Abbreviated Name

LTP2 Results Analysis-Schabath-Moffitt-2024

Lead Investigator

Schabath, Matthew — H. Lee Moffitt Cancer Center & Research Institute, Inc.

Coordinating Investigator

Zheng, Yingye — Fred Hutchinson Cancer Center

Involved Investigators

• Schabath, Matthew — H. Lee Moffitt Cancer Center & Research Institute, Inc.
• Zheng, Yingye — Fred Hutchinson Cancer Center

Abstract

The CT images that will be curated and archived will be subjected to our in-house radiomics pipeline detailed in Figure 7.4 below and available for future use as part of the reference set. From high-quality images, the nodule/tumor is identified and semi-automatically segmented. This segmented volume is rendered from the rest of the image. From the rendered volume and its surrounding stroma, quantitative radiomic features are extracted and then databased. The databased features are merged with patient data and then analyzed. Radiomic features (Semantic and Agnostic) will be extracted from the baseline and follow-up CT images from the participants. Semantic features are those that are commonly used in the radiology lexicon to describe regions of interest. Each feature is rated as ordinal (0-5) or binary (present vs. not present) and work is ongoing to automate the extraction of these features. Agnostic features capture lesion heterogeneity through quantitative descriptors. Descriptively, agnostic features include shape, size, voxel intensity, texture, and wavelets. We can extract nearly 500 agnostic features that divided into first-, second-, and higher-order outputs. First-order features describe the distribution of values of individual voxels (volume + pixel) without concern for spatial relationships. Shape and size-based features consist of the larger-scale form of the nodule (or tumor) such as sphericity, compactness, surface area, surface to volume ratio, and volume. Nodule/tumor voxel intensities describe the histogram of intensities within the CT image through basic metrics such as mean, median, maximum, minimum and uniformity or randomness (entropy) of the intensities in the image, as well as the skewness (asymmetry) and kurtosis (flatness) of the values. Second-order features are described as “texture” features; they describe statistical interrelationships between voxels with similar (or dissimilar) contrast values.

Aims

The CT images that will be curated and archived will be subjected to our in-house radiomics pipeline detailed in Figure 7.4 below and available for future use as part of the reference set. From high-quality images, the nodule/tumor is identified and semi-automatically segmented. This segmented volume is rendered from the rest of the image. From the rendered volume and its surrounding stroma, quantitative radiomic features are extracted and then databased. The databased features are merged with patient data and then analyzed. Radiomic features (Semantic and Agnostic) will be extracted from the baseline and follow-up CT images from the participants. Semantic features are those that are commonly used in the radiology lexicon to describe regions of interest. Each feature is rated as ordinal (0-5) or binary (present vs. not present) and work is ongoing to automate the extraction of these features. Agnostic features capture lesion heterogeneity through quantitative descriptors. Descriptively, agnostic features include shape, size, voxel intensity, texture, and wavelets. We can extract nearly 500 agnostic features that divided into first-, second-, and higher-order outputs. First-order features describe the distribution of values of individual voxels (volume + pixel) without concern for spatial relationships. Shape and size-based features consist of the larger-scale form of the nodule (or tumor) such as sphericity, compactness, surface area, surface to volume ratio, and volume. Nodule/tumor voxel intensities describe the histogram of intensities within the CT image through basic metrics such as mean, median, maximum, minimum and uniformity or randomness (entropy) of the intensities in the image, as well as the skewness (asymmetry) and kurtosis (flatness) of the values. Second-order features are described as “texture” features; they describe statistical interrelationships between voxels with similar (or dissimilar) contrast values.

Analytic Method

See LTP2 Protocol

Outcome

The CT images that will be curated and archived will be subjected to our in-house radiomics pipeline detailed in Figure 7.4 below and available for future use as part of the reference set. From high-quality images, the nodule/tumor is identified and semi-automatically segmented. This segmented volume is rendered from the rest of the image. From the rendered volume and its surrounding stroma, quantitative radiomic features are extracted and then databased. The databased features are merged with patient data and then analyzed. Radiomic features (Semantic and Agnostic) will be extracted from the baseline and follow-up CT images from the participants. Semantic features are those that are commonly used in the radiology lexicon to describe regions of interest. Each feature is rated as ordinal (0-5) or binary (present vs. not present) and work is ongoing to automate the extraction of these features. Agnostic features capture lesion heterogeneity through quantitative descriptors. Descriptively, agnostic features include shape, size, voxel intensity, texture, and wavelets. We can extract nearly 500 agnostic features that divided into first-, second-, and higher-order outputs. First-order features describe the distribution of values of individual voxels (volume + pixel) without concern for spatial relationships. Shape and size-based features consist of the larger-scale form of the nodule (or tumor) such as sphericity, compactness, surface area, surface to volume ratio, and volume. Nodule/tumor voxel intensities describe the histogram of intensities within the CT image through basic metrics such as mean, median, maximum, minimum and uniformity or randomness (entropy) of the intensities in the image, as well as the skewness (asymmetry) and kurtosis (flatness) of the values. Second-order features are described as “texture” features; they describe statistical interrelationships between voxels with similar (or dissimilar) contrast values.

Publications

• No publications available at this time for this protocol.

Biomarkers

• No biomarkers available at this time for this protocol.

Data Collections

• No data collections available at this time for this protocol.

Protocol ID

557

Protocol Type

Reference Set

Fields of Research

• Other

Collaborative Group

Lung and Upper Aerodigestive Cancers Research Group

Cancer Types

• Malignant neoplasm of bronchus and lung

Phased Status

2