Use of scores for predicting the risk of cancer progression in a subject suffering from lung cancer

Abstract

The present invention relates to the use of a Lesion Progression Probability Score (LPPS) for predicting the risk of progression of a tumor lesion in a subject suffering from lung cancer when treated by immunotherapy, wherein the tumor lesion comprises a core tumoral region and a peritumoral region referred to as a ring and the LPPS of each lesion comprises 6 specific radiomics features. The invention also relates to the use of a patient progression probability score (PPPS) for predicting the risk of disease progression in a subject suffering from lung cancer when treated by immunotherapy, wherein the PPPS combines the LPPS values of all analyzed tumor lesions of the subject with a volume equal to or higher than a minimum volume Vmin. The invention also relates to therapeutic uses of immunotherapy in subjects with lung cancer with a low or moderate predicted risk of disease progression using the PPPS.

Description

[0001] USE OF SCORES FOR PREDICTING THE RISK OF CANCER PROGRESSION IN A SUBJECT SUFFERING FROM LUNG CANCER

[0002] TECHNICAL FIELD OF THE INVENTION

[0003] The present invention is in the field of cancer treatment. It relates to the use of a Lesion Progression Probability Score (LPPS) for predicting the risk of progression of a tumor lesion in a subject suffering from lung cancer when treated by immunotherapy, wherein the tumor lesion comprises a core tumoral region and a peritumoral region referred to as a ring and the LPPS of each lesion comprises 6 specific radiomics features. The invention also relates to the use of a patient progression probability score (PPPS) for predicting the risk of disease progression in a subject suffering from lung cancer when treated by immunotherapy, wherein the PPPS combines the LPPS values of all analyzed tumor lesions of the subject with a volume equal to or higher than a minimum volume Vmin. The invention also relates to therapeutic uses of immunotherapy in subjects with lung cancer with a low or moderate predicted risk of disease progression using the PPPS.

[0004] BACKGROUND ART

[0005] Immune-checkpoint inhibitors (ICI) have been a major breakthrough in cancer treatment over the last decade (Robert C. A decade of immune-checkpoint inhibitors in cancer therapy. Nat Commun 2020; 1 1 : 3801 ; Chen DS, Mellman I. Oncology Meets Immunology: The Cancer-Immunity Cycle. Immunity 2013; 39: 1 -10). Although ICI may reveal long-term remission in the metastatic setting, only 20-40% of patients respond to these treatments (). Identifying patients who are most likely to respond is particularly challenging as the conventional companion biomarkers have only moderate performance. For instance, a meta-analysis of 8135 patients estimated the AUC (area under the curve) of PD-L1 (programmed cell death ligand 1 ) immunohistochemistry (IHC) to be 0.65 while the AUC of the tumor mutational burden (TMB) was estimated to be 0.69. Thus, development of effective biomarkers to identify the patients that are most likely to respond to these treatments is of utmost importance (Bellesoeur A, Torossian N, Amigorena S, Romano E. Advances in theranostic biomarkers for tumor immunotherapy. Current Opinion in Chemical Biology 2020; 56: 79-90). There is growing interest in quantitative imaging biomarkers and radiomics. Indeed, in contrast to traditional biopsy methods, medical imaging is non-invasive and is widely utilized in routine clinical practice. Additionally, it provides a comprehensive assessment of the entire tumor burden, enabling the analysis of all lesions (Henry T, Sun R, Lerousseau M, et al. Investigation of radiomics based intra-patient inter-tumor heterogeneity and the impact of tumor subsampling strategies. Sci Rep 2022; 12: 17244). Radiomics converts images into high-throughput quantitative data in order to develop imaging biomarkers, and has demonstrated promising outcomes in predicting responses to immunotherapy (Sun R, Limkin EJ, Vakalopoulou M, et al. A radiomics approach to assess tumour-infiltrating CD8 cells and response to anti-PD-1 or anti-PD-L1 immunotherapy: an imaging biomarker, retrospective multicohort study. The Lancet Oncology 2018; 19: 1180-91 ; Trebeschi S, Drago SG, Birkbak NJ, et al. Predicting Response to Cancer Immunotherapy using Non- invasive Radiomic Biomarkers. Ann Oncol 2019; published online March 21. D0l:10.1093 / annonc / mdz108 ; Tunali I, Gray JE, Qi J, et al. Novel clinical and radiomic predictors of rapid disease progression phenotypes among lung cancer patients treated with immunotherapy: An early report. Lung Cancer 2019; 129: 75-9; Yang Y, Yang J, Shen L, et al. A multi -omics-based serial deep learning approach to predict clinical outcomes of single-agent anti-PD-1 / PD-L1 immunotherapy in advanced stage non-small-cell lung cancer. Am J Transl Res 2021 ; 13: 743-56). Using RNA sequencing data, we previously developed a radiomics signature that quantifies tumor-infiltrating CD8 T cells (CD8- Rscore). This signature has shown to be associated with overall survival and treatment response in a cohort of advanced pan-solid tumor cancer patients undergoing anti- PD1 / PD-L1 immunotherapy (Sun R, Limkin EJ, Vakalopoulou M, et al. A radiomics approach to assess tumour-infiltrating CD8 cells and response to anti-PD-1 or anti-PD-L1 immunotherapy: an imaging biomarker, retrospective multicohort study. The Lancet Oncology 2018; 19: 1180-91 ) or immunoradiotherapy combinations (Sun R, Sundahl N, Hecht M, et al. Radiomics to predict outcomes and abscopal response of patients with cancer treated with immunotherapy combined with radiotherapy using a validated signature of CD8 cells. J Immunother Cancer 2020; 8: e001429 ; Korpics MC, Polley M-Y, Bhave SR, et al. A Validated T Cell Radiomics Score Is Associated With Clinical Outcomes Following Multisite SBRT and Pembrolizumab. International Journal of Radiation Oncology, Biology, Physics 2020; 0. D0l:10.1016 / j.ijrobp.2020.06.026; Korpics MC, Onderdonk BE, Dadey RE, et al. Partial tumor irradiation plus pembrolizumab in treating large advanced solid tumor metastases. J Clin Invest 2023; 133. DOI:10.1172 / JCI162260). The CD8-Rscore has also been shown to be interesting in advanced melanoma patients (Sun R, Lerousseau M, Briend-Diop J, et al. Radiomics to evaluate interlesion heterogeneity and to predict lesion response and patient outcomes using a validated signature of CD8 cells in advanced melanoma patients treated with anti-PD1 immunotherapy. J Immunother Cancer 2022; 10: e004867).

[0006] Despite the general ability of the CD8-Rscore to any cancer, each cancer is known to have specificities, and a signature that has been designed based on patients suffering from various solid cancers may not be optimal for a specific cancer.

[0007] Lung cancer is one of the solid cancers with the lowest survival rate at 5 years after diagnosis, and tools permitting to predict response to a particular treatment, such as immunotherapy, with the highest possible performance are thus particularly needed.

[0008] SUMMARY OF THE INVENTION

[0009] In the context of the present invention, the inventors surprisingly found that the previously established CD8-Rscore could predict lesion progression when treated by immunotherapy (durvalumab), but with a limited performance (AUC=0.59 for all lesions, P-value<0.0001 ), even when focusing on liver lesions, which were the best predicted (AUC=0.66, P-value=0.0002).

[0010] However, when adding further radiomics features to the CD8-Rscore, it was unexpectedly possible to obtain a Lesion Progression Probability Score (LPPS) able to predict lesion progression with a significantly increased performance (AUC of 0.82 and 0.78 in the train and test cohorts respectively, P-values<0.0001 ).

[0011] In a first aspect, the present invention thus relates to the use of a Lesion Progression Probability Score (LPPS) for predicting the risk of progression of a tumor lesion in a subject suffering from lung cancer when treated by immunotherapy, wherein the tumor lesion comprises a core tumoral region and a peritumoral region referred to as a ring and the LPPS of each lesion comprises the 6 following radiomics features:

[0012] (i) the standard deviation of the intensity distribution in the ring (abbreviated as ring_std Value);

[0013] (ii) the Low Grey Level Run Emphasis (LGLRE) feature of the Grey Level Run Length Matrix (GLRLM) of the ring (abbreviated as ring_GLRLM_LGRE);

[0014] (iii) the minimum intensity in the ring (abbreviated as ring_minValue);

[0015] (iv) the Intensity Histogram Kurtosis feature of the core tumoral region (abbreviated as tu m_H I STO_K u rtosi s) ;

[0016] (v) the Correlation feature of the Grey Level Cooccurrence Matrix (GLCM) of the ring (abbreviated as ring_GLCM_Correlation); and (vi) a complex radiomics feature combining the values of:

[0017] • the minimum intensity in the core tumoral region (abbreviated as tum_MinValue),

[0018] • two or more (e.g. 2, 3 or 4) Gray-level Run Length Matrix (GLRLM) features selected from the group consisting of: o GLRLM short-run high gray-level emphasis of the core tumoral region (abbreviated as tum_GLRLM_SRHGE), o GLRLM short-run low gray-level emphasis of the ring (abbreviated as ring_GLRLM_SRLGE), o GLRLM low gray-level run emphasis of the ring (abbreviated as ring_GLRLM_LGRE), and o GLRLM long-run low gray-level emphasis of the ring (abbreviated as ring_GLRLM_LRLGE),

[0019] • the imaging-acquisition feature kVp,

[0020] • optionally, the location feature lymph node metastasis (abbreviated as VOI_Adenopathy), wherein the value of VOI_Adenopathy is 1 when the analyzed lesion is a lymph node metastasis and 0 when the analyzed lesion is at any other location, and

[0021] • optionally, the location feature head and neck lesion (abbreviated as VOI_head_and_neck), wherein VOI_head_and_neck is 1 when the analyzed lesion is a tumor lesion in the pharynx, larynx, oral cavity, or salivary gland and 0 when the analyzed lesion is at any other location.

[0022] Preferably, the complex radiomics feature (vi) is the lung cancer CD8-Rscore, a complex radiomics feature consisting of a linear regression of:

[0023] • the minimum intensity in the core tumoral region (abbreviated as tum_MinValue),

[0024] • four Gray-level Run Length Matrix (GLRLM) features consisting of: o GLRLM short-run high gray-level emphasis of the core tumoral region (abbreviated as tum_GLRLM_SRHGE), o GLRLM short-run low gray-level emphasis of the ring (abbreviated as ring_GLRLM_SRLGE), o GLRLM low gray-level run emphasis of the ring (abbreviated as ring_GLRLM_LGRE), and o GLRLM long-run low gray-level emphasis of the ring (abbreviated as ring_GLRLM_LRLGE), the imaging-acquisition feature kVp, and two location features: o lymph node metastasis (abbreviated as VOI_Adenopathy), and o head and neck lesion (VOI_head_and_neck), wherein the value of VOI_Adenopathy is 1 when the analyzed lesion is a lymph node metastasis and 0 when the analyzed lesion is at any other location and VOI_head_and_neck is 1 when the analyzed lesion is a tumor lesion in the pharynx, larynx, oral cavity, or salivary gland and 0 when the analyzed lesion is at any other location.

[0025] The present invention also relates to the use of a patient progression probability score (PPPS) for predicting the risk of disease progression in a subject suffering from lung cancer when treated by immunotherapy, wherein the PPPS combines the LPPS values of all analyzed tumor lesions of the subject with a volume equal to or higher than a minimum volume Vmin , wherein the LPPS of each tumor lesion is as defined herein.

[0026] The present invention also relates to a method for predicting the risk of progression of a tumor lesion in a subject suffering from lung cancer when treated by immunotherapy, wherein the tumor lesion comprises a core tumoral region and a peritumoral region referred to as a ring from previously obtained image data obtained by using non-invasive imagining technologies, comprising: a) calculating a Lesion Progression Probability Score (LPPS) as defined herein based on the previously obtained image data obtained by using non-invasive imagining technologies; and b) predicting the risk of progression of the tumor lesion in the subject when treated by immunotherapy based on the LPPS calculated in step a).

