# US 2024/0005502 A1 · Machine learning to predict lung cancer involvement of lymph nodes

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2024/0005502 A1
- Document ID: US20240005502A1
- Status: Published application
- Kind: A1
- Application number: 18/255,432
- Filed: 2021-11-30
- Published: 2024-01-04
- Earliest claimed: 2020-12-01
- Assignees: Johnson & Johnson Enterprise Innovation Inc.
- Applicants: Johnson & Johnson Enterprise Innovation Inc.
- Inventors: George R. Washko, JR.; Raul San Jose Estepar; Charles Matthew Kinsey; Christopher Scott Stevenson
- Web page: https://patentgrep.com/patent/US20240005502A1

## CPC

- G06T 7/0012: {Biomedical image inspection}
- G06T 7/0014: {using an image reference approach}
- G06V 10/764: using classification, e.g. of video objects
- G06V 10/7715: {Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods}
- G06V 10/774: Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
- G16H 30/40: for processing medical images, e.g. editing
- G16H 50/20: for computer-aided diagnosis, e.g. based on medical expert systems
- G16H 50/30: for calculating health indices; for individual health risk assessment
- G06T 2207/10068: Endoscopic image
- G06T 2207/10081: Computed x-ray tomography [CT]
- G06T 2207/10088: Magnetic resonance imaging [MRI]
- G06T 2207/10104: Positron emission tomography [PET]
- G06T 2207/10116: X-ray image
- G06T 2207/10132: Ultrasound image
- G06T 2207/20064: Wavelet transform [DWT]
- G06T 2207/20076: Probabilistic image processing
- G06T 2207/20081: Training; Learning
- G06T 2207/20084: Artificial neural networks [ANN]
- G06T 2207/30061: Lung
- G06T 2207/30064: Lung nodule
- G06T 2207/30096: Tumor; Lesion
- G06T 2207/30101: Blood vessel; Artery; Vein; Vascular
- G06V 2201/03: Recognition of patterns in medical or anatomical images
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA

## Abstract

Disclosed herein are methods for determining a subject level risk of metastatic cancer involving the training and/or deployment of models to determine 1) a lymph node level risk of individual lymph node involvement and/or 2) a subject level risk of lymph node involvement. Thus, the methods can identify patients who are high or low risk for having nodal disease and optionally enable the guided intervention of cancer patients, for example, via treatment.

## Claims

### Claim 1

A method of determining a subject level risk of metastatic cancer of a subject, the method comprising:
obtaining one or more images captured from the subject comprising a plurality of lymph nodes of the subject; and
predicting the subject level risk of metastatic cancer by applying a risk model to extracted features of the obtained one or more images, the risk model trained to predict lymph node (LN) level risks of metastatic cancer for lymph nodes in images,
wherein the risk model is trained using at least labels derived from imputed LN-level risks of metastatic cancer for a set of lymph nodes of a reference individual, the imputed LN-level risks of the set of lymph nodes determined using at least an imputation model that discriminates between cancerous and non-cancerous lymph nodes of training images.

### Claim 2 (depends on 1)

The method of claim 1, wherein predicting the subject level risk of metastatic cancer further comprises:
selecting one or more of the plurality of lymph nodes based on their LN-level risk of metastatic cancer predicted by the risk model; and
determining the subject level risk of metastatic cancer using the LN-level risk of metastatic cancer predicted for the one or more lymph nodes.

### Claim 3 (depends on 2)

The method of claim 2, wherein selecting one or more of the plurality of lymph nodes comprises identifying the lymph node with the highest probability of LN-level risk, and wherein determining the subject level risk of metastatic cancer comprises assigning the highest probability of LN-level risk as the subject level risk of metastatic cancer.

### Claim 4 (depends on 1, 2, 3)

The method of any one of claims 1-3, wherein the risk model predicts LN-level risks of metastatic cancer with a greater range of probabilities than the imputed LN-level risks predicted by the imputation model.

