. 2023 Jun 19;7:59. doi: 10.1038/s41698-023-00413-9

Table 2.

Studies investigating the association between imaging and immune-related data.

Study	Year	Study size^a	Tumor grade	Imaging modality	ROI definition	Extracted features	Immune sampling	Database	Findings	Association with survival
Assessment of immune signatures and immune-related pathways
⁹⁷	2014	23	Grade IV	T1, T1CE, T2	Necrosis, edema, infiltrating tumor, enhancing tumor	Morphology + volumetric features	mRNA + DNA microarray analysis	Single-center retrospective study	Absence of mass effect was associated with interleukin 3 and transforming growth factor ß pathways.	n/a
⁸⁴	2014	55	Grade IV	T1, T1CE, T2-FLAIR	Enhancing tumor, necrosis, edema (including non-enhancing tumor)	Morphology + high-order features	Gene expression modules based on various cancer driver genes	TCGA + TCIA (from 4 centers in the US)	Edge blurriness (vs sharpness) of necrotic ROI was positively correlated with IL4 pathway involved in T-cell differentiation and proliferation.	Only enhancement ROI features were associated with survival. (Edge sharpness positively correlated with PFS and OS. Border regularity positively correlated with OS.)
⁸⁵	2016	91	Grade IV	T1, T1CE, T2-FLAIR	Enhancing tumor(CE), necrosis (NE), edema (ED) (including non-enhancing tumor); Tumor bulk (TB) = necrosis + enhancing tumor, Total tumor volume (TV) = edema + tumor bulk	Volumetric features	GSEA	TCGA + TCIA (from 2 centers in the US)	NE, TB, NE/CE, TB/TV were negatively associated with immune response pathways. ED/TV, CE/TB was positively associated with regulation of T-cell activation and proliferation.	NE, CE, TB significantly associated with survival. (CE was the strongest)
⁹⁸	2016	50	Grade IV	T1, T1CE, DWI	Necrotic core, enhancing active tumor, peritumoral edema	Volumetric features + ADC histogram (average mean, standard deviation, skewness, kurtosis and entropy)	n/a	TCGA + TCIA	Mean ADC was significantly negatively correlated with pathways involved in dendritic cell maturation and immune response through following genes: CD4, CD86, MHC class I and class II and MGMT gene.	n/a
⁸⁶	2017	92	Grade IV	T1, T1CE, T2, T2-FLAIR	Edema, enhancing tumor, non-enhancing tumor, necrosis	VASARI features	mRNa and miRNa expression	TCGA + TCIA	T helper cell differentiation, NK cell and B cell activation, interferon gamma response pathways were enriched in less aggressive phenotype (volume-class:T1/FLAIR ratio: hemorrhage ≤2).	Combinatorial phenotype of volume-class, hemorrhage, and T1/FLAIR- ratio significantly stratified survival. A low value for any of these 3 features indicated favorable survival characteristics.
⁹⁹	2018	155 (Training set = 91; Test set = 64)	Grade III	T1, T1CE	Enhancing tumor(CE), non-enhancing tumor	First‐order statistics, shape‐ and size‐based features, textual features	RNA microarray analysis	TCGA + CGGA + TCIA	Genes driving immune system response were significantly enriched in contrast-enhancement regions.	All of the prognostic features were textual features. Seven genes derived from the CE‐specific signature could stratify patients into two subgroups based on overall survival time.
¹⁰⁰	2018	47	Grade II and III	T2	Whole tumor (abnormal T2 hyperintensity signals)	First-order, textural, wavelet, shape- and size-based features	RNA microarray analysis	CGGA + TCGA + TCIA	High radiomic risk score was associated with immune responses (lymphocyte activation and positive regulation of immune system processes), programmed cell death and, I-kappaB and NF-kappaB signaling.	The radiomic risk score was an independent prognostic factor for PFS and provided significant stratification of PFS in both cohorts.
¹⁰¹	2021	95 (Training set = 78; Test set = 17)	Grade IV	T1, T1CE, T2, T2-FLAIR	Solid tumor core, edema; whole tumor (core + edema)	Shape, intensity, texture features	RNA sequencing	TCGA + TCIA + internal data set + external data set	Prognostic radiomics phenotypes were correlated with distinct immune pathway.	Significant association of radiomics signature with overall survival in the training subset (HR = 4.80) and validation subset (HR = 3.68). The C index achieved 0.73 in the training subset and 0.70 in the validation subset.