[0027] The present invention also relates to a method for predicting the risk of disease progression in a subject suffering from lung cancer treated by immunotherapy, from previously obtained image data obtained by using non-invasive imagining technologies, comprising: a) calculating a Lesion Progression Probability Score (LPPS) as defined herein for all analyzed tumor lesions with a volume equal to or higher than a minimum volume Vmin of the subject based on the previously obtained image data obtained by using non-invasive imagining technologies; b) calculating a patient progression probability score (PPPS) combining the LPPS values of all tumor lesions of the subject; and c) predicting the risk of disease progression when treated by immunotherapy based on the PPPS value calculated in step b).

[0028] The present invention also relates to an immunotherapy for treating lung cancer in a subject, wherein the immunotherapy is for administration to subjects for whom a low or moderate risk of disease progression when treated by immunotherapy has been predicted using the PPPS as defined herein or the method for predicting the risk of disease progression according to the invention.

[0029] DESCRIPTION OF THE FIGURES

[0030] Figure 1 : Study design.

[0031] Figure 2: Flowchart of patients.

[0032] Figure 3. Performance of the radiomics score of CD8 T-cells in predicting lesion progression. A: Tumor diameter changes at the first evaluation according to the Baseline CD8 Radiomics score. B. Boxplot of CD8 radiomic score level according to lesion progression. C Receiver operating characteristic curve of association between Lesion Progression and CD8 Radiomics score. D-E: AUC of the lesion progression prediction of the CD8 Radiomics score level, according to the volume of the lesion (D) and the location of the lesion (E). F-G: Kaplan-Meier curves of overall survival (F) and progression free survival (G) according to the minimal value of the CD8-Rscore of lesions in the whole cohort.

[0033] FIGURE 4. Dispersion metrics of the CD8 radiomics score and overall survival (A), or progression free survival (B).

[0034] FIGURE 5: importance of the feature retained in the fined-tuned XGBoost model, including the CD8-radiomics score.

[0035] Figure 6. Performance of the Lesion Progression Probability score (LPPS). A-B: AUC (A) and boxplot (B) of the lesion progression prediction according to the LPPS in the training set. C-D: AUC (C) and boxplot (D) of the lesion progression prediction according to the LPPS in the test set. E: AUC of the lesion progression prediction of the LPPS, according to the location of the lesion.

[0036] Figure 7. Kaplan-Meier curves of overall survival (A, B ) and progression free survival (C,D) according to the Patient progression probability score, a fined-tuned model including the CD8-Radiomics score, in the Training Set (A,C) and in the Test set (B,D). DETAILED DESCRIPTION OF THE INVENTION

[0037] In the context of the present invention, the inventors surprisingly found that the previously established CD8-Rscore could predict lesion progression when treated by immunotherapy (durvalumab), but with a limited performance (AUC=0.59 for all lesions, P-value<0.0001 ), even when focusing on liver lesions, which were the best predicted (AUC=0.66, P-value=0.0002).

[0038] However, when adding further radiomics features to the CD8-Rscore, it was unexpectedly possible to obtain a Lesion Progression Probability Score (LPPS) able to predict lesion progression with a significantly increased performance (AUC of 0.82 and 0.78 in the train and test cohorts respectively, P-values<0.0001 ).

[0039] Definitions

[0040] The scores used in the present invention combines radiomics and non-radiomics features. “Radiomics” consists in the analysis of quantitative data extracted from standard medical imaging to generate imaging biomarkers. The process of radiomics consists of discrete steps: image acquisition and segmentation, feature extraction, and statistical learning. A considerable number of features can be used to assess the characteristics of a target zone.

[0041] As used herein a “radiomics feature” refers to a quantitative variable extracted from images of a volume of interest in a consistent manner.

[0042] A “volume of interest” (abbreviated as “VOI”) refers to a delimitated zone in an image, in which the radiomics features are extracted. In the context of cancer, “tumor lesions” (i.e. areas comprising cancer cells) are analyzed. Tumor lesions can generally be divided into two distinct areas that may both be used as VOI for extraction of radiomics features: a “core tumoral region” (also referred to as “intratumoral region”, i.e. the area within the tumor, which may be abbreviated as “turn”) and a “peritumoral region” (corresponding to a peripheral ring of 1 to 3 mm, preferably 2 mm, on both sides of the tumor boundaries, which may be abbreviated as “ring”). Preferred VOIs in the context of the present invention are the core tumoral region and the peritumoral region of any tumor lesion of at least 1 mL of the subject of interest.

[0043] Radiomics features may be classified into several categories. There are quantitatively extracted descriptors of size, shape, and other radiologic terminologies which characterize the tumor surface. Here, the categories defined by the Image biomarker standardisation initiative’s (IBSI) are used. Such categories are used by various softwares, including the LIFEx software (http: / / www.lifexsoft.org, in particular as described in https: / / www.lifexsoft.org / index.php / resources / texture / radiomic-features). The definition of any radiomics feature mentioned herein can be found in IBSI website, and also in LIFEX website or in the LIFEx documentation relating to Features for LIFEx version 7.6. n, as updated on 2024 / 01 / 31 , available from https: / / www.lifexsoft.org / index.php / resources / documentation), which also refers to the Image biomarker standardisation initiative’s (IBSI) Reference manual version 1.0 of December 2019).

[0044] First-order features are used to study the distribution of voxel values without considering spatial relationships. They include features of the following categories: INTENSITY-BASED (also known as “conventional indices”), INTENSITY-HISTOGRAM (also known as “Discretized Indices”), and DISCRETIZED_HISTO-First order features.

[0045] INTENSITY-BASED features are conventional indices derived from the intensity distribution in the volume of interest. Intensity I may be expressed in HU (“Housfiled Units”, a dimensionless unit universally used in CT scanning to express CT numbers in a standardized and convenient form) for CT (“computed tomography”) or SUV (“standardized uptake value”, also known as “dose uptake ratio” (DUR), a mathematically derived ratio of tissue radioactivity concentration at a point in time C(T) at a specific region of interest (ROI) and the injected dose of radioactivity per kilogram of the patient's body weight) for PET (“Positron Emission Tomography”) . There are a number of INTENSITY-BASED features, including:

[0046] • The minimum intensity value in the Volume of Interest, which may be abbreviated as “VOI_MinValue”, wherein “VOI” is the name or abbreviation of the volume of interest, and

[0047] • The standard deviation of the intensity distribution in the Volume of Interest, which may be abbreviated as “VOI_stdValue”, wherein “VOI” is the name or abbreviation of the volume of interest.

[0048] There are a number of INTENSITY-HISTOGRAM features, including:

[0049] • The Intensity Histogram Kurtosis feature, which may be abbreviated as “VOI_HISTO_Kurtosis”, wherein “VOI” is the name or abbreviation of the volume of interest,

[0050] • The Intensity Histogram Skewness feature, which may be abbreviated as “VOI _HISTO_Skewness”, wherein “VOI” is the name or abbreviation of the volume of interest, • The Intensity Histogram Entropy_log10 feature, which may be abbreviated as “VOI_HISTO_Entropy_log10”, wherein “VOI” is the name or abbreviation of the volume of interest,

[0051] Second-order (also referred to as “texture”) features characterize spatial relationships between voxels. There are a number of second order or texture features, which may be derived from various matrices:

[0052] • the co-occurrence matrix (GLCM), which takes into account the arrangements of pairs of voxels to calculate textural indices;

[0053] • the gray-level run length matrix (GLRLM), which gives the size of homogeneous runs for each grey level;

[0054] • the gray-level size zone matrix (GLZLM), which provides information on the size of homogeneous zones for each grey-level in 3 dimensions and is also named the Grey Level Size Zone Matrix (GLSZM);

[0055] • the neighborhood gray-level different matrix (NGLDM), which corresponds to the difference of grey-level between one voxel and its 26 neighbours in 3 dimensions (8 in 2D).

[0056] There are a number of GLCM features, including:

[0057] • the Correlation feature, which may be abbreviated as “VOI_GLCM_Correlation”, wherein “VOI” is the name or abbreviation of the volume of interest,

[0058] • the Contrast feature, which may be abbreviated as “VOI_GLCM_Contrast”, wherein “VOI” is the name or abbreviation of the volume of interest.

[0059] There are a number of GLRLM features, including:

[0060] • the Low Grey Level Run Emphasis (LGLRE) feature, which may be abbreviated as “VOI_GLRLM_LGRE”, wherein “VOI” is the name or abbreviation of the volume of interest,

[0061] • the Short-Run High Gray-Level Emphasis (SRHGE) feature, which may be abbreviated as “VOI_GLRLM_SRHGE”, wherein “VOI” is the name or abbreviation of the volume of interest,

[0062] • the Short-Run Low Gray-Level Emphasis (SRLGE) feature, which may be abbreviated as “VOI_GLRLM_SRLGE”, wherein “VOI” is the name or abbreviation of the volume of interest,

[0063] • the low gray-level run emphasis (LGRE) feature, which may be abbreviated as “VOI_GLRLM_LGRE”, wherein “VOI” is the name or abbreviation of the volume of interest, • the long-run low gray-level emphasis (LRLGE) feature, which may be abbreviated as “VOI_GLRLM_LRLGE”, wherein “VOI” is the name or abbreviation of the volume of interest

[0064] There are a number of GLZLM features, including:

[0065] • the Small Zone Emphasis (SZE) feature, which may be abbreviated as “VOI GLZLM_SZE”, wherein “VOI” is the name or abbreviation of the volume of interest,

[0066] • the Large Zone Emphasis (LZE) feature, which may be abbreviated as “VOI_GLZLM_LZE”, wherein “VOI” is the name or abbreviation of the volume of interest,

[0067] • the Small Zone Low Grey Level Emphasis (SZLGLE) feature, which may be abbreviated as “VOI_GLZLM_SZLGLE”, wherein “VOI” is the name or abbreviation of the volume of interest.

[0068] There are a number of NGLDM features, including:

[0069] • The Contrast feature, which may be abbreviated as “VOI_NGLDM_Contrast”, wherein “VOI” is the name or abbreviation of the volume of interest,

[0070] • The Busyness feature, which may be abbreviated as “VOI_NGLDM_Busyness”, wherein “VOI” is the name or abbreviation of the volume of interest

[0071] Shape features describe the shape / morphology of volumes of interest delimited on image data. There are a number of SHAPE features, including:

[0072] • The Compacity feature when three dimensional images of the volume of interest are available, which may be abbreviated as “VOI_SHAPE_Compacity.onlyFor3DROI”, wherein “VOI” is the name or abbreviation of the volume of interest,

[0073] • The Sphericity feature when three dimensional images of the volume of interest are available, which may be abbreviated as “VOI_SHAPE_Sphericity.onlyFor3DROI”, wherein “VOI” is the name or abbreviation of the volume of interest,

[0074] • The Surface feature in mm2when three dimensional images of the tumor lesion are available, which may be abbreviated as “VOI_SHAPE_Surface.mm2..onlyFor3DROI”, wherein “VOI” is the name or abbreviation of the volume of interest,

[0075] • The Volume feature in mL, which may be abbreviated as “VOI_SHAPE_Volume.mL”, wherein “VOI” is the name or abbreviation of the volume of interest. Filter grids such as Gabor and Fourier may be used both in the pre-processing step and for extracting spatial or spatio-temporal features. A limitation is that some extracted values are dependent on the VOIs contoured.