### Claim 5 (depends on 1, 2, 3, 4)

The method of any one of claims 1-4, wherein the set of lymph nodes of a reference individual is selected by:
determining a median risk value; and
comparing imputed LN-level risks of the lymph nodes of the reference individual to the median risk value.

### Claim 6 (depends on 5)

The method of claim 5, wherein the set of lymph nodes of a reference individual is further selected by:
including one or more lymph nodes with LN-level risks greater than the median risk value in the set of lymph nodes.

### Claim 7 (depends on 5, 6)

The method of claim 5 or 6, wherein one or more lymph nodes with LN-level risks less than the median risk value in the set of lymph nodes are excluded from the set of lymph nodes.

### Claim 8 (depends on 5, 6, 7)

The method of any one of claims 5-7, wherein the median risk value is a median intrasubject LN-level risk of the reference individual.

### Claim 9 (depends on 1, 2, 3, 4, 5, 6, 7, 8)

The method of any one of claims 1-8, wherein the one or more images comprises a computed tomography (CT) image.

### Claim 10 (depends on 1, 2, 3, 4, 5, 6, 7, 8, 9)

The method of any one of claims 1-9, wherein the one or more images comprises are obtained from a thoracic CT scan.

### Claim 11 (depends on 1, 2, 3, 4, 5, 6, 7, 8, 9, 10)

The method of any one of claims 1-10, further comprising determining a stage of the cancer according to the predicted subject level risk of metastatic cancer.

### Claim 12 (depends on 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11)

The method of any one of claims 1-11, further comprising selecting a diagnostic or treatment for the subject according to the predicted subject level risk of metastatic cancer.

### Claim 13 (depends on 12)

The method of claim 12, wherein selecting a diagnostic or treatment for the subject comprises selecting either surgical tumor resection or combined bronchoscopic or endoscopic diagnosis and treatment when the predicted subject level risk of metastatic cancer is below a threshold risk value.

### Claim 14 (depends on 12)

The method of claim 12, wherein selecting a diagnostic or treatment for the subject comprises selecting additional diagnostic testing from one or both of performing a PET or PET-CT scan or performing a lymph node biopsy when the predicted subject level risk of metastatic cancer is above a threshold risk value.

### Claim 15 (depends on 12)

The method of claim 12, wherein selecting a diagnostic or treatment for the subject comprises selecting additional diagnostic testing of a PET or PET-CT scan when the predicted subject level risk of metastatic cancer is below a threshold risk value.

### Claim 16 (depends on 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15)

The method of any one of claims 1-15, wherein the imputation model is trained using training images obtained from thoracic CT scans.

### Claim 17 (depends on 16)

The method of claim 16, wherein the training images obtained from thoracic CT scans comprise one or more lymph nodes.

### Claim 18 (depends on 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17)

The method of any one of claims 1-17, wherein the imputation model is trained using training images of the National Lung Screening Trial (NLST).

### Claim 19 (depends on 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17)

The method of any one of claims 1-17, wherein the imputation model is trained using training images of a custom dataset.

### Claim 20 (depends on 16, 17, 18, 19)

The method of any one of claims 16-19, wherein the imputation model is trained by generating two or more radiomic panels from at least one of the training images.

### Claim 21 (depends on 20)

The method of claim 20, wherein the two or more radiomic panels comprise a Laplacian of Gaussians transformation of the training image and a wavelet transform of the training image.

### Claim 22 (depends on 20, 21)

The method of claim 20 or 21, wherein the imputation model is further trained by extracting features from each of the two or more radiomic panels; and training the imputation model using at least the extracted features.

### Claim 23 (depends on 20, 21)

The method of claim 20 or 21, wherein the imputation model is further trained by:
defining a region of interest (ROI) in the radiomic panels, the defined region of interest comprising a lymph node;
extracting features from the ROI in the radiomic panels; and
training the imputation model using at least the extracted features.

