This study demonstrates that using radiomics, it is possible to predict programmed death ligand 1 (PD-L1) expression status in non-small cell lung cancer (NSCLC) patients based on 2-deoxy-2-[18 F]-fluoroglucose-positron-emission tomography/computed tomography (18 F-FDG-PET/CT) images. This non-invasive approach can help identify patients who may benefit from immune checkpoint inhibitor (ICI) therapy [1]. By utilizing this method, we aim to support clinical decision-making and improve patient outcomes.

Malignancies are marked by significant intra-tumoral heterogeneity; this presents challenges for the clinical management of cancer patients. Spatial heterogeneity refers to the variation in the genetic and phenotypic characteristics of tumor cells within different regions of the same tumor or even between primary and metastatic sites. Temporal heterogeneity involves changes in these characteristics over time, often due to treatment and other environmental factors [2,3,4].

Medical imaging plays a crucial role in understanding the biology of malignancies. Firstly, unlike tissue sampling for histology, it is a non-invasive, in vivo diagnostic tool. Secondly, medical imaging enables the assessment of tumor system dynamics over an extended period by repeating the examinations. Thirdly, it provides insights at different depth levels to characterize various cancer processes, such as molecular, functional, and anatomic imaging, providing an opportunity to examine intra-tumoral heterogeneity. Lastly, unlike biopsies, which provide limited information from a single tissue sample, tomographic techniques such as PET, CT, and magnetic resonance imaging (MRI) offer a comprehensive, three-dimensional picture of the entire cancerous region, including nearby structures and potential metastases [2].

Radiomics exploits these advantages by using medical images to extract quantitative information that is difficult for human eyes to recognize or quantify. The primary hypothesis of radiomics is that the extracted features should correlate with gene expression pattern for therapeutically valuable markers, providing the basis for digital biopsy [4,5,6].

According to the latest global cancer statistics, the most prevalent form of malignancy worldwide is lung cancer, which is responsible for the highest number of cancer-related deaths [7]. Early and accurate diagnosis and staging are crucial to improve patient outcomes. In over 50% of cases, curative surgery is not an option [8]. Tissue sampling is essential as the future treatment depends on the histological type of the tumor. NSCLC represents ∼ 85% of diagnosed lung cancer cases and includes various subtypes, such as adenocarcinoma (ACC) and squamous cell carcinoma (SCC).

The immune system has several checkpoints to prevent the destruction of normal cells, but malignancies can manipulate these to establish tolerance. Among the checkpoints, programmed death receptor 1 (PD-1) has received particular interest recently due to its high potential in treating many cancers. Figure 1. shows the interaction between PD-1 and PD-L1 and PD-L2 proteins. Currently, there are antibodies approved for targeting PD-1 and PD-L1. According to the latest version of the National Comprehensive Cancer Network (NCCN) Clinical Practice Guidelines in Oncology, depending on the PD-L1 expression status, they can be prescribed as first-line or subsequent treatment with or without chemoradiotherapy or as part of neoadjuvant and adjuvant therapy for squamous and non-squamous NSCLC. They may be suitable for inoperative or potentially operative cases, and compared to platinum-based chemotherapy, they have fewer, usually mild side effects [9]. Currently, PD-L1 expression is the most reliable biomarker for identifying patients who can benefit from PD-1 or PD-L1 inhibitor therapy. Immunohistochemistry (IHC) remains the gold standard for assessing PD-L1 expression status. Still, there has been particular interest recently in quantifying PD-1/PD-L1 expression and its heterogeneity in vivo with 89Zr-labeled anti-PD-1/PD-L1 monoclonal antibodies [1, 8, 10]. In lung cancers, PD-L1 expression status is routinely assessed only in ACC and SCC.

Pictorial illustration showing the mechanism of action by ICIs in cancer immunotherapy. TCR: T-cell receptor, MHC: major histocompatibility complex

Histological sampling is often time-consuming, complicated, or unsuccessful (e.g., due to extensive emphysema or the undesirable location of the tumor). Complications can also occur, such as pneumothorax, bleeding, and infections. In some institutions, placing patients after the procedure presents a challenge. The performance status of some patients may not allow sampling at all. Additionally, the pathological analysis of the sample is usually a time-consuming process. Also, PD-L1 expression status can be altered by the heterogeneity of the tumor and by treatment [11]. This highlights the need for new non-invasive techniques, such as radiomic analysis of imaging data. Radiomics has potential clinical value in situations where a biopsy is not possible. Additionally, it can be a viable alternative for cases requiring repeated monitoring or reassessment of treatment effects, as it offers more detailed information about tissue changes than traditional methods. In instances where multiple biopsies would be difficult or impractical, radiomic analysis can help determine whether tissue sampling is necessary. Furthermore, radiomics lays the groundwork for further research, which may uncover stronger connections to biological markers.

Per the NCCN recommendations, all patients with suspected NSCLC should undergo a PET/CT scan, which provides information on the metabolic activity of the cancer processes. 18 F-FDG is the most commonly used PET radiotracer in the management of NSCLC. 18 F-FDG uptake varies among different lung histological subtypes: squamous cell carcinoma typically shows higher 18 F-FDG avidity than adenocarcinoma [12].

18 F-FDG-PET/CT can efficiently differentiate between benign and malignant lesions. A maximum standardized uptake value (SUVmax) exceeding the mean mediastinal blood pool activity or 2.5 is suggestive of malignancy. Due to limited spatial resolution, 18 F-FDG-PET/CT is not recommended for lesions smaller than 8 mm. 18 F-FDG-PET/CT should also be avoided in the characterization of subsolid opacities to prevent false-negative (e.g., atypical adenomatous hyperplasia, carcinoma in situ, minimally invasive carcinoma) and false-positive results (e.g., inflammation or infection) [13]. Small lesions smaller than 15 mm in the lower lobes may lead to inaccuracies due to artifacts caused by breathing motion; thus, respiratory gating is strongly recommended in these scenarios [14].

One of the most promising advancements in oncologic imaging is the field of radiomics. Radiomics involves extracting features from medical images and analysing the acquired complex data using advanced machine learning and statistical methods. Radiomic features may not only characterize tumor heterogeneity but also provide additional complementary information to other available sources: demographics, pathology, blood biomarkers, and genomics, which could play essential roles in diagnostics, patient-specific treatment monitoring, and patient prognosis [4, 5].