Table 2.
Details of included studies.
| Disease | Study [Ref] | Total Cohort # |
Validation | CD8 Evaluation |
Joint Analysis |
Imaging Modality | Radiomics Software |
Features Extracted |
Tumour Region |
Relevant Radiomic Signatures |
Modelling | End Point | ||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| R | S | SQ | # | Features | ||||||||||||
| Lung | NSCLC | Lopci et al. [47] | 55 | N | IHC | N | PET ([18F]-FDG) | NA | N | N | Y | Intratumoural | 2 | SQ: SUVmean, SUVmax | Cox regression | DFS |
| NSCLC | Castello et al. [48] | 44 | N | IHC | N | PET ([18F]-FDG) | LIFEx | Y | N | Y | Intratumoural | 5 | R: First-order SQ: SUVmax, SUVpeak, SUVmean, MTV |
Cox regression | DFS | |
| NSCLC | Mazzaschi et al. [42] | 100 | Y | IHC | Y (CD3, PD-1) |
CT | SlicerRadiomics | Y | Y | N | Intratumoural | 11 | R: First-order, GLCM, GLRLM, GLDM S: Texture, effect (parenchyma reaction), margins |
Cox regression | OS, DFS | |
| NSCLC | Mitchell et al. [49] | 59 | N | IHC | N | PET ([18F]-FDG) | NA | N | N | Y | Intratumoural | 0 | None significant | Cox regression | OS, DFS | |
| NSCLC | Zhou et al. [45] | 91 | N | IHC | Y (PD-L1) |
PET ([18F]-FDG) | NA | N | N | Y | Intratumoural | 5 | SQ: SUVmax, SUVmean, TLG | Logistic regression | Tumour immuno- phenotype |
|
| NSCLC | Min et al. [37] | 97 | Y | FACS | N | CT | PyRadiomics | Y | Y | N | Intratumoural | 4 | R: GLCM, GLDM S: Boundary type, lymphatic metastasis |
Neural network-based | High/low CD8 levels | |
| NSCLC | Zhou et al. [46] | 103 | Y | IHC | Y (PD-L1) |
PET/CT ([18F]-FDG) | LIFEx | Y | N | Y | Intratumoural | 1 | R: NGLDM | Logistic regression | Tumour immuno-phenotype | |
| Hepatobiliary | PDAC | Li et al. [54] | 184 | Y | IHC | N | CE-CT | PyRadiomics | Y | Y | N | Intratumoural | 11 | R: First-order, GLSZM S: Tumour size |
Logistic regression, XGBoost | High/low CD8 levels |
| PDAC | Bian et al. [55] | 156 | Y | IHC | N | MRI (T1W, T2W, post-contrast [AP PPP, PVP]) | PyRadiomics | Y | Y | N | Intratumoural | 14 | R: First-order, GLCM, GLRLM, GLSZM, NGTDM S: Lesion location, tumour size |
Linear regression, XGBoost | High/low CD8 levels | |
| PDAC | Bian et al. [56] | 144 | Y | IHC | N | MRI (T1W, T2W) | PyRadiomics | Y | Y | N | Intratumoural | 13 | R: First-order, GLCM, GLRLM, GLSZM | LDA classifier | High/low CD8 levels | |
| HCC | Chen et al. [43] | 207 | Y | IHC | Y (CD3) |
MRI (CE) | Analysis Kit (GE Healthcare) | Y | N | N | Intratumoural, peritumoural, combined | 70 | R: Shape, GLCM, GLRLM, GLSZM | Extra-Trees, logistic regression | Immunoscore prediction | |
| HCC | Liao et al. [57] | 142 | Y | IHC | N | CE-CT | Analysis Kit (GE Healthcare) | Y | N | N | Intratumoural | 7 | R: GLCM, GLRLM | Elastic-net | OS, DFS | |
| ICC | Zhang et al. [58] | 78 | N | IHC | N | MRI (T1W, T2W, post-contrast [AP, PVP], DW) | PyRadiomics | Y | N | N | Intratumoural | 4 | R: Shape, first-order, GLSZM | Logistic regression, Cox regression | Tumour immuno-phenotype, OS | |
| Brain | LGG | Zhang et al. [38] | 107 | Y | TIMER | N | MRI (T1W, T1CE, T2W, T2-FLAIR) | CaPTK | Y | N | Y | Multiple subregions | 3 | R: Shape, GLRLM | Cox regression | OS |
| GBM | Hsu et al. [59] | 116 | Y | RNA-seq | N | MRI (T1CE, DW) | ND | Y | N | N | Intratumoural | 15 | R: First-order, GLRLM | Logistic regression | High/low CD8 levels | |
| HGG | Kim et al. [36] | 51 | N | FACS | Y (CD4) |
MRI (T1W, T1CE, T2W, T2-FLAIR, DW, DSC) | PyRadiomics | Y | N | N | Intratumoural | 5 | R: GLCM, GLRLM, GLSZM, GLDM | sPLS-DA | OS | |
| Glioma | Chaddad et al. [40] | 151 | Y | CIBERSORT | N | MRI (T1W, T1CE, FLAIR, T2W) | MATLAB | Y | Y | N | Intratumoural | 3 | R: GLSZM | Neural network-based | High/low CD8 levels | |
| Gastrointestinal | Gastric cancer | Jiang et al. [44] | 1778 | Y | IHC | Y (CD3, CD45RO, CD66b) |