### Claim 24 (depends on 22, 23)

The method of claim 22 or 23, wherein the extracted features include one or more feature categories of first order features, shape features, gray level co-occurrence matrix 1 (GLCM) features, gray level run length matrix (GLRLM) features, gray level size zone matrix (GLSZM) features, and neighborhood gray tone difference matrix (NGTDM) features.

### Claim 25 (depends on 16, 17, 18, 19, 20, 21, 22, 23, 24)

The method of any of claims 16-24, wherein the imputation model is further trained using reference ground truth values indicating a presence or absence of metastasis in training images.

### Claim 26 (depends on 25)

The method of claim 25, wherein the reference ground truth values are derived from Nstage covariable values indicating the stage of the cancer, wherein a cancer stage of “0” or “1” indicate an absence of metastasis and wherein a cancer stage of “2” or “3” indicate a presence of metastasis.

### Claim 27 (depends on 26)

The method of claim 26, wherein the reference ground truth values indicate whether individual lymph nodes are cancerous or non-cancerous.

### Claim 28 (depends on 27)

The method of claim 27, wherein the reference ground truth values indicating whether individual lymph nodes are cancerous or non-cancerous are back-calculated from the Nstage covariable values based on characteristics of the tumor or lymph nodes.

### Claim 29 (depends on 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28)

The method of any one of claims 1-28, wherein the imputation model is a random forest classifier or a LASSO classifier.

### Claim 30 (depends on 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29)

The method of any one of claims 1-29, wherein the risk model is a random forest classifier or a LASSO classifier.

### Claim 31 (depends on 30)

The method of claim 30, wherein the risk model is trained using supervised learning techniques.

### Claim 32 (depends on 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31)

The method of any one of claims 1-31, wherein the plurality of lymph nodes comprises 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, or 21 lymph nodes.

### Claim 33 (depends on 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31)

The method of any one of claims 1-31, wherein the plurality of lymph nodes comprises between 8 to 20 lymph nodes, between 10 to 19 lymph nodes, between 12 to 18 lymph nodes, or between 14 and 16 lymph nodes.

### Claim 34 (depends on 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33)

The method of any one of claims 1-33, wherein the metastatic cancer is metastatic lung cancer.

### Claim 35 (depends on 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34)

The method of any one of claims 1-34, wherein the risk model exhibits a performance of at least AUC=78%, Sensitivity=82%, or Specificity=60%.

### Claim 36 (depends on 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35)

A method of training the risk model of any one of claims 1-35, the method comprising:
applying an imputation model to generate imputed LN-level risks for a plurality of lymph nodes in training images;
selecting a set of lymph nodes, wherein the lymph nodes in the set have LN-level risks that are greater than a median risk value;
using the LN-level risks of the lymph nodes in the set of lymph nodes as reference ground truths for training the risk model, the risk model able to predict LN-level risks with a greater range of probabilities than the imputed LN-level risks predicted by the imputation model.

### Claim 37 (depends on 36)

The method of claim 36, wherein the risk model is further trained by extracting features from one or more radiomic panels; and training the risk model using at least the extracted features.

### Claim 38 (depends on 37)

The method of claim 37, wherein extracting features from two or more radiomic panels comprises:
defining a region of interest (ROI) in the one or more radiomic panels, the defined region of interest comprising a lymph node;
extracting features from the ROI in the one or more radiomic panels.

### Claim 39 (depends on 37, 38)

The method of claim 37 or 38, wherein the extracted features include one or more feature categories of first order features, shape features, gray level co-occurrence matrix 1 (GLCM) features, gray level run length matrix (GLRLM) features, gray level size zone matrix (GLSZM) features, and neighborhood gray tone difference matrix (NGTDM) features.

### Claim 40

A system for determining a subject level risk of metastatic cancer of a subject, the system comprising:
an imaging device configured to capture one or more images of the subject; and
a computing device configured to perform the steps of:
obtaining one or more images captured from the subject comprising a plurality of lymph nodes of the subject; and
predicting the subject level risk of metastatic cancer by applying a risk model to extracted features of the obtained one or more images, the risk model trained to predict lymph node (LN) level risks of metastatic cancer for lymph nodes in images,
wherein the risk model is trained using at least labels derived from imputed LN-level risks of metastatic cancer for a set of lymph nodes of a reference individual, the imputed LN-level risks of the set of lymph nodes determined using at least an imputation model that discriminates between cancerous and non-cancerous lymph nodes of training images.