[0076] Furthermore, complex radiomics features combining first order, second order and shape radiomics features (and optionally non-radiomics features) can also be computed and used as a new radiomics feature. An example of such a complex radiomics feature is the “CD8-Rscore”, which is a linear regression of:

[0077] • the minimum intensity in the core tumoral region (abbreviated as tum_MinValue),

[0078] • four Gray-level Run Length Matrix (GLRLM) features consisting of: o GLRLM short-run high gray-level emphasis of the core tumoral region (abbreviated as tum_GLRLM_SRHGE), o GLRLM short-run low gray-level emphasis of the ring (abbreviated as ring_GLRLM_SRLGE), o GLRLM low gray-level run emphasis of the ring (abbreviated as ring_GLRLM_LGRE), and o GLRLM long-run low gray-level emphasis of the ring (abbreviated as ring_GLRLM_LRLGE),

[0079] • the imaging-acquisition feature kVp, and

[0080] • two location features: o lymph node metastasis (abbreviated as VOI_Adenopathy), and o head and neck lesion (VOI_head_and_neck), wherein the value of VOI_Adenopathy is 1 when the analyzed lesion is a lymph node metastasis and 0 when the analyzed tumor lesion is at any other location, and VOI_head_and_neck is 1 when the analyzed lesion is a tumor lesion in the pharynx, larynx, oral cavity, or salivary gland, and 0 when theanalyzed tumor lesion is at any other location.

[0081] Preferably, the CD8-Rscore may use the regression coefficients presented in Table 1 below for the above-defined features:

[0082] Table 1. Preferred regression coefficients used in CD8-Rscore linear regression of the indicated features.

[0083] In the CD8-Rscore linear regression of the above features, the correlation between the CD8-Rscore and the risk of lesion progression depends on the features and their associated coefficients as defined in Table 1 above, not on the value of the intercept. Indeed, if the intercept is changed, only the value of the cut-off separating low and high risk of lesion progression needs to be adapted accordingly. As a result, any intercept value may be used in the CD8-Rscore linear regression of the above features, in particular when the above preferred coefficients are used.

[0084] The extracted features can be global (one value for the whole ROI), or local (a value per image patch) when inhomogeneous patterns are present in the image, where dimensionality significantly increases if simple concatenation of local descriptors is carried out. For this, more advanced frameworks explore compact statistical representations based on coding structures / dictionaries. A more detailed review on texture analysis methods focusing on microscopy images of cells or tissues can be found in [Cataldo SD, Ficarra E. Mining textural knowledge in biological images: applications, methods and trends. Comput Struct Biotechnol J 2016]. Radiomic features in PET scan or CT scan imagery cannot be accessed or extracted by pencil and paper or acquired by the human mind. Radiomic features present in CT scan imagery are sub- visual features that are not visible to the human eye.

[0085] The scores used in the present invention also comprise non-radiomics features, including:

[0086] • the peak kilovoltage (abbreviated as “kVp”), which is an acquisition feature corresponding to the peak potential applied to the x-ray tube, which accelerates electrons from the cathode to the anode in radiography or scanner, including computed tomography (CT) scan and positron emission tomography (PET) scan.

[0087] • location features: o “VOI_Adenopathy”, which value is 1 when the analyzed lesion is a lymph node metastasis and 0 when the analyzed tumor lesion is at any other location, and o “VOI_head and neck”, which value is 1 when the analyzed lesion is a tumor lesion in the pharynx, larynx, oral cavity or salivary gland and 0 when the analyzed tumor lesion is at any other location.

[0088] “Computed tomography” (abbreviated as “CT”) refers to a computerized x-ray imaging procedure in which a narrow beam of x-rays is aimed at a patient and quickly rotated around the body, producing signals that are processed by the machine’s computer to generate cross-sectional images, or “slices.” These slices are called tomographic images and can give a clinician more detailed information than conventional x-rays. Once a number of successive slices are collected by the machine’s computer, they can be digitally “stacked” together to form a three-dimensional (3D) image of the patient that allows for easier identification of basic structures as well as possible tumors or abnormalities.

[0089] “Positron emission tomography” (abbreviated as “PET”) refers to a procedure in which a small amount of radioactive glucose (sugar) is injected into a vein, and a scanner is used to make detailed, computerized pictures of areas inside the body where the glucose is taken up. Because cancer cells often take up more glucose than normal cells, the pictures can be used to find cancer cells in the body. Use of a Lesion Progression Probability Score (LPPS) for predicting the risk of progression of a tumor lesion in a subject suffering from lung cancer when treated by immunotherapy

[0090] The present invention first relates to the use of a Lesion Progression Probability Score (LPPS) for predicting the risk of progression of a tumor lesion in a subject suffering from lung cancer when treated by immunotherapy, wherein the tumor lesion comprises a core tumoral region and a peritumoral region referred to as a ring and the LPPS of each lesion comprises the 6 following radiomics features:

[0091] (i) the standard deviation of the intensity distribution in the ring (abbreviated as ring_std Value);

[0092] (ii) the Low Grey Level Run Emphasis (LGLRE) feature of the Grey Level Run Length Matrix (GLRLM) of the ring (abbreviated as ring_GLRLM_LGRE);

[0093] (iii) the minimum intensity in the ring (abbreviated as ring_minValue);

[0094] (iv) the Intensity Histogram Kurtosis feature of the core tumoral region (abbreviated as tu m_H I STO_K u rtosi s) ;

[0095] (v) the Correlation feature of the Grey Level Cooccurrence Matrix (GLCM) of the ring (abbreviated as ring_GLCM_Correlation); and

[0096] (vi) a complex radiomics feature combining the values of:

[0097] • the minimum intensity in the core tumoral region (abbreviated as tum_MinValue),

[0098] • two or more (e.g. 2, 3 or 4) Gray-level Run Length Matrix (GLRLM) features selected from the group consisting of: o GLRLM short-run high gray-level emphasis of the core tumoral region (abbreviated as tum_GLRLM_SRHGE), o GLRLM short-run low gray-level emphasis of the ring (abbreviated as ring_GLRLM_SRLGE), o GLRLM low gray-level run emphasis of the ring (abbreviated as ring_GLRLM_LGRE), and o GLRLM long-run low gray-level emphasis of the ring (abbreviated as ring_GLRLM_LRLGE),

[0099] • the imaging-acquisition feature kVp,

[0100] • optionally, the location feature lymph node metastasis (abbreviated as VOI_Adenopathy), wherein the value of VOI_Adenopathy is 1 when the analyzed lesion is a lymph node metastasis and 0 when the analyzed lesion is at any other location, and • optionally, the location feature head and neck lesion (abbreviated as VOI_head_and_neck), wherein VOI_head_and_neck is 1 when the analyzed lesion is a tumor lesion in the pharynx, larynx, oral cavity, or salivary gland and 0 when the analyzed lesion is at any other location.

[0101] In a preferred embodiment, the two or more GLRLM features included in the complex radiomics feature (vi) comprise tum_GLRLM_SRHGE and ring_GLRLM_LRLGE. When only two GLRLM features are included in the complex radiomics feature (vi), they thus preferably consist of ring_GLRLM_LRLGE and tum_GLRLM_SRHGE. When 3 GLRLM features are included in the complex radiomics feature (vi), they may be selected from:

[0102] • GLRLM features tum_GLRLM_SRHGE, ring_GLRLM_LRLGE, and ring_GLRLM_SRLGE,

[0103] • GLRLM features tum_GLRLM_SRHGE, ring_GLRLM_LRLGE, and ring_GLRLM_LGRE,

[0104] • GLRLM features tum_GLRLM_SRHGE, ring_GLRLM_SRLGE, and ring_GLRLM_LGRE, or

[0105] • ring_GLRLM_LRLGE, ring_GLRLM_SRLGE, and ring_GLRLM_LGRE.

[0106] Preferably, the complex radiomics feature (vi) combines the values of all four GLRLM features tum_GLRLM_SRHGE, ring_GLRLM_LRLGE, ring_GLRLM_SRLGE, and ring_GLRLM_LGRE.

[0107] The location features VOI_Adenopathy and VOI_head_and_neck is optional as it is only useful when the LPPS is used for predicting the risk of progression of a lymph node metastasis tumor lesion (for VOI_Adenopathy) or of a tumor lesion located in the pharynx, larynx, oral cavity, or salivary gland (for VOI_head_and_neck).

[0108] When the LPPS is developed for a population of cancer patients without lymph node metastasis tumor lesions, then VOI_Adenopathy does not need to be included in the complex radiomics feature (vi). However, most lung cancer patients populations include patients with lymph node metastasis tumor lesions, and VOI_Adenopathy is thus preferably included in the complex radiomics feature (vi).

[0109] In contrast, many lung cancer patients do not have tumor lesions located in the pharynx, larynx, oral cavity, or salivary gland, so that VOI_head_and_neck can easily be absent from the complex radiomics feature (vi), although is may still be included.

[0110] The features included in the complex radiomics feature (vi) may be combined using different types of algorithms, such as machine learning algorithms, Linear regression, Logistic regression, Decision tree, SVM algorithms, Naive Bayes algorithms, KNN algorithms, K-means algorithms, Random forest algorithms, Dimensionality reduction algorithms, boosting algorithms (in particular gradient boosting algorithms), deep learning algorithms and neural networks. Some machine learning algorithms are preferred, including but not limited to linear regression, logistic regression, boosting algorithms, random forest algorithms and deep learning algorithms. In a preferred embodiment, the complex radiomics feature (vi) comprises and preferably consists of a linear regression of the features included in it, but other types of algorithms may also be used.

[0111] Most preferably, the complex radiomics feature (vi) is the lung cancer CD8-Rscore, a complex radiomics feature consisting of a linear regression of:

[0112] • the minimum intensity in the core tumoral region (abbreviated as tum_MinValue),

[0113] • four Gray-level Run Length Matrix (GLRLM) features consisting of: o GLRLM short-run high gray-level emphasis of the core tumoral region (abbreviated as tum_GLRLM_SRHGE), o GLRLM short-run low gray-level emphasis of the ring (abbreviated as ring_GLRLM_SRLGE), o GLRLM low gray-level run emphasis of the ring (abbreviated as ring_GLRLM_LGRE), and o GLRLM long-run low gray-level emphasis of the ring (abbreviated as ring_GLRLM_LRLGE),

[0114] • the imaging-acquisition feature kVp, and

[0115] • two location features: o lymph node metastasis (abbreviated as VOI_Adenopathy), and o head and neck lesion (VOI_head_and_neck), wherein the value of VOI_Adenopathy is 1 when the analyzed lesion is a lymph node metastasis and 0 when the analyzed lesion is at any other location and VOI_head_and_neckis 1 when the analyzed lesion is a tumor lesion in the pharynx, larynx, oral cavity, or salivary gland, and 0 when the analyzed lesion is at any other location.

[0116] In the linear regression, coefficients associated with each of the features are preferably in the ranges presented in Table 2 below:

[0117] Analyzed image data

[0118] Any image data obtained by using scanners can be used in the method of the invention, including image data obtained by using Computed tomography (CT) scan or PET (Positron Emission Tomography) scan.

[0119] Indeed, a positron emission tomography / computed tomography (PET / CT) scan always comprises a computed tomography (CT) component, and the radiomic signature developed by the inventors from conventional diagnostic computed tomography (CT) images is also applicable to the CT component of a positron emission tomography / computed tomography (PET / CT) scan. Both types of CT imaging rely on the same physical principles and hardware, producing anatomically detailed X-ray-based tomographic images. Although variations in acquisition parameters (such as radiation dose, slice thickness, or reconstruction algorithms) may occur between diagnostic CT and PET / CT acquisitions, such variations can be handled using standard radiomic harmonization or imaging preprocessing techniques. Consequently, radiomic features extracted from PET-associated CT images are compatible with those derived from conventional CT, allowing for direct application of the radiomic signature independently of PET data or image fusion.

[0120] Preferably, image data obtained by using a CT scan is used.