CE-CT | MATLAB | Y | N | N | Intratumoural, peritumoural | 13 | R: Shape, GLCM, GLRLM, GLSZM, NGTDM | Logistic regression, Cox regression | Immunoscore prediction, DFS, OS |
| ESCC | Wen et al. [60] | 220 | Y | IHC | N | CE-CT | IBEX | Y | N | N | Intratumoural | 8 | R: First-order, GLCM, GLRLM | Logistic regression | High/low CD8 levels | |
| Rectal cancer | Jeon et al. [61] | 113 | Y | IHC | N | MRI (T2W) | MATLAB | Y | N | N | Intratumoural | 6 | R: First-order, GLCM, GLRLM, GLSZM | Linear regression | Chemoradiotherapy-induced changes | |
| Head and neck | HNSCC | Katsoulakis et al. [52] | 160 | Y | RNA-seq | N | CE-CT | Radiomics Toolbox in CERR | Y | N | N | Intratumoural | 67 | R: First-order, GLCM, GLRLM, GLSZM, NGTDM, NGLDM | Random forest | High/low CD8 levels |
| HNSCC | Wang et al. [41] | 71 | Y | Chemokine gene expression | N | CE-CT | SlicerRadiomics | Y | N | N | Intratumoural | 8 | R: GLCM, GLSZM, GLDM, NGTDM | Logistic regression | Tumour immuno-phenotype | |
| Multiple | Multiple | Sun et al. [51] | 491 | Y | RNA-seq | N | CE-CT | LIFEx | Y | Y | N | Intratumoural, peritumoural | 8 | R: First-order, GLRLM S: Lesion location (adenopathy; head and neck), CT parameters (kVp) |
Elastic-net | Objective response, OS |
| Multiple | Ligero et al. [53] | 198 | Y | IHC | N | CE-CT | PyRadiomics | Y | Y | N | Intratumoural | 16 | R: Shape, first-order, GLCM, GLDM S: Lesion location (liver; other) |
Elastic-net, Cox regression | Objective response, OS | |
| Others | Breast cancer | Arefan et al. [39] | 73 | Y | MCP-Counter | N | MRI (DCE) | PyRadiomics | Y | N | Y | Intratumoural | 2 | R: Shape SQ: Tumour mean peak enhancement |
XGBoost | High/low CD8 levels |
| UPS | Toulmonde et al. [62] | 14 | N | IHC; RNA-seq | N | MRI (T1CE) | OleaSphere® Software | Y | N | N | Intratumoural | 9 | R: First-order, GLRLM | Cox regression | OS, MFS | |
| Melanoma | Aoude et al. [50] | 52 | N | RNA-seq; mIF; Histomorphometry | N | PET/CT ([18F]-FDG) | NA | Y | N | Y | Intratumoural | 1 | R: First-order | Cox regression | OS, PFS | |
Acronyms: AP = arterial phase; CaPTK = cancer imaging phenomics toolkit; CE = contrast-enhanced; CE-CT = contrast-enhanced CT; CERR = computational environment for radiological research; CIBERSORT = cell-type identification by estimating relative subsets of RNA transcript; DCE = dynamic contrast-enhanced; DFS = disease-free survival; DSC = dynamic susceptibility contrast-enhanced; ESCC = esophageal squamous cell carcinoma; Extra-Trees = extremely randomized tree algorithm; FACS = fluorescence-activated cell sorting; GBM = glioblastoma; HCC = hepatocellular cancer; HGG = high-grade glioma; HNSCC = head and neck squamous cell carcinoma; ICC = intrahepatic cholangiocarcinoma; IHC = immunohistochemistry; kVp = peak kilovoltage; LDA = linear discriminant analysis; LGG = lower-grade glioma; MCP-counter = microenvironment cell populations-counter; MFS = metastasis-free survival; mIF = multiplex immunofluorescence; N = no; NA = not available/applicable; ND = not declared; NSCLC = non-small cell lung cancer; OS = overall survival; PDAC = pancreatic ductal adenocarcinoma; PFS = progression-free survival; PPP = pancreatic parenchymal phase; PVP = portal venous phase; R = radiomic features; RNA-seq = RNA sequencing; S = semantic features; sPLS-DA = sparse partial least squares discriminant analysis; SQ = semi-quantitative features; SSF = spatial scaling factor; T1-FLAIR = T1-weighted fluid attenuated inversion recovery; T1CE = T1-weighted contrast-enhanced; T1W = T1-weighted; T2-FLAIR = T2-weighted fluid attenuated inversion recovery; T2W = T2-weighted; TIMER = tumour immune estimation resource; UPS = undifferentiated pleomorphic sarcoma; XGBoost = binary logistic extreme gradient boosting framework; Y = yes.