### Claim 41 (depends on 40)

The system of claim 40, wherein predicting the subject level risk of metastatic cancer further comprises:
selecting one or more of the plurality of lymph nodes based on their LN-level risk of metastatic cancer predicted by the risk model; and
determining the subject level risk of metastatic cancer using the LN-level risk of metastatic cancer predicted for the one or more lymph nodes.

### Claim 42 (depends on 41)

The system of claim 41, wherein selecting one or more of the plurality of lymph nodes comprises identifying the lymph node with the highest probability of LN-level risk, and wherein determining the subject level risk of metastatic cancer comprises assigning the highest probability of LN-level risk as the subject level risk of metastatic cancer.

### Claim 43 (depends on 40, 41, 42)

The system of any one of claims 40-42, wherein the risk model predicts LN-level risks of metastatic cancer with a greater range of probabilities than the imputed LN-level risks predicted by the imputation model.

### Claim 44 (depends on 40, 41, 42, 43)

The system of any one of claims 40-43, wherein the set of lymph nodes of a reference individual is selected by:
determining a median risk value; and
comparing imputed LN-level risks of the lymph nodes of the reference individual to the median risk value.

### Claim 45 (depends on 44)

The system of claim 44, wherein the set of lymph nodes of a reference individual is further selected by:
including one or more lymph nodes with LN-level risks greater than the median risk value in the set of lymph nodes.

### Claim 46 (depends on 44, 45)

The system of claim 44 or 45, wherein one or more lymph nodes with LN-level risks less than the median risk value in the set of lymph nodes are excluded from the set of lymph nodes.

### Claim 47 (depends on 44, 45, 46)

The system of any one of claims 44-46, wherein the median risk value is a median intrasubject LN-level risk of the reference individual.

### Claim 48 (depends on 40, 41, 42, 43, 44, 45, 46, 47)

The system of any one of claims 40-47, wherein the imaging device is one of a computed tomography (CT) scanner, magnetic resonance imaging (MRI) scanner, positron emission tomography (PET) scanner, x-ray scanner, or an ultrasound imaging device.

### Claim 49 (depends on 40, 41, 42, 43, 44, 45, 46, 47, 48)

The system of any one of claims 40-48, wherein the imaging device is a CT scanner and wherein the one or more images comprise a computed tomography (CT) image.

### Claim 50 (depends on 40, 41, 42, 43, 44, 45, 46, 47, 48, 49)

The system of any one of claims 40-49, wherein the one or more images comprise a thoracic CT image.

### Claim 51 (depends on 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50)

The system of any one of claims 40-50, wherein the computing device is further configured to perform the step of: determining a stage of the cancer according to the predicted subject level risk of metastatic cancer.

### Claim 52 (depends on 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51)

The system of any one of claims 40-51, wherein the computing device is further configured to perform the step of: selecting a diagnostic or treatment for the subject according to the predicted subject level risk of metastatic cancer.

### Claim 53 (depends on 52)

The system of claim 52, wherein selecting a diagnostic or treatment for the subject comprises selecting either surgical tumor resection or combined bronchoscopic or endoscopic diagnosis and treatment when the predicted subject level risk of metastatic cancer is below a threshold risk value.

### Claim 54 (depends on 53)

The system of claim 53, wherein selecting a diagnostic or treatment for the subject comprises selecting additional diagnostic testing of one or both of performing a PET or PET-CT scan or performing a lymph node biopsy when the predicted subject level risk of metastatic cancer is above a threshold risk value.

### Claim 55 (depends on 53)

The system of claim 53, wherein selecting a diagnostic or treatment for the subject comprises selecting additional diagnostic testing of a PET or PET-CT scan when the predicted subject level risk of metastatic cancer is below a threshold risk value.