[0121] Further optional features

[0122] The above-described radiomics features are the most important features of the LPPS for predicting the risk of progression of a tumor lesion in a subject suffering from lung cancer when treated by immunotherapy (see Figure 5). However, for further fine-tuning of the prediction, the LPPS may preferably further comprise one or more of the following radiomics and non-radiomics features:

[0123] (a) radiomics features:

[0124] (vii) Intensity Histogram Kurtosis of the ring (abbreviated as ring_HISTO_Kurtosis);

[0125] (viii ) Compacity of the ring when three dimensional images of the tumor lesion are available (abbreviated as ring_SHAPE_Compacity.onlyFor3DROI);

[0126] (x) Intensity Histogram Skewness of the ring (abbreviated as ri ng_H ISTO_Skewness) ;

[0127] (xi) Small Zone Emphasis of the Grey-Level Zone Length Matrix (abbreviated as ring_GLZLM_SZE);

[0128] (xii) the minimum intensity in the core tumoral region (abbreviated as tum_minValue);

[0129] (xiii) Contrast of the Neighborhood Grey-Level Difference Matrix of the core tumoral region (abbreviated as tum_NGLDM_Contrast);

[0130] (xiv) Busyness of the Neighborhood Grey-Level Difference Matrix of the core tumoral region (abbreviated as tum_NGLDM_Busyness);

[0131] (xv) Sphericity of the core tumoral region when three dimensional images of the tumor lesion are available (abbreviated as tum_SHAPE_Sphericity.onlyFor3DROI);

[0132] (xvi) Large Zone Emphasis of the Grey-Level Zone Length Matrix of the ring (abbreviated as ring_GLZLM_LZE);

[0133] (xvii) Small Zone Low Grey Level Emphasis of the Grey-Level Zone Length Matrix of the ring (abbreviated as ring_GLZLM_SZLGLE);

[0134] (xviii) Intensity Histogram Skewness of the core tumoral region (abbreviated as tu m_H I ST O_Skewness ) ;

[0135] (xix) Surface in mm2of the core tumoral region when three dimensional images of the tumor lesion are available (abbreviated as tum_SHAPE_Surface.mm2..onlyFor3DROI);

[0136] (xx) Contrast of the Grey Level Cooccurrence Matrix of the core tumoral region (abbreviated as tum_GLCM_Contrast) ;

[0137] (xxi) Contrast of the Grey Level Cooccurrence Matrix of the ring (abbreviated as ring_GLCM_Contrast) ;

[0138] (xxii) Correlation of the Grey Level Cooccurrence Matrix of the core tumoral region (abbreviated as tum_GLCM_Correlation) ;

[0139] (xxiii) Intensity Histogram Entropy_log10 of the core tumoral region (abbreviated as tum_HISTO_Entropy_log10); and (xxiv) Volume in mL of the core tumoral region (abbreviated as tum_SHAPE_Volume.mL); and

[0140] (b) non-radiomics feature :

[0141] (ix) the peak kilovoltage (abbreviated as kVp).

[0142] All of features (vii ) to (xxiv) have been shown to have some importance for predicting the risk of progression of a tumor lesion in a subject suffering from lung cancer when treated by immunotherapy (see Figure 5). In a preferred embodiment, the LPPS thus further comprises all of features (vii) to (xxiv) above. However, the order of importance of features (vii) to (xxiv) is decreasing when the feature number increases. As a result, the LPPS may not comprise all of features (vii) to (xxiv) above, but only some of them. In this case, the LPPS may preferably comprise the most important features. For instance, the LPPS may comprise feature (vii), features (vii) and (viii), features (vii) to (ix), features (vii) to (x), features (vii) to (xi), features (vii) to (xii), features (vii) to (xiii), features (vii) to (xiv), features (vii) to (xv), features (vii) to (xvi), features (vii) to (xvii), features (vii) to (xviii), features (vii) to (xix), features (vii) to (xx), features (vii) to (xxi), features (vii) to (xxii), or features (vii) to (xxiii). However, the LPPS may also comprise any other combination of features (vii) to (xxiv).

[0143] LPPS score

[0144] Preferably, the LPPS is a model returning for each analyzed tumor lesion an LPPS value depending on the values of each of the features included in the LPPS (see above), wherein the LPPS value is positively or negatively correlated to the risk of progression of the tumor lesion when treated by immunotherapy, and wherein the model has been obtained by training an algorithm on the features to be included in the LPPS extracted from image data of a reference population of lung cancer patients.

[0145] Based on features to be included in the LPPS extracted from image data of a reference population of lung cancer patients, an algorithm can be trained to obtain an LPPS model returning for each newly analyzed tumor lesion an LPPS value that is positively or negatively correlated to the risk of progression of the tumor lesion when treated by immunotherapy.

[0146] Any algorithm may be trained based on features to be included in the LPPS extracted from image data of a reference population of lung cancer patients, using machine learning algorithms, such as Linear regression, Logistic regression, Decision tree, SVM algorithms, Naive Bayes algorithms, KNN algorithms, K-means algorithms, Random forest algorithms, Dimensionality reduction algorithms, boosting algorithms (in particular gradient boosting algorithms), deep learning algorithms and neural networks.

[0147] However, some machine learning algorithms are preferred, including but not limited to boosting algorithms, linear regression, logistic regression, random forest algorithms and deep learning algorithms.

[0148] A boosting algorithm may preferably be used.

[0149] A boosting algorithm is an ensemble learning algorithm that combines the predictive power of several base estimators to improve robustness. In short, it combines multiple weak or average predictors to build a strong predictor. While boosting is not algorithmically constrained, most boosting algorithms consist of iteratively learning weak classifiers with respect to a distribution and adding them to a final strong classifier. When they are added, they are weighted in a way that is related to the weak learners' accuracy. In adaptive boosting algorithms, which are preferred, after a weak learner is added, the data weights are readjusted, known as "re-weighting". Misclassified input data gain a higher weight and examples that are classified correctly lose weight. Thus, future weak learners focus more on the examples that previous weak learners misclassified.

[0150] Among boosting algorithms, a gradient boosting algorithm may preferably be used. Gradient boosting is a machine learning technique based on boosting in a functional space, where the target is pseudo-residuals rather than the typical residuals used in traditional boosting. It gives a prediction model in the form of an ensemble of weak prediction models, i.e., models that make very few assumptions about the data, which are typically simple decision trees (it may then be referred to as “gradient-boosted trees”). Non-limiting examples of gradient boosting algorithm include XGBoost, Gradient Boosting, AdaBoost (adaptive boosting), CatBoost, LightGBM (Light Gradient Boosting Machine).

[0151] Reference population

[0152] The reference population of lung cancer patients is preferably representative of the general population of lung cancer patients. In particular, it preferably comprises:

[0153] • at least one and preferably several lung cancer patients for which image data (preferably CT scans) of tumor lesions that were later found progressive are available; and at least one and preferably several lung cancer patients for which image data (preferably CT scans) of tumor lesions that were later found non-progressive are available.

[0154] More preferably, the reference population of lung cancer patients comprises a reasonable number (such as at least 10%, preferably at least 30%) of lung cancer patients for which image data of later-found progressive tumor lesions are available and a reasonable number (such as at least 10%, preferably at least 30%) of lung cancer patients for which image data of later-found non-progressive tumor lesions are available.

[0155] Determinins the risk of lesion progression when treated by immunotherapy

[0156] Once the LPPS has been calculated, the risk of progression of the tumor lesion when treated by immunotherapy is determined based on the correlation found by the inventors between the LPPS and the risk of progression of the tumor lesion when treated by immunotherapy.

[0157] The LPPS value may be used for predicting the risk of progression of the tumor lesion in several different manners.

[0158] Preferably, the LPPS value is used to determine the risk of progression of the tumor lesion as follows: a) the risk of progression of the tumor lesion is determined by comparison of the LPPS value to a cut-off value, wherein the cut-off value has preferably been previously determined based on the reference population of lung cancer patients by the algorithm used for obtaining the model; or b) the LPPS value directly represents the probability that the analyzed tumor lesion progresses or that the analyzed tumor lesion does not progress.

[0159] Comparison with a cut-off value

[0160] In an embodiment, the risk of progression of the tumor lesion when treated by immunotherapy may be determined by comparison with a cut-off value.

[0161] When the LPPS increases with the risk of progression (i.e. when the LPPS is positively correlated to the risk of progression of the tumor lesion when treated by immunotherapy), the risk of progression of the tumor lesion when treated by immunotherapy may preferably be considered:

[0162] • high if the LPPS is higher than the cut-off value, and

[0163] • low if the LPPS is lower or equal to the cut-off value. When the LPPS decreases with the risk of progression (i.e. when the LPPS is negatively correlated to the risk of progression of the tumor lesion when treated by immunotherapy), the risk of progression of the tumor lesion when treated by immunotherapy may preferably be considered:

[0164] • high if the LPPS is lower than the cut-off value, and

[0165] • low if the LPPS is higher or equal to the cut-off value.

[0166] In this embodiment, to be able to reliably predict the risk of progression of the tumor lesion, the cut-off value, to which the LPPS is compared for determining the risk, has preferably been previously determined based on the reference population of lung cancer patients (which may be referred to as a “training set”) by the algorithm used for obtaining the model.

[0167] Different methods may be used to determine the cut-off value based on the reference population. In an embodiment, the median LPPS value of lesions analyzed in the reference population may be used. Several alternative methods may be used in order to optimize the cut-off value for better discrimination between progressive and non-progressive tumor lesions (see Sook Young Woo, Seonwoo Kim. Determination of cutoff values for biomarkers in clinical studies. Precision and Future Medicine 2020;4(1 ):2-8. https: / / doi.org / 10.23838 / pfm.2019.00135), including: the minimum P -value approach (the cut-off value that corresponds to the most significant difference in the prognosis of an outcome between the two groups and provides the minimum P-value among all potential cut-offs is selected), without or preferably with a well-known adjustment method to correct P-values selected from Bonferroni method, Miller and Siegmund method, Lausen and Schumacher methods 1 and 2, Hothorn and Lausen method, Contal and O’Quigley method, Woo et al method (see Sook Young Woo, Seonwoo Kim. Determination of cutoff values for biomarkers in clinical studies. Precision and Future Medicine 2020;4(1 ):2-8. https: / / doi.org / 10.23838 / pfm.2019.00135). Software packages for the Miller and Siegmund method, the Lausen and Schumacher methods 1 and 2, and the Hothorn and Lausen method are available in the R package (R Foundation for Statistical Computing, Vienna, Austria) “maxstat”. Also, package “survMisc” in the R can be used for the Contal and O’Quigley method. Woo et al. method was implemented in a SAS macro (SAS Institute Inc., Cary, NC, USA), and code is available upon request. the Youden index method (Youden WJ. Index for rating diagnostic tests. Cancer 1950; 3(1 ): 32-35), the concordance probability criterion (Liu X. Classification accuracy and cut point selection. Stat. Med. 2012; 31 (23): 2676- 2686) and the point closest- to- (0,1 ) corner in the ROC plane approach (Perkins NJ, Schisterman EF. The inconsistency of "optimal" cutpoints obtained using two criteria based on the receiver operating characteristic curve. Am. J. Epidemiol. 2006; 163(7): 670-675). of lesion non¬

[0168] In another embodiment, no cut-off value is used but the LPPS directly represents the probability that the analyzed tumor lesion progresses or that the analyzed tumor lesion does not progress.

[0169] This embodiment is preferred, in particular when the LPPS directly represents the probability that the analyzed tumor lesion progresses or that the analyzed tumor lesion does not progress, more preferably when the LPPS directly represents the probability that the analyzed tumor lesion progresses.

[0170] Luns cancer

[0171] The LPPS may be used for predicting the risk of progression of a tumor lesion in a subject suffering from any lung cancer when treated by immunotherapy.

[0172] However, it is particularly applicable for predicting the risk of progression of a tumor lesion in a subject suffering from non-small cells lung cancer (NSCLC), when treated by immunotherapy.

[0173] Immunotherapy

[0174] The LPPS may be used for predicting the risk of progression of a tumor lesion in a subject suffering from lung cancer (in particular NSCLC) when treated by any immunotherapy.

[0175] However, it is particularly applicable for predicting the risk of progression of a tumor lesion in a subject suffering from lung cancer (in particular NSCLC) when treated by a direct or indirect T cell-based immunotherapy.

