Imaging-guided precision medicine in glioblastoma patients treated with immune checkpoint modulators: research trend and future directions in the field of imaging biomarkers and artificial intelligence

Abstract

Immunotherapies that employ immune checkpoint modulators (ICMs) have emerged as an effective treatment for a variety of solid cancers, as well as a paradigm shift in the treatment of cancers. Despite this breakthrough, the median survival time of glioblastoma patients has remained at about 2 years. Therefore, the safety and anti-cancer efficacy of combination therapies that include ICMs are being actively investigated. Because of the distinct mechanisms of ICMs, which restore the immune system's anti-tumor capacity, unconventional immune-related phenomena are increasingly being reported in terms of tumor response and progression, as well as adverse events. Indeed, immunotherapy response assessments for neuro-oncology (iRANO) play a central role in guiding cancer patient management and define a "wait and see strategy" for patients treated with ICMs in monotherapy with progressive disease on MRI. This article deciphers emerging research trends to ameliorate four challenges unaddressed by the iRANO criteria: (1) patient selection, (2) identification of immune-related phenomena other than pseudoprogression (i.e., hyperprogression, the abscopal effect, immune-related adverse events), (3) response assessment in combination therapies including ICM, and (4) alternatives to MRI. To this end, our article provides a structured approach for standardized selection and reporting of imaging modalities to enable the use of precision medicine by deciphering the characteristics of the tumor and its immune environment. Emerging preclinical or clinical innovations are also discussed as future directions such as immune-specific targeting and implementation of artificial intelligence algorithms.

Keywords: Artificial Intelligence; Durvalumab; Gliblastoma; Imaging; Immunotherapy; MR; Nivolumab; PET; Pembrolizumab; Pidilizumab; RANO; Radiomics; iRANO.

Fig. 1

Imaging of actionable molecular pathways in patients with glioblastoma: the concept of supervoxels. Imaging allows non-invasive evaluation of the action of immune checkpoint modulators in patients with glioblastoma. Currently, most clinicians perform a visual and qualitative assessment. Alternatively, artificial intelligence can be trained to extract imaging biomarkers by measuring the signal in each unique voxel of a region of interest provided by each imaging technique. Ultimately, artificial intelligence can resume the information provided by multiple voxels from multiple imaging modality to provide one single quantitative probability map using supervoxels (synthetic summary of all voxels from the same volume of interest using different imaging modalities)

Fig. 2

Detection of a potential hyperprogression in a patient with glioblastoma. This case illustrates the potential risk of hyperprogression. Imaging of an 18 year old patient with a diagnosis of glioblastoma treated with anti-PD-1. MRIs were obtained at 3-month intervals (baseline, a–e; 3 months, f, g). a–e Baseline T1 post-contrast MRI prior to immunotherapy and re-gamma knife therapy demonstrating an enhancing lesion with increased perfusion. f, g MRI post-initiation of immunotherapy showing fast interval growth of the lesion, as well as a life-threatening mass effect. This case illustrates the potential life-threatening local complications of hyperprogression

Fig. 3

Multimodal image-guided management in a PD-1, PD-L1, TILs glioblastoma. This case illustrates the potential interest of pre-immunotherapy immuno-PET imaging biomarkers since the immune escaping environment (i.e., pathology was negative for PD-1, PD-L1 and, tumor infiltrating lymphocytes) explaining the insensitivity of this patient to immunotherapy was demonstrated only on the pathology post-resection at the end of immunotherapy. Existing imaging techniques demonstrated treatment insensitivity (a–h) but were not able to decipher the immune contexture for an early prediction of outcome. Imaging of a patient with recurrent glioblastoma in the left parietal lobe treated with combined immunotherapy (nivolumab) and re-gamma knife. MRIs were obtained at 3-month intervals. a Baseline T1 post-contrast MRI prior to immunotherapy and re-gamma knife therapy demonstrating a 6 × 5 mm enhancing lesion in the left parietal lobe. b MRI post-initiation of immunotherapy and pre-re-gamma knife therapy showing interval growth of the lesion. c MRI perfusion demonstrating growth and increased flow along the anterior margin of the tumor. d, e PET/CT demonstrating continued growth and increased FDG activity along the margin of the lesion. f Subsequent MRI demonstrating significant growth, increased peripheral nodular enhancement, and central necrosis. g Post-contrast MRI post-resection showing mild non-specific enhancement around the resection margin. h Follow-up MRI 7 months after resection demonstrating progression of disease

Fig. 4

Multimodal image-guided management using artificial intelligence in glioblastoma. This case illustrates the potential interest of imaging biomarkers extracted using artificial intelligence. Imaging of a patient with glioblastoma. a Baseline T1 post-contrast MRI prior to therapy demonstrating an enhancing lesion. b Baseline 18F-Dopa PET showing an increased amino acid uptake outside of the enhancing lesion on MRI. c Analysis of the MRI by artificial intelligence demonstrating areas with high heterogeneity (red) and low heterogeneity in normal healthy brain tissue (blue). This map is a parametric map of local entropy computed using the baseline T1 post-contrast MRI. The only limit in the analysis of the local heterogeneity/entropy is that contours/edge/interface are always heterogeneous. d Fused image of the Baseline 18F-Dopa PET (b) and of the parametric map of local entropy (c)