### Claim 56 (depends on 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54)

The system of any one of claims 40-54, wherein the imputation model is trained using training images obtained from thoracic CT scans.

### Claim 57 (depends on 56)

The system of claim 56, wherein the training images obtained from thoracic CT scans comprise one or more lymph nodes.

### Claim 58 (depends on 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57)

The system of any one of claims 40-57, wherein the imputation model is trained using training images of the National Lung Screening Trial (NLST).

### Claim 59 (depends on 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57)

The system of any one of claims 40-57, wherein the imputation model is trained using training images of a custom dataset.

### Claim 60 (depends on 56, 57, 58, 59)

The system of any one of claims 56-59, wherein the imputation model is trained by generating two or more radiomic panels from at least one of the training images.

### Claim 61 (depends on 60)

The system of claim 60, wherein the two or more radiomic panels comprise a Laplacian of Gaussians transformation of the training image and a wavelet transform of the training image.

### Claim 62 (depends on 60, 61)

The system of claim 60 or 61, wherein the imputation model is further trained by extracting features from each of the two or more radiomic panels; and training the imputation model using at least the extracted features.

### Claim 63 (depends on 60, 61)

The system of claim 60 or 61, wherein the imputation model is further trained by:
defining a region of interest (ROI) in the radiomic panels, the defined region of interest comprising a lymph node;
extracting features from the ROI in the radiomic panels; and
training the imputation model using at least the extracted features.

### Claim 64 (depends on 62, 63)

The system of claim 62 or 63, wherein the extracted features include one or more feature categories of first order features, shape features, gray level co-occurrence matrix 1 (GLCM) features, gray level run length matrix (GLRLM) features, gray level size zone matrix (GLSZM) features, and neighborhood gray tone difference matrix (NGTDM) features.

### Claim 65 (depends on 56, 57, 58, 59, 60, 61, 62, 63, 64)

The system of any of claims 56-64, wherein the imputation model is further trained using reference ground truth values indicating a presence or absence of metastasis in training images.

### Claim 66 (depends on 65)

The system of claim 65, wherein the reference ground truth values are derived from Nstage covariable values indicating the stage of the cancer, wherein a cancer stage of “0” or “1” indicate an absence of metastasis and wherein a cancer stage of “2” or “3” indicate a presence of metastasis.

### Claim 67 (depends on 66)

The system of claim 66, wherein the reference ground truth values indicate whether individual lymph nodes are cancerous or non-cancerous.

### Claim 68 (depends on 67)

The system of claim 67, wherein the reference ground truth values indicating whether individual lymph nodes are cancerous or non-cancerous are back-calculated from the Nstage covariable values based on characteristics of the tumor or lymph nodes.

### Claim 69 (depends on 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68)

The system of any one of claims 40-68, wherein the imputation model is a random forest classifier or a LASSO classifier.

### Claim 70 (depends on 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69)

The system of any one of claims 40-69, wherein the risk model is a random forest classifier or a LASSO classifier.

### Claim 71 (depends on 70)

The system of claim 70, wherein the risk model is trained using supervised learning techniques.

### Claim 72 (depends on 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71)

The system of any one of claims 40-71, wherein the plurality of lymph nodes comprises 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, or 21 lymph nodes.

### Claim 73 (depends on 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72)

The system of any one of claims 40-72, wherein the plurality of lymph nodes comprises between 8 to 20 lymph nodes, between 10 to 19 lymph nodes, between 12 to 18 lymph nodes, or between 14 and 16 lymph nodes.

### Claim 74 (depends on 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73)

The system of any one of claims 40-73, wherein the metastatic cancer is metastatic lung cancer.

### Claim 75 (depends on 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74)

The system of any one of claims 40-74, wherein the risk model exhibits a performance of at least AUC=78%, Sensitivity=82%, or Specificity=60%.