[0176] T cell-based immunotherapies may be classified into several categories, including (Lee HM. Strategies for Manipulating T Cells in Cancer Immunotherapy. Biomol Ther (Seoul). 2022 Jul 1 ;30(4):299-308):

[0177] • Direct T cell-based immunotherapies are based on adoptive T cell transfer, which utilizes processed or modified T cells in different way such as tumor-infiltrating lymphocytes, chimeric antigen receptor (CAR) T cells, and engineered T cell receptor (TCR) T cells.

[0178] • Indirect T cell-based immunotherapies mainly rely on T cell-targeting antibodies, including: o immune checkpoint inhibitors (ICI ) , and o bispecific T cell engagers.

[0179] All T cell-based immunotherapy are based on boosting natural or genetically engineered T cells to improve their ability to kill of cancer cells. As a result, it may be expected that the LPPS may be used for predicting the risk of progression of a tumor lesion in a subject suffering from lung cancer (in particular NSCLC) when treated by any direct or indirect T cell-based immunotherapy.

[0180] However, the LPPS may particularly be used for predicting the risk of progression of a tumor lesion in a subject suffering from lung cancer (in particular NSCLC) when treated by an indirect T cell-based immunotherapy selected from immune checkpoint inhibitors (ICI) and bispecific T cell engagers. Even more preferably, the LPPS may be used for predicting the risk of progression of a tumor lesion in a subject suffering from lung cancer (in particular NSCLC) when treated by an immune checkpoint inhibitor (ICI).

[0181] Several molecules are known to deliver negative signals to activated T cells to regulate the magnitude of immune response, thereby acting as an “immune checkpoint”. Immune checkpoints help keep immune responses from being too strong but sometimes can keep T cells from killing cancer cells, as cancer cells sometimes find ways to use these checkpoints to avoid being attacked by the immune system. In particular, cancer cells often express ligands (e.g. PD-L1 ) of immune checkpoints (e.g. PD-1 ) at their surface. Upon binding of the ligand on cancer cells to the immune checkpoint on T cells, a negative signal is sent to the T cells, thus preventing cancer cell killing. “Immune checkpoint inhibitors” (abbreviated as “ICI”, also referred to as “immune checkpoint blockers” abbreviated as “ICB”) prevent the binding of the cancer cell ligand to the immune checkpoint at the surface of T cells. As no negative signal is sent to the T cells, they may kill cancer cells. ICI are generally antibodies or antigen-binding fragments thereof specifically binding to an immune checkpoint.

[0182] Immune checkpoints commonly found on T cells, their corresponding ligands commonly found on cancer cells or in the tumor microenvironment, and associated ICI are listed in Table 3 below:

[0183] Table 3. immune checkpoints, their ligands and associated immune checkpoint inhibitors.

[0184] The LPPS may especially be used for predicting the risk of progression of a tumor lesion in a subject suffering from lung cancer (in particular NSCLC) when treated by an immune checkpoint inhibitor (ICI) selected from anti-PD-1 , anti-PD-L1 and anti-PD-L2 antibodies, more preferably selected from anti-PD-1 and anti-PD-L1 antibodies, even more preferably selected from anti-PD-1 antibodies These antibodies may particularly be selected from those disclosed in Table 3 above. Use of a patient progression probability score (PPPS) for predicting the risk of disease progression in a subject suffering from lung cancer when treated by immunotherapy

[0185] The present invention also relates to the use of a patient progression probability score (PPPS) for predicting the risk of disease progression in a subject suffering from lung cancer when treated by immunotherapy, wherein the PPPS combines the LPPS values of all analyzed tumor lesions of the subject with a volume equal to or higher than a minimum volume Vmin (preferably at least 1 mL), wherein the LPPS of each tumor lesion is as defined above. Patient progression probability score (PPPS)

[0186] The PPPS combines the LPPS values of all analyzed tumor lesions of the subject with a volume equal to or higher than a minimum volume Vmin (preferably at least 1 mL) in order to provide information regarding disease progression.

[0187] In an embodiment, when the risk of progression of each analyzed tumor lesion of the subject with a volume equal to or higher than a minimum volume Vmin is determined by comparison of the LPPS value to a cut-off value and each analyzed tumor lesion of the subject with a volume equal to or higher than a minimum volume Vmin is classified as progressive or non-progressive, the PPPS may return the information whether at least one of the analyzed tumor lesions of the subject with a volume equal to or higher than a minimum volume Vmin is classified as progressive, or whether none of the analyzed tumor lesions of the subject with a volume equal to or higher than a minimum volume Vmin is classified as progressive (or whether all the analyzed tumor lesions of the subject with a volume equal to or higher than a minimum volume Vmin are classified as non-progressive). However, in a preferred embodiment, the LPPS directly represents the probability that the analyzed tumor lesion progresses or that the analyzed tumor lesion does not progress (preferably, the LPPS directly represents the probability that the analyzed tumor lesion progresses) and the PPPS is equal to:

[0188] PPPS = 1 - n"=i(l - LPPSO (Formula I) wherein n is the number of analyzed tumor lesions with a volume equal to or higher than a minimum volume Vmin (preferably at least 1 mL) of the subject and for each tumor lesion i, LPPSi is the LPPS of lesion i.

[0189] In this preferred embodiment, the PPPS value directly represents the probability that the disease progresses or that the disease does not progress.

[0190] Determining the risk of disease progression when treated by immunotherapy

[0191] Once the PPPS has been determined, the risk of disease progression when treated by immunotherapy is determined based on the correlation found by the inventors between the PPPS and the risk of disease progression when treated by immunotherapy.

[0192] Here also, the PPPS may be used for predicting the risk of progression of the tumor lesion in several different manners.

[0193] In particular, the following methods may be used: a) when the risk of progression of each analyzed tumor lesion of the subject with a volume equal to or higher than a minimum volume Vmin is determined by TJ comparison of the LPPS value to a cut-off value, each analyzed tumor lesion of the subject with a volume equal to or higher than a minimum volume Vmin is classified as progressive or non-progressive. The PPPS may then return the information whether at least one of the analyzed tumor lesion of the subject with a volume equal to or higher than a minimum volume Vmin is classified as progressive, or whether none of the analyzed tumor lesions of the subject with a volume equal to or higher than a minimum volume Vmin is classified as progressive (or whether all the analyzed tumor lesions of the subject with a volume equal to or higher than a minimum volume Vmin are classified as non-progressive). As this is a binary information, one possibility may be represented by a first value and the other possibility by another value, for instance 1 when at least one of the analyzed tumor lesion of the subject with a volume equal to or higher than a minimum volume Vmin is classified as progressive and 0 when none of the analyzed tumor lesion of the subject with a volume equal to or higher than a minimum volume Vmin is classified as progressive.

[0194] Then, the disease may be classified as:

[0195] • progressive if at least one of the analyzed tumor lesions of the subject with a volume equal to or higher than a minimum volume Vmin is classified as progressive (or not all the analyzed tumor lesion of the subject with a volume equal to or higher than a minimum volume Vmin are classified as non-progressive), or

[0196] • non-progressive if none of the analyzed tumor lesions of the subject with a volume equal to or higher than a minimum volume Vmin is classified as progressive (or all the analyzed tumor lesions of the subject with a volume equal to or higher than a minimum volume Vmin are classified as non- progressive). b) when the LPPS value directly represents the probability that the analyzed tumor lesion progresses or that the analyzed tumor lesion does not progress (preferably, the LPPS value directly represents the probability that the analyzed tumor lesion progresses), the PPPS may return a value directly representing the probability that the disease progresses or that the disease does not progress, and the risk of disease progression may either be left as a probability, or classified in two or more risk classes by comparing the PPPS value (i.e. the probability value) to one or more cut-off values.

[0197] In a preferred embodiment: • the LPPS value directly represents the probability that the analyzed tumor lesion progresses or that the analyzed tumor lesion does not progress (preferably, the LPPS value directly represents the probability that the analyzed tumor lesion progresses) and the PPPS returns a value directly representing the probability that the disease progresses or that the disease does not progress (preferably, the PPPS is of formula I disclosed above), and

[0198] • the risk of disease progression is classified in two or more risk classes by comparing the PPPS value to one or more cut-off values.

[0199] In an embodiment, the risk may be determined low or high, based on comparison of the PPPS value to one cut-off value.

[0200] Preferably, the risk is considered low, moderate or high, based on comparison of the PPPS value to two cut-off values.

[0201] In a more preferred embodiment, the LPPS value directly represents the probability that the analyzed tumor lesion progresses, the PPPS is of formula I disclosed above and returns a value directly representing the probability that the disease progresses, and the risk of disease progression when treated by immunotherapy is:

[0202] • High if the PPPS value is higher than 0.95,

[0203] • Moderate if the PPPS value is higher than 0.5 and lower or equal to 0.95; and

[0204] • Low if the PPPS value is lower or equal to 0.5.

[0205] Lung cancer

[0206] The PPPS may be used for predicting the risk of disease progression in a subject suffering from any lung cancer when treated by immunotherapy.

[0207] However, it is particularly applicable for predicting the risk of disease progression in a subject suffering from non-small cells lung cancer (NSCLC), when treated by immunotherapy.

[0208] Immunotherapy

[0209] The PPPS may be used for predicting the risk of disease progression in a subject suffering from lung cancer (in particular NSCLC) when treated by any immunotherapy.

[0210] However, preferred embodiments disclosed above in sub-section “Immunotherapy” of the section directed to the use of LPPS for predicting the risk of progression of a tumor lesion in a subject suffering from lung cancer when treated by any immunotherapy are also preferred embodiments when using the PPPS for predicting the risk of disease progression in a subject suffering from lung cancer (in particular NSCLC).

[0211] Method for predicting the risk of progression of a tumor lesion in a subject suffering from lung cancer when treated by immunotherapy

[0212] The present invention also relates to a method for predicting the risk of progression of a tumor lesion in a subject suffering from lung cancer when treated by immunotherapy, wherein the tumor lesion comprises a core tumoral region and a peritumoral region referred to as a ring from previously obtained image data obtained by using non-invasive imagining technologies, comprising: a) calculating a Lesion Progression Probability Score (LPPS) as defined in the section entitled “Use of a Lesion Progression Probability Score (LPPS) for predicting the risk of progression of a tumor lesion in a subject suffering from lung cancer when treated by immunotherapy” above based on the previously obtained image data obtained by using non-invasive imagining technologies; and b) predicting the risk of progression of the tumor lesion in the subject when treated by immunotherapy based on the LPPS calculated in step a).

[0213] In this method, the risk of progression of the tumor lesion may be determined using any method disclosed in the sub-section entitled “Determining the risk of lesion progression when treated by immunotherapy” of the above section entitled “Use of a Lesion Progression Probability Score (LPPS) for predicting the risk of progression of a tumor lesion in a subject suffering from lung cancer when treated by immunotherapy”.

[0214] Method for predicting the risk of disease progression in a subject suffering from lung cancer treated by immunotherapy

[0215] The present invention also relates to a method for predicting the risk of disease progression in a subject suffering from lung cancer treated by immunotherapy, from previously obtained image data obtained by using non-invasive imagining technologies, comprising: a) calculating a Lesion Progression Probability Score (LPPS) as defined in the section entitled “Use of a Lesion Progression Probability Score (LPPS) for predicting the risk of progression of a tumor lesion in a subject suffering from lung cancer when treated by immunotherapy” above for all analyzed tumor lesions with a volume equal to or higher than a minimum volume Vmin of the subject based on the previously obtained image data obtained by using non-invasive imagining technologies; b) calculating a patient progression probability score (PPPS) combining the LPPS values of all tumor lesions of the subject; and c) predicting the risk of disease progression when treated by immunotherapy based on the PPPS value calculated in step b).

[0216] In a preferred embodiment, the LPPS value of each analyzed lesion directly represents the probability that the analyzed tumor lesion progresses or that the analyzed tumor lesion does not progress and the PPPS is of formula I:

[0217] PPPS = 1 - nr=i(l - LPPSt (Formula I) wherein n is the number of tumor lesions of the subject and for each tumor lesion i and LPPSi is the LPPS of lesion i.