⁶⁹	2023	149	Grade IV	T1CE, T2	Enhancing tumor, necrosis, edema, peritumoral region (10 mm from the enhanced boundaries of the tumor)	Shape, first-order, textural features, wavelet features	CIBERSORT + ESTIMATE	TCGA + CGGA + CPTAC + TCIA	A model that combined 11 radiomic features was able to distinguish tumors with different Immune cell infiltration (ICI) scores (AUC = 0.96, accuracy = 94%). GBM with a low ICI score exhibits greater necrosis in T1CE and lower expression of the original GLCM texture feature and wavelet feature around the peritumoral regions. GBM with a high ICI score exhibits smaller necrosis in T1CE and higher expression of the original GLCM texture feature and wavelet feature around the peritumoral regions in T2 sequence.	The survival outcomes of patients could be stratified according to the ICI groups predicted by radiogenomic features
Assessment of specific immune cell subset infiltration
⁶⁴	2017	69	Grade IV	T1, T1CE, T2-FLAIR	Enhancing tumor, T2-FLAIR hyperintensity (solid tumor + infiltrating tumor + edema)	Volumetric + textural + intensity features	mRNA expression of CD3D/E/G	TCGA + TCIA	Prediction of CD3 infiltration: Training set: Accuracy = 97.1% and AUC = 0.993. Test set: Accuracy= 76.5% and AUC = 0.847. GLSZM was the best single predictor.	n/a
⁶⁶	2018	60	Grade IV	T2-FLAIR, T1CE, T1, T2, DSC perfusion MRI, DWI	Entire volume of contrast-enhancing lesions, T2 high signal intensity lesions, and necrosis (defined as a hypointense area without contrast enhancement on T1CE within the mass on the FLAIR images)	Volumetrics, mean ADC and CBV values	RNA-level analysis of 14 immune cell markers using quantitative RT-PCR	Single center retrospective study	CD68 (TAMs), CSF1R (TAMs), CD33 (myeloid-derived suppressor cell) and CD4 (helper T-cell, regulatory T-cell) levels were highly positively correlated with nCBV values based on ROIs from both FLAIR and T1CE. CD11b had a significant positive correlation with nCBV values only from T1CE. CD3e (helper T-cell, cytotoxic T-cell) and CD49d showed a significantly negative correlation with ADC. CD33 and CD123, and CD25 were negatively correlated with ADC values from FLAIR and T1CE, respectively. Tumor volumes based on FLAIR or T1CE had significant negative correlations with the expression levels of CD123, CD49d, and CD117, but no immune cell markers showed a significant correlation with tumor necrosis or necrosis ratio.	CD49d was an independent factor for PFS indicating CD49d expression levels correlated with ADC can be a candidate biomarker for predicting progression.
⁶⁸	2020	116 (Training set = 84; Test set = 32)	Grade IV	T1CE, ADC	Total tumor region	First-order statistics, gray-level run length matrix (GLRLM), gray-level co-occurrence matrix (GLCM), shape and size features	RNA sequencing	TCGA, ICGA, TCIA	For T1CE imaging data, the average accuracies of the CTL, aDC, Treg, and MDSC models were 0.72, 0.75, 0.81, and 0.88, respectively. For ADC imaging data, the average accuracies of the aforementioned models were 0.71, 0.61, 0.68, and 0.79, respectively. T1CE features yielded better distinguishability of the enrichment levels of all immune cell subsets relative to ADC features.	The developed radiomics models could reliably identify three immunophenotype groups and aid in the prediction of prognosis.
¹⁰²	2020	107 (Training set = 85; Test set = 22)	Grade II and III	T1, T1CE, T2, T2-FLAIR	Enhancing part of the tumor core, non-enhancing part of the tumor core and peritumoural edema	Intensity, volumetric, histogram-based, and textural features	Tumor Immune Estimation Resource (TIMER) based on RNA sequencing data	TCIA + TCGA	The infiltration levels of B cells, CD8+ T-cells, neutrophils and macrophages estimated by radiomics correlated with those estimated by TIMER in the testing cohort.	n/a