### Claim 76 (depends on 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75)

The system of any one of claims 40-75, wherein the risk model is trained by:
applying an imputation model to generate imputed LN-level risks for a plurality of lymph nodes in training images;
selecting a set of lymph nodes, wherein the lymph nodes in the set have LN-level risks that are greater than a median risk value;
using the LN-level risks of the lymph nodes in the set of lymph nodes as reference ground truths for training the risk model, the risk model able to predict LN-level risks with a greater range of probabilities than the imputed LN-level risks predicted by the imputation model.

### Claim 77 (depends on 76)

The system of claim 76, wherein the risk model is further trained by extracting features from one or more radiomic panels; and training the risk model using at least the extracted features.

### Claim 78 (depends on 77)

The system of claim 77, wherein extracting features from two or more radiomic panels comprises:
defining a region of interest (ROI) in the one or more radiomic panels, the defined region of interest comprising a lymph node;
extracting features from the ROI in the one or more radiomic panels.

### Claim 79 (depends on 77, 78)

The system of claim 77 or 78, wherein the extracted features include one or more feature categories of first order features, shape features, gray level co-occurrence matrix 1 (GLCM) features, gray level run length matrix (GLRLM) features, gray level size zone matrix (GLSZM) features, and neighborhood gray tone difference matrix (NGTDM) features.

### Claim 80

A non-transitory computer readable medium comprising instructions that, when executed by a processor, cause the processor to:
obtain one or more images captured from the subject comprising a plurality of lymph nodes of the subject; and
predict the subject level risk of metastatic cancer by applying a risk model to extracted features of the obtained one or more images, the risk model trained to predict lymph node (LN) level risks of metastatic cancer for lymph nodes in images,
wherein the risk model is trained using at least labels derived from imputed LN-level risks of metastatic cancer for a set of lymph nodes of a reference individual, the imputed LN-level risks of the set of lymph nodes determined using at least an imputation model that discriminates between cancerous and non-cancerous lymph nodes of training images.

### Claim 81 (depends on 80)

The non-transitory computer readable medium of claim 80, wherein the instructions that cause the processor to predict the subject level risk of metastatic cancer further comprises instructions that, when executed by the processor, cause the processor to:
select one or more of the plurality of lymph nodes based on their LN-level risk of metastatic cancer predicted by the risk model; and
determine the subject level risk of metastatic cancer using the LN-level risk of metastatic cancer predicted for the one or more lymph nodes.

### Claim 82 (depends on 81)

The non-transitory computer readable medium of claim 81, wherein the instructions that cause the processor to select one or more of the plurality of lymph nodes further comprises instructions that, when executed by the processor, cause the processor to identify the lymph node with the highest probability of LN-level risk, and wherein the instructions that cause the processor to determine the subject level risk of metastatic cancer comprises instructions that, when executed by the processor, cause the processor to assign the highest probability of LN-level risk as the subject level risk of metastatic cancer.

### Claim 83 (depends on 80, 81, 82)

The non-transitory computer readable medium of any one of claims 80-82, wherein the risk model predicts LN-level risks of metastatic cancer with a greater range of probabilities than the imputed LN-level risks predicted by the imputation model.

### Claim 84 (depends on 80, 81, 82, 83)

The non-transitory computer readable medium of any one of claims 80-83, wherein the set of lymph nodes of a reference individual is selected by:
determining a median risk value; and
comparing imputed LN-level risks of the lymph nodes of the reference individual to the median risk value.

### Claim 85 (depends on 84)

The non-transitory computer readable medium of claim 84, wherein the set of lymph nodes of a reference individual is further selected by including one or more lymph nodes with LN-level risks greater than the median risk value in the set of lymph nodes.

### Claim 86 (depends on 84, 85)

The non-transitory computer readable medium of claim 84 or 85, wherein one or more lymph nodes with LN-level risks less than the median risk value in the set of lymph nodes are excluded from the set of lymph nodes.

### Claim 87 (depends on 84, 85, 86)

The non-transitory computer readable medium of any one of claims 84-86, wherein the median risk value is a median intrasubject LN-level risk of the reference individual.