[0218] In this embodiment, the PPPS value directly represents the probability that the disease progresses or that the disease does not progress.

[0219] In this embodiment, it is further preferred that the LPPS value directly represents the probability that the analyzed tumor lesion progresses, the PPPS is of formula I, the PPPS value directly represents the probability that the disease progresses, and the risk of disease progression when treated by immunotherapy is preferably:

[0220] • High if the PPPS is higher than 0.95,

[0221] • Moderate if the PPPS is higher than 0.5 and lower or equal to 0.95; and

[0222] • Low if the PPPS is lower or equal to 0.5.

[0223] Therapeutic uses of immunotherapy for treating lung cancer in subject with a low or moderate risk of disease progression when treated by immunotherapy

[0224] The present invention also relates to an immunotherapy for treating lung cancer in a subject, wherein the immunotherapy is for administration to subjects for whom a low or moderate risk of disease progression when treated by immunotherapy has been predicted using the PPPS as defined the above section entitled “Use of a patient progression probability score (PPPS) for predicting the risk of disease progression in a subject suffering from lung cancer when treated by immunotherapy” or the method for predicting the risk of disease progression in a subject suffering from lung cancer treated by immunotherapy according to the invention.

[0225] The present invention also relates to the use of an immunotherapy for the manufacture of a drug for treating lung cancer in a subject, wherein the immunotherapy is for administration to subjects for whom a low or moderate risk of disease progression when treated by immunotherapy has been predicted using the PPPS as defined the above section entitled “Use of a patient progression probability score (PPPS) for predicting the risk of disease progression in a subject suffering from lung cancer when treated by immunotherapy” or the method for predicting the risk of disease progression in a subject suffering from lung cancer treated by immunotherapy according to the invention.

[0226] The present invention also relates to the use of an immunotherapy for treating lung cancer in a subject, wherein the immunotherapy is for administration to subjects for whom a low or moderate risk of disease progression when treated by immunotherapy has been predicted using the PPPS as defined the above section entitled “Use of a patient progression probability score (PPPS) for predicting the risk of disease progression in a subject suffering from lung cancer when treated by immunotherapy” or the method for predicting the risk of disease progression in a subject suffering from lung cancer treated by immunotherapy according to the invention.

[0227] The present invention also relates to a method for treating a subject suffering from lung cancer, comprising: a) obtaining image data by using non-invasive imagining technologies such as scanners, magnetic resonance imaging (MRI) or PET (Positron Emission Tomography), preferably image data obtained by using a CT scan; b) predicting a risk of disease progression when treated by immunotherapy using the PPPS as defined the above section entitled “Use of a patient progression probability score (PPPS) for predicting the risk of disease progression in a subject suffering from lung cancer when treated by immunotherapy” or the method for predicting the risk of disease progression in a subject suffering from lung cancer treated by immunotherapy according to the invention; c) if a low or moderate risk of disease progression when treated by immunotherapy has been predicted in step b), administering immunotherapy to the subject; and d) alternatively, if a high risk of disease progression when treated by immunotherapy has been predicted in step b), administering another treatment to the subject, the other treatment being preferably selected from chemotherapy, targeted therapy and combination treatments.

[0228] In the above method, immunotherapy is mainly administered (either alone or in combination with another treatment) when a low or moderate risk of disease progression when treated by immunotherapy has been predicted in step b), thus ensuring a minimum probability that the immunotherapy treatment will be useful to the treated subject. When a high risk of disease progression when treated by immunotherapy has been predicted in step b), immunotherapy alone is not administered to the subject. Instead, another treatment is administered to the subject to improve chances that the treated subject responds to the administered treatment.

[0229] The other treatment is preferably selected from chemotherapy, targeted therapy, radiotherapy and combination treatments.

[0230] Non-limiting examples of chemotherapies that may be administered to the lung cancer patient include platinum-based chemotherapy, pemetrexed, paclitaxel, docetaxel.

[0231] Non-limiting examples of targeted therapies that may be administered to the lung cancer patient include tyrosine kinase inhibitors for patients with specific mutations (e.g., EGFR, ALK, ROS1 ), bevacizumab.

[0232] Another alternative is to use combination treatments combining two or more distinct treatments to improve chances that the treated subject responds to the administered treatment. In this case, at least one of the treatments of the combination is selected from chemotherapy, targeted therapy and radiotherapy. The other treatment(s) of the combination may be selected from the same list or may be immunotherapy.

[0233] The following examples merely intend to illustrate the present invention.

[0234] EXAMPLES

[0235] Example 1 : Predicting Lesion-Level Outcomes and Survival in Advanced Non-Small cell lung cancers treated with Durvalumab

[0236] This study is an ancillary analysis of the NSCLC patients included in the CP1108 study, a first-time-in-human Phase 1 / 2 study with a dose-escalation phase evaluating durvalumab, an anti-PD-L1 immunotherapy in participants with advanced solid tumors (NCT01693562). The objective of this study was to evaluate the association between the CD8-Rscore of a given lesion and the progression of that same lesion at the first follow-up CT (Sun R, Limkin EJ, Vakalopoulou M, et al. A radiomics approach to assess tumour-infiltrating CD8 cells and response to anti-PD-1 or anti-PD-L1 immunotherapy: an imaging biomarker, retrospective multicohort study. The Lancet Oncology 2018; 19: 1180-91 ) and to evaluate the association between the spatial inter-lesion heterogeneity of the CD8-Rscores of the analyzed lesions with progression-free survival (PFS) and overall survival (OS) of advanced Non-small cell lung cancer patients treated with durvalumab anti-PD-L1 immunotherapy. We also designed a fine-tuned radiomic signature for this specific population. Materials and Methods

[0237] Data and study design

[0238] Patients with metastatic NSCLC and treated with durvalumab in the CP-1108 (NCT01693562) phase 1 / 2 study.

[0239] Patients with contrast-enhanced computed tomography (CT) available at baseline (E0) were included for analysis. For each patient, the first follow-up CT scan (E1 ) was included when available.

[0240] A training set and a test set were predefined (80 / 20 ratio) by the promotor of the study. Clinical data, baseline (E0) and follow-up (E1 ) CTs of the training set were available. Our team was blinded to the follow-up CTs of the test set and assessment of lesion response in the test set were done independently by the promotor (FIGURE 1 ).

[0241] PD-L1 high was defined as a tumor-cell Score > 25%.

[0242] Image analysis

[0243] Definition of analyzed lesions

[0244] The analyzed lesions were any tumors (primary or secondary) that were identifiable on baseline CTs. Small lesions <5mm were also included to assess the impact of lesion size on radiomics prediction. Lesions that could not be accurately discriminated from surrounding tissues (i.e. lung nodule adjacent or within atelectasis) or from other adjacent lesions at baseline or follow-up CTs (i.e. confluent metastases) were not delineated and excluded.

[0245] Feature extraction and computation of the CD8-Rscore

[0246] Contrast-enhanced CTs were used to extract radiomics features. All CT images were reconstructed using soft or standard convolution kernels and had a slice thickness of less than 5 mm. On baseline and follow-up scans, experienced physicians annotated the different lesions. Each lesion was divided into two volumes of interest (VOIs), the core tumor and a peripheral ring of 2 mm on both sides of the tumor boundaries. Images were resized to 1x1x1 mm3voxels and resampled into 400 discrete values using absolute discretization from to -1000 and 3000 HU. Radiomics features of each VOI were extracted using LIFEx software version 6.67 (Local Image Feature Extraction, freeware, www.lifexsoft.org) (Nioche C, Orlhac F, Boughdad S, et al. LIFEx: A Freeware for Radiomic Feature Calculation in Multimodality Imaging to Accelerate Advances in the Characterization of Tumor Heterogeneity. Cancer Res 2018; 78: 4786-9).

[0247] The radiomics analysis to compute the CD8-Rscore was performed according to the previously described method (Sun R, Limkin EJ, Vakalopoulou M, et al. A radiomics approach to assess tumour-infiltrating CD8 cells and response to anti-PD-1 or anti-PD-L1 immunotherapy: an imaging biomarker, retrospective multicohort study. The Lancet Oncology 2018; 19: 1180-91 ) without any fine-tuning.

[0248] The previously validated CD8-Rscore consisted of a linear regression on the basis of eight variables: five radiomics features extracted from each lesion, two variables about lesion location, and one imaging-acquisition related variable - the peak kilovoltage (kVp). The five features from the radiomics signature were extracted and normalized in the range 0 to 1 according to the data from the training set of the original publication. VOI location was labeled as adenopathy, head and neck, lung, liver, or other. Coefficients from the validated signature were applied to these eight variables to obtain the radiomics score of each analyzed lesion.

[0249] Response evaluation

[0250] Lesion response

[0251] A lesion-wise evaluation of relative change in diameter between baseline and follow-up was carried out using RECIST 1.1 criteria. A responding lesion was defined by a decrease in lesion size of at least 30%. A progressive lesion was defined by an increase in lesion size of at least 20%.

[0252] Patient patterns of response

[0253] To take into account the overall tumor burden and to assess the different patterns of response on a patient basis, mixed response was defined as the presence of both progressive and responding lesions, as opposed to patients presenting only responding (uniform response) or progressive lesions (uniform progression, progressive disease [PD]), irrespective of stable lesions, or only stable lesions (stable disease [SD]).

[0254] Spatial heterogeneity evaluation

[0255] To assess the impact of intra-patient inter-lesion CD8-Rscore heterogeneity, conventional first order histogram-based distribution metrics such as minimum value, maximal value, mean value, standard deviation, skewness, kurtosis and entropy of the pretreatment lesions’ CD8-Rscore were retrieved.

[0256] End points

[0257] The primary objective of the ancillary analysis was the association between the unrefined CD8-Rscore and lesion progression at a lesion level on the whole cohort. Spatial heterogeneity assessed according to the distribution metrics of the CD8-Rscore and especially the minimal value of the CD8-Rscore were used to evaluate their association with patient clinical outcomes (PFS, OS).

[0258] The primary endpoint for the fine-tuning of the signature was the prediction of lesion progression - at a lesion-level in the test set.

[0259] The patient progression probability was defined according to the probability to have at least one progressive lesion.

[0260] Secondary objectives were to evaluate the association between the concatenation of the lesion predictions at the patient level with clinical outcomes in the training and test sets.

[0261] Statistical analysis

[0262] Comparisons between variables were performed using Wilcoxon signed-rank test or Kruskal-Wallis test for continuous variables, and Fisher test for categorical variables.

[0263] To assess the association between radiomics and lesion response, the CD8-Rscore was evaluated as a continuous variable. Subgroup analyses according to lesion volume were performed as a sensitivity analysis.

[0264] To assess the association between radiomics and patient survival, patients were dichotomized into two groups of risk according to the spatial distribution metrics of the CD8-Rscore using the median value.

[0265] To assess the impact of potential clinical cofounding factors, clinical and biological variables were also analyzed in univariate and multivariate analyses. For continuous variables, patients were dichotomized into two distinct groups on the basis of median value. A threshold of <.05 was defined for double-tailed P-value’s significance. Statistical analyses were performed using R software version 3.6.0 (https: / / www.r-project.org / ) (R Core Team. R: a language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing, 2008 http: / / www.R-project.org / ). OS and PFS were computed according to the Kaplan-Meier method and Cox proportional-hazards survival estimates. Endpoints were death from any cause for OS, and any recurrence or death for PFS. Multivariate models included the clinically relevant variables that were not redundant with a P-value <0.05 in univariate analysis. A Backward Stepwise Regression was used to identify the most parsimonious model. No imputation was made for the missing data.

[0266] Machine learning

[0267] An XGBoost model was trained on the training data, and hyperparameters (max depth, learning rate, n_estimators) were optimized using a 10-fold cross-validation approach. Lesions of more than 1 ml were kept for the analysis. The best hyperparameters were retained for the final model, which is referred to the Lesion Progression Probability score (LPPS).

