, and infiltration into tumors. 66 An active area of investigation is varying doses and fractionation of radiation therapy in combination with various immunotherapy agents, as well as evaluating the antitumor immune response in various tumor types. 67,68 Recently, several clinical trials investigating the potential of the combined strategy have been reviewed for both stage III and advanced NSCLC. 69 It is reasonable to hypothesize that radiation could modulate immune functions in its applications, such as dose/fractionation/site/area of delivery. If the hypothesis is proven, radiomics and genomics studies would be valuable to elucidate the effect of radiation on the human body systematically. On occasion, it is difficult or impractical to deliver the higher dose per fraction ideal for eliciting an antitumor immune response, owing to tumor size or location. Under these circumstances, radiation therapy may be delivered by irradiating a fractional tumor volume.
Spatially fractionated radiation therapy is a way to deliver an inhomogeneous radiation dose to a whole or partial tumor. 70  showed that the host immune system could be activated in a different way using partial tumor irradiation. 71 Preclinical studies demonstrated that the hypoxic tumor cells showed a higher abscopal potential than the normoxic cells of the same tumor type. 72 An interesting animal study showed that targeting hypoxic tumor regions with a radiation boost improved the control of tumors relative to a controlled boost with the same dose to nonhypoxic regions of the tumor. This shows that small biological distinct subvolumes within a tumor can determine tumor curability. 73 Another study showed that hemi-irradiation of the mouse tumor volume was sufficient in controlling the tumor and eliciting CD8 + T cell-mediated immune response to the irradiated tumor; furthermore, partial radiation also significantly postponed tumor growth in the contralateral non-treated tumors. 74 One of the attempts to leverage the immunogenic potential of radiation was stereotactic body radiation therapy targeting hypoxic segment of bulky tumors. 75

Radiomics-guided precision radiation therapy
Ling et al. introduced the concept of "biological target volume (BTV)", which represents a subvolume of the tumor with specific characteristics on functional or molecular imaging techniques. 76 Radiomics potentially helps explore the heterogeneity of tumor spatial structure and map BTV. Regions identified by radiomics that either confer radioresistance or elicit immune reactions could be tailored for precision radiation planning, thus better local control or induction of systemic immune reactions can be achieved. We envision the concept of BTV inside the gross tumor could be defined by the aid of radiomics and biological information extracted from images, and then used in radiation planning to guide treatment in a systemic biological manner: various BTVs intratumoral and/or TME will be radiated differently to maximize local debulking effects and enhance immunity. Therefore, this concept is leading to radiomics-guided radiation therapy. 77 Wu et al. proposed a robust tumor-partitioning method using a twostage clustering procedure, 27

Potential to incorporate novel biomarkers in clinical trials
Biomarker-based trials are ongoing in oncology practice. A markerrelated trial includes the following steps: phase I attempts to validate the marker effectiveness, phase II validates the subpopulation based on certain markers responding to the given therapy, and phase III tries to confirm the clinical benefit from makers based on phase II studies. 78,79 The Biomarker-integrated Approaches of Targeted Therapy for Lung Cancer Elimination (BATTLE) trial was the first prospective, biomarker-based trial in patients with pretreated NSCLC. After profiling 11 prespecified biomarkers, patients were assigned to one of the five biomarker groups based on biopsy findings, and then initially equally randomized into one of the four treatment arms without considering their biomarker status. After this phase, patients were randomly assigned to treatment according to the Bayesian adaptive algorithm. The feature of the Bayesian adaptive algorithm is to randomize patients based on probabilities to treatments based on early outcome data. Patients are randomized based on their biomarker profiles into treatments with the highest likelihood of potential for efficacy in the BATTLE trial. 80 anaplastic lymphoma kinase. 83 Early-stage completely resected nonsquamous lung cancer patients who were eligible for the ALCHEMIST trial were assigned to different treatment arms based on their EGFR/anaplastic lymphoma kinase/PD-L1 status. In each arm, patients were randomized to receive targeted therapy or a placebo. 84 The ALCHEMIST trial using the umbrella platform has the potential to enhance efficiency and provide a precision medicine-based approach to add new substudies with novel biomarkers in a defined subset of patients. 85

Potential of radiomic biomarkers in immuno-oncology trials
Currently, patients are subject to ICIs based on their PD-L1 status in immune-oncology trials; however, the outcome study is only based on imaging of RECIST criteria. The current RECIST criteria were based on simple metrics, such as diameters of the lesion/node in the short or long axis, which do not consider the complete information carried with the imaging, such as the shape and heterogeneity of the lesion/node. One potential direction of the clinical trial is to identify patients based on a series of radiomic biomarkers and subgroup them into different categories: radiomics signatures will be utilized to define tumor aggressiveness and the level of the immune response. 86 Tentatively, the potential four categories are as follows based on different TME groups with potential implications for mechanism and therapy have been identified according to B7-H1 (PD-L1) expression and the presence of tumor-infiltrating lymphocytes in tumor biopsies. 87 The first category is hot tumor/adaptive resistance. This group has low aggressiveness or fewer hypoxic regions, but with a potential strong immune response, as manifested by more immune cell infiltration and a relatively active TME. The second category is intermediate (lazy) tumor/a situation of tolerance. This group has low aggressiveness or fewer hypoxic regions, but with weak immune responses, such as less immune cell infiltration or an immunosuppressive microenvironment.

The third category is intermediate (active) tumor/balanced situation.
The tumor has high aggressiveness and more hypoxic regions exit, yet with a strong immune response with more immune cell infiltration and an immunoactive microenvironment. The fourth category is cold tumor/immunological ignorant. The tumor has high aggressiveness, but a low immune response and has an immunosuppressive microenvironment. After patients are subgrouped based on radiomic signatures, they can enter biomarker-driven clinical trials where they would be randomized to the immune point inhibitors/placebo to elucidate their treatment response, which is used to validate the radiomic signatures.
Once the validity of the signatures is established, patients can be entered into an umbrella trial and randomized to treatments based on the status of radiomic signatures. Chemotherapy and radiation therapy are frequently used to prime tumors and harness the host's immune system to attack tumor cells. 88,89 Each category will be randomized to monotherapy with ICI, ICI + chemotherapy, ICI + whole tumor radiation therapy, ICI + partial tumor radiation therapy, and control to optimize current therapeutic methods with each biological distinct group. This is a bridge to radiomics-driven precision medicine.

Radiomics can provide biomarkers and easily predict important
clinical measures, such as therapy outcome, stage, grade, and other main biological pathways for selection of the optimal therapy regimen in lung immuno-oncology; radiomics-guided radiotherapy will improve the ablation effect and immunomodulatory effects of radiation therapy.

PERSPECTIVES
Radiomics has the potential to elucidate tumor biology and prognosis in a non-invasive, comprehensive, and efficient manner. It also has the potential to guide the selection of drug treatment and clinical planning of radiation therapy. The application of radiomics in lung immunooncology will need to be validated in large-scale prospective studies.
Integration of an expert panel of clinical practice, biology, data science, and artificial intelligence would be instrumental in the future development of lung immune oncology.

ETHICS STATEMENT
This study is not involved in any ethical issues.

CONFLICT OF INTEREST STATEMENT
The authors declare that they have read the article and there are no competing interests.