### Claim 88 (depends on 80, 81, 82, 83, 84, 85, 86, 87)

The non-transitory computer readable medium of any one of claims 80-87, wherein the one or more images comprises a computed tomography (CT) image.

### Claim 89 (depends on 80, 81, 82, 83, 84, 85, 86, 87, 88)

The non-transitory computer readable medium of any one of claims 80-88, wherein the one or more images comprises are obtained from a thoracic CT scan.

### Claim 90 (depends on 80, 81, 82, 83, 84, 85, 86, 87, 88, 89)

The non-transitory computer readable medium of any one of claims 80-89, further comprising instructions that, when executed by the processor, cause the processor to determine a stage of the cancer according to the predicted subject level risk of metastatic cancer.

### Claim 91 (depends on 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90)

The non-transitory computer readable medium of any one of claims 80-90, further comprising instructions that, when executed by the processor, cause the processor to select a diagnostic or treatment for the subject according to the predicted subject level risk of metastatic cancer.

### Claim 92 (depends on 91)

The non-transitory computer readable medium of claim 91, wherein the instructions that cause the processor to select a diagnostic or treatment for the subject further comprises instructions that, when executed by the processor, cause the processor to select either surgical tumor resection or combined bronchoscopic or endoscopic diagnosis and treatment when the predicted subject level risk of metastatic cancer is below a threshold risk value.

### Claim 93 (depends on 91)

The non-transitory computer readable medium of claim 91, wherein the instructions that cause the processor to select a diagnostic or treatment for the subject further comprises instructions that, when executed by the processor, cause the processor to select additional diagnostic testing of one or both of performing a PET or PET-CT scan or performing a lymph node biopsy when the predicted subject level risk of metastatic cancer is above a threshold risk value.

### Claim 94 (depends on 91)

The non-transitory computer readable medium of claim 91, wherein the instructions that cause the processor to select a diagnostic or treatment for the subject further comprises instructions that, when executed by the processor, cause the processor to select additional diagnostic testing of a PET or PET-CT scan when the predicted subject level risk of metastatic cancer is below a threshold risk value.

### Claim 95 (depends on 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94)

The non-transitory computer readable medium of any one of claims 80-94, wherein the imputation model is trained using training images obtained from thoracic CT scans.

### Claim 96 (depends on 95)

The non-transitory computer readable medium of claim 95, wherein the training images obtained from thoracic CT scans comprise one or more lymph nodes.

### Claim 97 (depends on 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96)

The non-transitory computer readable medium of any one of claims 80-96, wherein the imputation model is trained using training images of the National Lung Screening Trial (NLST).

### Claim 98 (depends on 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96)

The non-transitory computer readable medium of any one of claims 80-96, wherein the imputation model is trained using training images of a custom dataset.

### Claim 99 (depends on 95, 96, 97, 98)

The non-transitory computer readable medium of any one of claims 95-98, wherein the imputation model is trained by generating two or more radiomic panels from at least one of the training images.

### Claim 100 (depends on 99)

The non-transitory computer readable medium of claim 99, wherein the two or more radiomic panels comprise a Laplacian of Gaussians transformation of the training image and a wavelet transform of the training image.

### Claim 101 (depends on 99, 100)

The non-transitory computer readable medium of claim 99 or 100, wherein the imputation model is further trained by extracting features from each of the two or more radiomic panels; and training the imputation model using at least the extracted features.

### Claim 102 (depends on 99, 100)

The non-transitory computer readable medium of claim 99 or 100, wherein the imputation model is further trained by:
defining a region of interest (ROI) in the radiomic panels, the defined region of interest comprising a lymph node;
extracting features from the ROI in the radiomic panels; and
training the imputation model using at least the extracted features.