[0268] The cutoff to classify lesions as progressive or non-progressive was optimized using Youden Index Optimization and applied to the test set.

[0269] The Lesion Progression Probability Score of lesion progression was aggregated on a patient level to determine the Patient Progression Probability score. It was designed as the probability to get at least one lesion classified as progressive by the LPPS at a lesionlevel.

[0270] PPPS = 1 - n"=i(l - LPPSO (Formula I)

[0271] With LPPSi = probability of a lesion progression according to the Lesion Progression Probability score (LPPS) fitted using a logistic regression in the training set.

[0272] The cutoff used at the patient level to classify patients’ outcomes between high, moderate and low risk was chosen at 0.95 and 0.5.

[0273] Results

[0274] Patient characteristics and analyzed lesions

[0275] A total of 191 patients in the training set and 48 patients in the test set have been screened. After excluding patients for whom contrast enhanced CT at baseline were not available, 148 patients in the training set and 40 patients in the test set, included in the CP-1108 trial between May 2013 and February 2016 and treated with durvalumab (anti- PD-L1 ), were analyzed in this study (flowchart in FIGURE 2). Median follow-up was 9.2 months (IQR [3.1 , 24.0]).

[0276] Of the 188 patients, 44.1% were over 65 years old, 52.1% had non-squamous NSCLC and almost all had stage IV disease at enrollment (89.4%). PD-L1 status was available in 174 patients (92.6%), of which 93 were PD-L1 high (53.4%) (TABLE 4). TABLE 4

[0277] Of the 148 patients in the training set, a total of 906 tumor lesions were delineated at baseline (E0). Follow-up (E1 ) CT scans were available for 120 patients (81.1%), allowing a total of 701 lesions (77.4%) to be delineated at both time points. Median time between E0 and E1 was 50 days (IQR [44, 58]) (TABLE 5, 6). Regarding the 40 patients from the test set, 231 tumor lesions were delineated at baseline. The independent assessment of each lesion response was possible for 31 patients (77.5%) and 165 lesions (71.4%) (TABLE 5). TABLE 5. Description of the volume of interest analysed in the train and test sets at the start of the study

[0278] Table 6 Lesion delineated for radiomics analysis, according to their volume: Treatment response

[0279] The best overall response in the training set consisted of four CR (2.7%), 22 PR (14.9%), 38 SD (25.7%) and 64 PD (43.2%) and was not different from the test set with no CR (0%), five PR (12.5%), 14 SD (35.0%) and 14 PD (35.0%) (TABLE 7). Median OS and PFS was 9.7 months (95% Cl [7.9, 15]) and 1.6 months (95% Cl [1.4, 2.8]) in the training set, and 7.6 months (95% Cl [3.6, 15.5]) and 1.7 months (95% Cl [1.2, 3.9]) in the test set (HR=1.2 months, 95% Cl [0.85, 1.8], P-value = 0.27 and HR=1.1 months, 95% Cl [0.78, 1.7], P-value = 0.48 for OS and PFS respectively). Table 7. Treatment outcomes and characteristics in the train and test.

[0280] CD8-Rscore and association of lesion progression and patient outcomes

[0281] Lesion progression

[0282] At the first evaluation (E1 ), among the lesions 701 lesions of the training set and the 165 lesions of the test set for which lesion response was independently evaluated at E1 , the rate of progressive, responding and stable lesions were 34.7%, 7.9% and 57.4% in the training set, and 18.2%, 15.8%, and 66.1% (P-value<0.0001 ). The CD8-Rscore at baseline could be determined for 865 out of the 866 lesions (99.9%) (because of the size of the volume of interest). The baseline CD8 radiomics score was associated with the tumor diameter changes at E1 (Sperman’s Rho=-0.123, P- value=0.0003). Lower CD8 radiomics score were associated with progressive lesions (AUC=0.59, 95%CI[0.55, 0.63], P-value<0.0001 ) (FIGURE 3).

[0283] The performance of the CD8-Rscore for predicting progressive lesions was constant according to the volume of the lesion (AUC=0.59, 95%CI[0.55, 0.63], P-value<0.0001 for lesions >0.5 ml (n=778) and AUC=0.61 , 95%CI[0.56, 0.65], P-value<0.0001 for lesions >1 ml (n=687), although it did not reach significance for smaller lesions of less than 1 ml (n=178, AUC=0.55, 95%CI[0.47, 0.64], P-value=0.23) and 0.5ml (n=87, AUC=0.58, 95%CI[0.46, 0.71], P-value=0.19). The predictive performance of the CD8-Rscore varied according to the tumor location, with better performance for liver lesions (AUC=0.66, 95%CI[0.58, 0.75], P-value=0.0002) and adrenal glands (AUC=0.72, 95%CI[0.54,0.90], P-value=0.046). AUC of the prediction for lung lesions was 0.57, 95%CI[0.50, 0.64], P-value=0.069) when considering lesions of more than 0.5 ml (FIGURE 3).

[0284] Patient response

[0285] Patients with a higher CD8-Rscore when looking for the least infiltrated lesion (the minimum value of the CD8-Rscore across all the lesions of a patient) had a better prognosis than patients with a lower CD8-Rscore (HR=0.70, 95%CI[0.51 ; 0.96], P- value=0.029 for OS and HR=0.68, 95%CI[0.49 ; 0.92], P-value=0.014, for PFS) (FIGURE 3). The entropy of the inter-lesion distribution of the CD8-Rscore, which evaluates the heterogeneity among the different lesions, was also associated with OS and PFS (HR=1 .49, 95%CI [1.08, 2.05], P-value=0.016 and HR=1.63, 95%CI [1.19, 2.23], P-value=0.0022 respectively, FIGURE 4). of lesion

[0286] After feature selection, a total of 52 radiomic features were extracted from the tumor and the rim of each analyzed lesion, along with the CD8-Rscore. Only lesions of more than 1 ml were considered for the machine learning algorithm. After optimization of the XGboost model using the training set to predict lesion progression, the final signature was composed of 24 features, among which the CD8-Rscore was an important predictive feature (FIGURE 5), although five other radiomics features were found to be more important. Lesion Progression Probability Score

[0287] The Lesion Progression Probability score (LPPS) discriminated progressive lesions in the training set and in the test set with an AUC of 0.82, 95%CI[0.78, 0.85], P-value<0.0001 and an AUC of 0.78, 95%CI[0.68, 0.87], P-value<0.0001 respectively (TABLE 8). Performance according to location of the VOI are shown in FIGURE 6.

[0288] Table 8. Performance of the CD8-based Lesion Progression Score in the test set

[0289] Patient Progression Probability Score

[0290] Probability to have at least one progressive lesion was defined as “high” when the probability was estimated to be more than 95%, while “moderate risk” was defined as a probability of more than 50%. The patient progression probability score (PPPS) could discriminate OS of patients in both training (P-value = 0.006, and P-value = 0.022 respectively), as well as PFS (P-value = 0.026, and P-value = 0.015) (FIGURE 7). Univariate analyses of OS and PFS are summarized in supplemental data (TABLE 9) and show that PPPS value is correlated to both OS and PFS (see p-value for high vs low PPPS). Table 9. Univariate analysis for Overall survival and Progression-Free survival

[0291] The high risk group remained independently associated with OS and PFS in the test set when adjusting for PD-L1 status, haeomoglin level, ECOG status, or the number of analyzed lesions in multivariate analyses (TABLE 10)

[0292] Table 10. Multivariate analysis

[0293] Model with PDL1

[0294] Model with Hb, ECOG Model with number of lesions

[0295] Conclusions

[0296] These results confirm the predictive value of the CD8-Rscore at the individual lesion level in NSCLC patients undergoing durvalumab anti-PD-L1 immunotherapy.

[0297] These results also confirm the potential interest of such imaging biomarkers in the assessment of the spatial heterogeneity of the disease, emphasizing the prognostic value of the least infiltrated lesion.

[0298] Moreover, an optimized radiomics-based signature has been designed, which can be used for predicting survival when treated by immunotherapy and thus response to immunotherapy in NSCLC patients with a better accuracy than the previously disclosed CD8-Rscore.

[0299] In conclusion, this study underscores the contribution of imaging biomarkers in guiding clinicians towards a new area of radiomics-based precision medicine. BIBLIOGRAPHIC REFERENCES

[0300] Bellesoeur A, Torossian N, Amigorena S, Romano E. Advances in theranostic biomarkers for tumor immunotherapy. Current Opinion in Chemical Biology 2020; 56: 79-90.

[0301] Cataldo SD, Ficarra E. Mining textural knowledge in biological images: applications, methods and trends. Comput Struct Biotechnol J 2016.

[0302] Chen DS, Mellman I. Oncology Meets Immunology: The Cancer-Immunity Cycle. Immunity 2013; 39: 1-10

[0303] Henry T, Sun R, Lerousseau M, et al. Investigation of radiomics based intra-patient intertumor heterogeneity and the impact of tumor subsampling strategies. Sci Rep 2022; 12: 17244.

[0304] IBSI Reference manual version 1.0 of December 2019 available from https: / / buildmedia.readthedocs.org / media / pdf / ibsi / latest / ibsi.pdf

[0305] Korpics MC, Onderdonk BE, Dadey RE, et al. Partial tumor irradiation plus pembrolizumab in treating large advanced solid tumor metastases. J Clin Invest 2023; 133. DOI:10.1172 / JC1162260.

[0306] Korpics MC, Polley M-Y, Bhave SR, et al. A Validated T Cell Radiomics Score Is Associated With Clinical Outcomes Following Multisite SBRT and Pembrolizumab. International Journal of Radiation Oncology, Biology, Physics 2020; 0.

[0307] DOI : 10.1016 / j .ij robp.2020.06.026.

[0308] Lee HM. Strategies for Manipulating T Cells in Cancer Immunotherapy. Biomol Ther (Seoul). 2022 Jul 1 ;30(4) :299-308

[0309] LIFEx documentation relating to Features for LIFEx version 7.6.n, as updated on 2024 / 01 / 31 available from https: / / www.lifexsoft.org / index.php / resources / documentation

[0310] Liu X. Classification accuracy and cut point selection. Stat. Med. 2012; 31 (23): 2676- 2686

[0311] Nioche C, Orlhac F, Boughdad S, et al. LIFEx: A Freeware for Radiomic Feature Calculation in Multimodality Imaging to Accelerate Advances in the Characterization of Tumor Heterogeneity. Cancer Res 2018; 78: 4786-9

[0312] Perkins NJ, Schisterman EF. The inconsistency of "optimal" cutpoints obtained using two criteria based on the receiver operating characteristic curve. Am. J. Epidemiol. 2006; 163(7): 670-675

[0313] R Core Team. R: a language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing, 2008 http: / / www.R-project.org /

[0314] Robert C. A decade of immune-checkpoint inhibitors in cancer therapy. Nat Commun 2020; 11 : 3801 ;

[0315] Sook Young Woo, Seonwoo Kim. Determination of cutoff values for biomarkers in clinical studies. Precision and Future Medicine 2020;4(1 ):2-8. https: / / doi.org / 10.23838 / pfm.2019.00135 Sun R, Lerousseau M, Briend-Diop J, et al. Radiomics to evaluate interlesion heterogeneity and to predict lesion response and patient outcomes using a validated signature of CD8 cells in advanced melanoma patients treated with anti-PD1 immunotherapy. J Immunother Cancer 2022; 10: e004867.

[0316] Sun R, Limkin EJ, Vakalopoulou M, et al. A radiomics approach to assess tumourinfiltrating CD8 cells and response to anti-PD-1 or anti-PD-L1 immunotherapy: an imaging biomarker, retrospective multicohort study. The Lancet Oncology 2018; 19: 1180-91

[0317] Sun R, Limkin EJ, Vakalopoulou M, et al. A radiomics approach to assess tumourinfiltrating CD8 cells and response to anti-PD-1 or anti-PD-L1 immunotherapy: an imaging biomarker, retrospective multicohort study. The Lancet Oncology 2018; 19: 1180-91.