⁶⁷	2021	64	Grade III and IV (GRADE IV, anaplastic oligodendrogliomas, and anaplastic astrocytomas)	T1, T1CE, T2, T2-FLAIR, DSC, ADC	Enhancing tumor and edema (T1CE and FLAIR); whole tumor for rCBV and ADC maps	First-order, shape-based, textural features	Flow cytometry	Single-center retrospective study	The radiomic signatures showed the following diagnostic performance in predicting the immune phenotypes: (1) T-cell fraction (enriched vs deficient), AUC = 0.986; (2) T-cell subclass without Treg (T8 vs T4^* dominant), AUC = 0.783; (3) M2-TAM fraction (M2-TAM high vs low), AUC = 0.798. The IDH-projected radiomics signature score was significantly positively correlated with the following: M2-TAM; TAM; M2-monocyte/macrophage; and TIL. Regardless of immune phenotype, the majority of the top 10 features were from ADC maps	n/a
⁶⁵	2022	167	Grade II, III, IV	T2	Whole tumor (abnormal hyperintense signals on T2)	First-order, shape and size, textural and wavelet features	Single-cell RNA-sequencing	TCGA + TCIA + internal prospective cohort	The immune system process was significantly related to 14 radiomics features (RFs). Tumor-infiltrating macrophages showed a distinct and strong correlation with prognostic RFs.	Tumor-infiltrating macrophages were highly enriched in patients with higher RF scores and convey poor prognosis.
Assessment of checkpoint inhibitor expression
⁷²	2020	85	Grade II, III, IV	T1, T1CE, T2,T2-FLAIR	Enhancing tumor, non-enhancing tumor(NET), peri-tumoral edema (ED)	Intensity, volumetric, morphologic, histogram-based, textural and spatial features	RNA-seq + Gene functional enrichment analysis	TCGA + TCIA	Radiomic features effectively separated gliomas into two subgroups with distinct prognosis: C1 (higher survival) and C2 (lower survival). Patients with C2 radiomic subtype harbored higher CD8 + T-cells, PD1, PD-L1, and CTLA4.	The prognostic value of radiomics extracted from ED region was slightly lower compared with NET and ET region.
⁷⁰	2021	124 (Training set = 68; Test set = 56)	Grade II and III	T1, T1CE, T2, T2-FLAIR	Whole tumor based on FLAIR	n/a	mRNA microarray	TCGA + TCIA + internal cohort	Value of the ROC curve for radiomics-based prediction of IMRiskScore^b: 0.821 in the test group and 0.708 in the test group.	n/a
Assessment of prognosis and/or treatment response in immunotherapy trials
⁷⁴	2019	22	Grade IV	T1, T1CE, DWI, ADC, DSC perfusion MRI, T2-FLAIR, T2	Tumor volume (enhancing portion of the lesion on T1CE)	Volumetrics, rCBV_max, ADCmin	Flow cytometery, ELISA	Single center prospective study	Significant decrease in rADC_min was observed after 4 vaccinations only in patients with a persistent increase of natural killer cells (response effectors during immunotherapy) in peripheral blood. Also, difference in cerebral blood volume (ΔrCBV_max) distinguished TTP from PsP with a sensitivity of 67% and specificity of 75%.	Basal rADC_min > 1 significantly predicted longer progression free and overall survival.
⁷⁵	2022	162	Grade IV	T1, T1CE, T2, T2-FLAIR	Enhancing tumor volume	Tumor shape, intensity histogram, and textural features	n/a	Multi-national phase II clinical trial of durvalumab	n/a	Pretreatment radiomics model showed poor performance in predicting OS and PFS. Conversely, first post-treatment radiomics model showed a high C-index for the prediction of OS.

ROI region of interest, GBM glioblastoma multiforme, DSC Dynamic susceptibility contrast, ADC Apparent Diffusion Coefficient, T1CE T1 post-contrast, DWI diffusion-weighted imaging, TCGA The Cancer Genome Atlas, TCIA The Cancer Imaging Archive, CGGA Chinese Glioma Genome Atlas, ICGA Indian Cancer Genome Atlas, CPTAC The National Cancer Institute’s Clinical Proteomic Tumor Analysis Consortium, GSEA Gene set enrichment analysis, OS overall survival, PFS progression free survival, CBV cerebral blood volume, n/a not available.

^aThe number of patients included in the radiomic analysis.

^bIMriskScore-related mRNAs are derived from Immunophenotype-associated mRNA signatures and are associated with immune checkpoint expression and prognosis.