### Claim 103 (depends on 101, 102)

The non-transitory computer readable medium of claim 101 or 102, wherein the extracted features include one or more feature categories of first order features, shape features, gray level co-occurrence matrix 1 (GLCM) features, gray level run length matrix (GLRLM) features, gray level size zone matrix (GLSZM) features, and neighborhood gray tone difference matrix (NGTDM) features.

### Claim 104 (depends on 95, 96, 97, 98, 99, 100, 101, 102, 103)

The non-transitory computer readable medium of any of claims 95-103, wherein the imputation model is further trained using reference ground truth values indicating a presence or absence of metastasis in training images.

### Claim 105 (depends on 104)

The non-transitory computer readable medium of claim 104, wherein the reference ground truth values are derived from Nstage covariable values indicating the stage of the cancer, wherein a cancer stage of “0” or “1” indicate an absence of metastasis and wherein a cancer stage of “2” or “3” indicate a presence of metastasis.

### Claim 106 (depends on 105)

The non-transitory computer readable medium of claim 105, wherein the reference ground truth values indicate whether individual lymph nodes are cancerous or non-cancerous.

### Claim 107 (depends on 106)

The non-transitory computer readable medium of claim 106, wherein the reference ground truth values indicating whether individual lymph nodes are cancerous or non-cancerous are back-calculated from the Nstage covariable values based on characteristics of the tumor or lymph nodes.

### Claim 108 (depends on 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 107)

The non-transitory computer readable medium of any one of claims 80-107, wherein the imputation model is a random forest classifier or a LASSO classifier.

### Claim 109 (depends on 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108)

The non-transitory computer readable medium of any one of claims 80-108, wherein the risk model is a random forest classifier or a LASSO classifier.

### Claim 110 (depends on 109)

The non-transitory computer readable medium of claim 109, wherein the risk model is trained using supervised learning techniques.

### Claim 111 (depends on 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110)

The non-transitory computer readable medium of any one of claims 80-110, wherein the plurality of lymph nodes comprises 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, or 21 lymph nodes.

### Claim 112 (depends on 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111)

The non-transitory computer readable medium of any one of claims 80-111, wherein the plurality of lymph nodes comprises between 8 to 20 lymph nodes, between 10 to 19 lymph nodes, between 12 to 18 lymph nodes, or between 14 and 16 lymph nodes.

### Claim 113 (depends on 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111, 112)

The non-transitory computer readable medium of any one of claims 80-112, wherein the metastatic cancer is metastatic lung cancer.

### Claim 114 (depends on 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111, 112, 113)

The non-transitory computer readable medium of any one of claims 80-113, wherein the risk model exhibits a performance of at least AUC=78%, Sensitivity=82%, or Specificity=60%.

### Claim 115 (depends on 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111, 112, 113, 114)

The non-transitory computer readable medium of any one of claims 80-114, wherein the risk model is trained by:
applying an imputation model to generate imputed LN-level risks for a plurality of lymph nodes in training images;
selecting a set of lymph nodes, wherein the lymph nodes in the set have LN-level risks that are greater than a median risk value;
using the LN-level risks of the lymph nodes in the set of lymph nodes as reference ground truths for training the risk model, the risk model able to predict LN-level risks with a greater range of probabilities than the imputed LN-level risks predicted by the imputation model.

### Claim 116 (depends on 115)

The non-transitory computer readable medium of claim 115, wherein the risk model is further trained by extracting features from one or more radiomic panels; and training the risk model using at least the extracted features.

### Claim 117 (depends on 116)

The non-transitory computer readable medium of claim 116, wherein extracting features from two or more radiomic panels comprises:
defining a region of interest (ROI) in the one or more radiomic panels, the defined region of interest comprising a lymph node;
extracting features from the ROI in the one or more radiomic panels.

### Claim 118 (depends on 115, 116)

The non-transitory computer readable medium of claim 115 or 116, wherein the extracted features include one or more feature categories of first order features, shape features, gray level co-occurrence matrix 1 (GLCM) features, gray level run length matrix (GLRLM) features, gray level size zone matrix (GLSZM) features, and neighborhood gray tone difference matrix (NGTDM) features.