[0318] Sun R, Sundahl N, Hecht M, et al. Radiomics to predict outcomes and abscopal response of patients with cancer treated with immunotherapy combined with radiotherapy using a validated signature of CD8 cells. J Immunother Cancer 2020; 8: e001429.

[0319] Trebeschi S, Drago SG, Birkbak NJ, et al. Predicting Response to Cancer Immunotherapy using Non-invasive Radiomic Biomarkers. Ann Oncol 2019; published online March 21. DOI : 10.1093 / annonc / mdz108.

[0320] Tunali I, Gray JE, Qi J, et al. Novel clinical and radiomic predictors of rapid disease progression phenotypes among lung cancer patients treated with immunotherapy: An early report. Lung Cancer 2019; 129: 75-9.

[0321] Yang Y, Yang J, Shen L, et al. A multi -omics-based serial deep learning approach to predict clinical outcomes of single-agent anti-PD-1 / PD-L1 immunotherapy in advanced stage non-small-cell lung cancer. Am J Transl Res 2021 ; 13: 743-56.

[0322] Youden WJ. Index for rating diagnostic tests. Cancer 1950; 3(1 ): 32-35

Claims

CLAIMS1. Use of a Lesion Progression Probability Score (LPPS) for predicting the risk of progression of a tumor lesion in a subject suffering from lung cancer when treated by immunotherapy, wherein the tumor lesion comprises a core tumoral region and a peritumoral region referred to as a ring and the LPPS of each lesion comprises the 6 following radiomics features:(i) the standard deviation of the intensity distribution in the ring (abbreviated as ring_std Value);(ii) the Low Grey Level Run Emphasis (LGLRE) feature of the Grey Level Run Length Matrix (GLRLM) of the ring (abbreviated as ring_GLRLM_LGRE);(iii ) the minimum intensity in the ring (abbreviated as ring_minValue);(iv) the Intensity Histogram Kurtosis feature of the core tumoral region (abbreviated as tu m_H I STO_K u rtosi s) ;(v) the Correlation feature of the Grey Level Cooccurrence Matrix (GLCM) of the ring (abbreviated as ring_GLCM_Correlation); and(vi) a complex radiomics feature combining the values of:• the minimum intensity in the core tumoral region (abbreviated as tum_MinValue),• two or more Gray-level Run Length Matrix (GLRLM) features selected from the group consisting of: o GLRLM short-run high gray-level emphasis of the core tumoral region (abbreviated as tum_GLRLM_SRHGE), o GLRLM short-run low gray-level emphasis of the ring (abbreviated as ring_GLRLM_SRLGE), o GLRLM low gray-level run emphasis of the ring (abbreviated as ring_GLRLM_LGRE), and o GLRLM long-run low gray-level emphasis of the ring (abbreviated as ring_GLRLM_LRLGE),• the imaging-acquisition feature kVp,• optionally, the location feature lymph node metastasis (abbreviated as VOI_Adenopathy), wherein the value of VOI_Adenopathy is 1 when the analyzed lesion is a lymph node metastasis and 0 when the analyzed lesion is at any other location, and• optionally, the location feature head and neck lesion (abbreviated as VOI_head_and_neck), wherein VOI_head_and_neck is 1 when the analyzed lesionis a tumor lesion in the pharynx, larynx, oral cavity, or salivary gland and 0 when the analyzed lesion is at any other location.

2. The use according to claim 1 , wherein the two or more GLRLM features included in the complex radiomics feature (vi) comprise tum_GLRLM_SRHGE and ring_GLRLM_LRLGE.

3. The use according to claim 1 or claim 2, wherein the complex radiomics feature (vi) combines the values of all four GLRLM features tum_GLRLM_SRHGE, ring_GLRLM_LRLGE, ring_GLRLM_SRLGE, and ring_GLRLM_LGRE.

4. The use according to any one of claims 1 to 3, wherein the complex radiomics feature (vi) is the lung cancer CD8-Rscore, a complex radiomics feature consisting of a linear regression of:• the minimum intensity in the core tumoral region (abbreviated as tum_MinValue),• four Gray-level Run Length Matrix (GLRLM) features consisting of: o GLRLM short-run high gray-level emphasis of the core tumoral region (abbreviated as tum_GLRLM_SRHGE), o GLRLM short-run low gray-level emphasis of the ring (abbreviated as ring_GLRLM_SRLGE), o GLRLM low gray-level run emphasis of the ring (abbreviated as ring_GLRLM_LGRE), and o GLRLM long-run low gray-level emphasis of the ring (abbreviated as ring_GLRLM_LRLGE),• the imaging-acquisition feature kVp, and• two location features: o lymph node metastasis (abbreviated as VOI_Adenopathy), and o head and neck lesion (VOI_head_and_neck), o wherein the value of VOI_Adenopathy is 1 when the analyzed lesion is a lymph node metastasis and 0 when the analyzed lesion is at any other location and VOI_head_and_neck is 1 when the analyzed lesion is a tumor lesion in the pharynx, larynx, oral cavity, or salivary gland, and 0 when the analyzed lesion is at any other location.

5. The use according to any one of claims 1 to 4, wherein the features are extracted from image data obtained by using non-invasive imagining technologies such as scanners,including Computed tomography (CT) scan and PET (Positron Emission Tomography) scan, preferably based on image data obtained by using a CT scan.

6. The use according to any one of claims 1 to 5, wherein the LPPS further comprises one or more and preferably all of the following radiomics and non-radiomics features:(a) radiomics features:(vii) Intensity Histogram Kurtosis of the ring (abbreviated as ring_HISTO_Kurtosis);(viii ) Compacity of the ring when three dimensional images of the tumor lesion are available (abbreviated as ring_SHAPE_Compacity.onlyFor3DROI);(x) Intensity Histogram Skewness of the ring (abbreviated as ri ng_H ISTO_Skewness) ;(xi) Small Zone Emphasis of the Grey-Level Zone Length Matrix (abbreviated as ring_GLZLM_SZE);(xii) the minimum intensity in the core tumoral region (abbreviated as tum_minValue);(xiii) Contrast of the Neighborhood Grey-Level Difference Matrix of the core tumoral region (abbreviated as tum_NGLDM_Contrast);(xiv) Busyness of the Neighborhood Grey-Level Difference Matrix of the core tumoral region (abbreviated as tum_NGLDM_Busyness);(xv) Sphericity of the core tumoral region when three dimensional images of the tumor lesion are available (abbreviated as tum_SHAPE_Sphericity.onlyFor3DROI);(xvi) Large Zone Emphasis of the Grey-Level Zone Length Matrix of the ring (abbreviated as ring_GLZLM_LZE);(xvii) Small Zone Low Grey Level Emphasis of the Grey-Level Zone Length Matrix of the ring (abbreviated as ring_GLZLM_SZLGLE);(xviii) Intensity Histogram Skewness of the core tumoral region (abbreviated as tu m_H I ST O_Skewness ) ;(xix) Surface in mm2of the core tumoral region when three dimensional images of the tumor lesion are available (abbreviated as tum_SHAPE_Surface.mm2..onlyFor3DROI);(xx) Contrast of the Grey Level Cooccurrence Matrix of the core tumoral region (abbreviated as tum_GLCM_Contrast) ;(xxi) Contrast of the Grey Level Cooccurrence Matrix of the ring (abbreviated as ring_GLCM_Contrast) ;(xxii) Correlation of the Grey Level Cooccurrence Matrix of the core tumoral region (abbreviated as tum_GLCM_Correlation) ;(xxiii) Intensity Histogram Entropy_log10 of the core tumoral region (abbreviated as tum_HISTO_Entropy_log10); and(xxiv) Volume in mL of the core tumoral region (abbreviated as tum_SHAPE_Volume.mL); and(b) non-radiomics feature :(ix) the peak kilovoltage (abbreviated as kVp).

7. The use according to any one of claims 1 to 6, wherein the LPPS is a model returning for each analyzed tumor lesion an LPPS value depending on the values of each of the features included in the LPPS, wherein the LPPS value is positively or negatively correlated to the risk of progression of the tumor lesion when treated by immunotherapy, and wherein the model has been obtained by training an algorithm on the features to be included in the LPPS extracted from image data of a reference population of lung cancer patients.

8. The use according to claim 7, wherein the algorithm is a machine learning algorithm, preferably selected from boosting algorithms, linear regression, logistic regression, random forest algorithms and deep learning algorithms.

9. The use according to claim 8, wherein the machine learning algorithm is a boosting algorithm, preferably a gradient boosting algorithm such as XGBoost.

10. The use according to any one of claims 7 to 9, wherein the LPPS value is used to determine the risk of progression of the tumor lesion as follows: a) the risk of progression of the tumor lesion is determined by comparison of the LPPS value to a cut-off value, wherein the cut-off value has preferably been previously determined based on the reference population of lung cancer patients by the algorithm used for obtaining the model; or b) the LPPS value directly represents the probability that the analyzed tumor lesion progresses or that the analyzed tumor lesion does not progress.11 . Use of a patient progression probability score (PPPS) for predicting the risk of disease progression in a subject suffering from lung cancer when treated by immunotherapy, wherein the PPPS combines the LPPS values of all analyzed tumor lesions of the subject with a volume equal to or higher than a minimum volume Vmin , wherein the LPPS of each tumor lesion is as defined in any one of claims 1 to 10.

12. The use according to claim 11 , wherein the LPPS value directly represents the probability that the analyzed tumor lesion progresses or does not progress and the PPPS is of formula I:PPPS = 1 - nr=i(l - LPPSi) (Formula I) wherein n is the number of tumor lesions of the subject and for each tumor lesion i and LPPSi is the LPPS of lesion I, and the PPPS value directly represents the probability that the disease progresses or does not progress.

13. The use according to claim 12, wherein the LPPS value directly represents the probability that the analyzed tumor lesion progresses, the PPPS is of formula I, the PPPS value directly represents the probability that the disease progresses, and the risk of disease progression when treated by immunotherapy is:• High if the PPPS is higher than 0.95,• Moderate if the PPPS is higher than 0.5 and lower or equal to 0.95; and• Low if the PPPS is lower or equal to 0.5.

14. The use of any one of claims 1 to 13, wherein the lung cancer is non-small cells lung cancer (NSCLC).

15. The use of any one of claims 1 to 14, wherein the immunotherapy is a direct or indirect T cell-based immunotherapy, preferably an immune checkpoint inhibitor (ICI), in particular an anti-PD-1 or an anti-PD-L1 antibody.

16. A method for predicting the risk of progression of a tumor lesion in a subject suffering from lung cancer when treated by immunotherapy, wherein the tumor lesion comprises a core tumoral region and a peritumoral region referred to as a ring from previously obtained image data obtained by using non-invasive imagining technologies, comprising: a) calculating a Lesion Progression Probability Score (LPPS) as defined in any one of claims 1 to 10 based on the previously obtained image data obtained by using non-invasive imagining technologies; and b) predicting the risk of progression of the tumor lesion in the subject when treated by immunotherapy based on the LPPS calculated in step a).

17. A method for predicting the risk of disease progression in a subject suffering from lung cancer treated by immunotherapy, from previously obtained image data obtained by using non-invasive imagining technologies, comprising: a) calculating a Lesion Progression Probability Score (LPPS) as defined in any one of claims 1 to 10 for all analyzed tumor lesions with a volume equal to or higher than a minimum volume Vmin of the subject based on the previously obtained image data obtained by using non-invasive imagining technologies; b) calculating a patient progression probability score (PPPS) combining the LPPS values of all tumor lesions of the subject; and c) predicting the risk of disease progression when treated by immunotherapy based on the PPPS value calculated in step b).

18. An immunotherapy for treating lung cancer in a subject, wherein the immunotherapy is for administration to subjects for whom a low or moderate risk of disease progression when treated by immunotherapy has been predicted using the PPPS as defined in any one of claims 11 to 13 or the method of claim 17.

Images