Impact of a Radiologic and Immunologic Signature on Cerebral Efficacy of Treatments in Lung Cancers With Brain Metastasis

Non-small cell lung cancer (NSCLC) is a major public health concern and is frequently diagnosed at an advanced or metastatic stage. Brain metastases occur frequently in patients with advanced NSCLC and are associated with poor prognosis and limited therapeutic response.

The tumor immune microenvironment plays an important role in tumor progression and treatment resistance. Several immune cell populations, including regulatory T lymphocytes, tumor-associated macrophages, and myeloid-derived suppressor cells, contribute to an immunosuppressive environment. The proportion and distribution of these immune populations may provide prognostic information and may influence response to systemic treatments, including immunotherapy. However, characterization of the immune microenvironment generally requires tissue sampling and may not provide a dynamic assessment of treatment response.

Radiomics is a quantitative imaging approach allowing automated extraction of multiple imaging features from medical images, including tumor shape, texture, signal characteristics, and characteristics of the surrounding tissue. Radiomic features can be integrated into machine-learning approaches to identify imaging patterns associated with clinically relevant biological characteristics. In NSCLC, previous studies have reported associations between radiomic signatures and molecular characteristics, treatment response, progression-free survival, and overall survival. However, no radiomic signature has been prospectively validated for routine clinical use in NSCLC.

The combination of radiomic and immune or transcriptomic data may provide complementary information regarding the biological characteristics of brain metastases. The present study will investigate whether MRI-derived radiomic signatures are associated with immune profiles obtained from bulk RNA sequencing of resected brain metastasis tissue.

The study will include 38 patients with metastatic NSCLC and at least one brain metastasis at diagnosis who were managed at Bordeaux University Hospital between 2017 and 2025. The study is retrospective and monocentric and will use clinical, imaging, and biological data collected as part of routine care.

Brain MRI examinations performed at diagnosis will be collected and pseudonymized. Tumor lesions and their surrounding tissues will undergo segmentation and quantitative radiomic analysis. Available tumor tissue samples from resected brain metastases will be identified, and RNA extraction followed by bulk RNA sequencing will be performed. Bioinformatic analyses will be used to characterize transcriptomic profiles and to investigate associations between radiomic and transcriptomic signatures.

Clinical and treatment data will be collected from medical records, including demographic characteristics, performance status, date of diagnosis, treatment initiation, histological characteristics, PD-L1 status, molecular status, characteristics of brain metastases, corticosteroid treatment, radiotherapy, cerebral progression, death, and last follow-up. Cerebral progression-free survival will be assessed using follow-up data.

The primary analysis will evaluate the correlation between the MRI-based radiomic signature and the transcriptomic signature obtained from bulk RNA sequencing. Secondary analyses will evaluate associations between radiomic, transcriptomic, and combined radio-transcriptomic signatures and cerebral progression-free survival.

The study will also explore whether distinct radiomic patterns are associated with different immune profiles, including relatively immune-low and immune-high profiles. An immune-low profile is expected to be associated with lower levels of cytotoxic T lymphocytes and antitumor M1 macrophages and higher levels of immunosuppressive populations such as myeloid-derived suppressor cells, M2 macrophages, and regulatory T lymphocytes.

The study may provide preliminary evidence supporting the use of MRI-derived radiomic signatures as a non-invasive surrogate of the immune characteristics of brain metastases. Such an approach could potentially allow longitudinal assessment of brain metastases using routine MRI and contribute to the development of predictive algorithms for treatment response. Further prospective validation would be required before clinical implementation.