Radiomics of Lung Screening
cancer within the NLST screening and follow-up periods?
This project will analyze the Moffitt dataset with other screening datasets who underwent initial and two annual NLSTscreenings and who did not develop lung cancer within the screening or follow-up periods. Features will be developed and extracted from left and right lung fields of all patients and algorithms will be developed to analyze the longitudinal progression of individual features. These analyses will be able to quantitatively describe the natural range of all features as well as their reproducibility. The impact of these studies will be identification of a panel of features that characterize patients who have “lower risk” wherein the screening interval could be lengthened to, e.g. 5-10 years.
Aim 1. Develop robust methods for segmenting lung fields of normal patients. Although we have developed methods for segmenting lung cancers, we have not yet optimized segmentation of the entire lungs. Our newly-developed multi-seed point approach will be used.
Aim 2. Collate and validate Moffitt dataset. The data from the NLST trial was prospectively structured to produce high levels of compliance and follow up. However, to avoid mining of unreliable data, considerable effort will be expended to collate and verify the CT and clinical data available in the Moffitt dataset. Inclusion criteria include (a) the presence of at least three longitudinal scans; (b) extractable radiology reports; and (c) sufficient clinical data to classify participants as true negative, equivocal negative or true positive. These data will be tabulated and entered into the database developed by the core.
Aim 3. Expand Feature set. To date, we have expanded the feature set for lung cancers to over 320 2- and 3-Dimensional elements. These are specific for lung lesions and may not capture the additional features that are relevant to normal lung fields. The major feature to be developed in this aim will be quantitative descriptors of the normal bronchiolar tree and terminal bronchiolar architecture.
Aim 4. Extract and analyze data from the Moffitt data set. Features will be extracted from all validated patients in the Moffitt dataset and deposited into the database developed by the core. Both L and R lung fields will be extracted from at least three longitudinal scans from each participant. Data will be parsed for unequivocally negative normal participants and these will be used to develop a baseline normal set of features. These will be used to determine which features do not exhibit progression throughout the screening and follow-up periods. In addition to Radiomic variables, feature selection will consider other potentially confounding variables archived by the Core, including demographic variables (age, race, sex), physiologic variables (BMI, BP, airway obstruction), and respiratory exposure variables (smoking, asbestos exposures). Patterns of selected features will be compared to participants who were marginally normal, i.e. those with moderate to severe airway obstruction (COPD) and severe inflammatory residua.
Aim 5. Extract Quantitative descriptors in Pathology Images:
We would like use the digital pathology images (H&E or others) to extract quantitative descriptors at the individual cell level and quantitatively describe the surrounding characteristics. These metrics across the population would be additional information to aid the CT extracted features in the model to predict and or prognosticate patient characteristics to become malignant over time. Pathology features will also be related to CT extracted features both to understand the relationship at two different scales.
Moffitt Cancer Center:
Robert Gillies, Ph. D (PI)
Robert Gatenby, M.D, Ph.D
Steven Estrich, Ph.D
Greg Bloom, Ph.D
Yoga Balagurunathan, Ph.D
Yuhua Gu, Ph.D
Moffitt NLST Collaborators:
Melvyn Tockman, M.D, Ph.D
Lynn Coppage, M.D, Ph.D
University of South Florida:
Lawrence Hall, Ph.D
Dmitry Goldgoff, Ph.D
Mastro Cancer Center, Netherlands,
Internal-transfer Weighting of Multi-task Learning for Lung Cancer Detection.
Yang Y, Gao R, Tang Y, Antic SL, Deppen S, Huo Y, Sandler KL, Massion PP, Landman BA
Proc SPIE Int Soc Opt Eng. 2020; Volume 11313 PUBMED
Semi-supervised Machine Learning with MixMatch and Equivalence Classes.
Hansen CB, Nath V, Gao R, Bermudez C, Huo Y, Sandler KL, Massion PP, Blume JD, Lasko TA, Landman BA
Lect Notes Monogr Ser. 2020; Volume 12446: Pages 112-121 PUBMED
Time-distanced gates in long short-term memory networks.
Gao R, Tang Y, Xu K, Huo Y, Bao S, Antic SL, Epstein ES, Deppen S, Paulson AB, Sandler KL, Massion PP, Landman BA
Med Image Anal. 2020 Oct; Volume 65: Pages 101785 PUBMED
Convolutional Neural Network ensembles for accurate lung nodule malignancy prediction 2 years in the future.
Paul R, Schabath M, Gillies R, Hall L, Goldgof D
Comput Biol Med. 2020 Jul; Volume 122: Pages 103882 PUBMED
Multi-path x-D Recurrent Neural Networks for Collaborative Image Classification.
Gao R, Huo Y, Bao S, Tang Y, Antic SL, Epstein ES, Deppen S, Paulson AB, Sandler KL, Massion PP, Landman BA
Neurocomputing. 2020 Jul 15; Volume 397: Pages 48-59 PUBMED
Lung Nodule Sizes Are Encoded When Scaling CT Image for CNN's.
Cherezov D, Paul R, Fetisov N, Gillies RJ, Schabath MB, Goldgof DB, Hall LO
Tomography. 2020 Jun; Volume 6 (Issue 2): Pages 209-215 PUBMED
Peritumoral and intratumoral radiomic features predict survival outcomes among patients diagnosed in lung cancer screening.
Pérez-Morales J, Tunali I, Stringfield O, Eschrich SA, Balagurunathan Y, Gillies RJ, Schabath MB
Sci Rep. 2020 Jun 29; Volume 10 (Issue 1): Pages 10528 PUBMED
Hybrid models for lung nodule malignancy prediction utilizing convolutional neural network ensembles and clinical data.
Paul R, Schabath MB, Gillies R, Hall LO, Goldgof DB
J Med Imaging (Bellingham). 2020 Mar; Volume 7 (Issue 2): Pages 024502 PUBMED
Multi-window CT based Radiomic signatures in differentiating indolent versus aggressive lung cancers in the National Lung Screening Trial: a retrospective study.
Lu H, Mu W, Balagurunathan Y, Qi J, Abdalah MA, Garcia AL, Ye Z, Gillies RJ, Schabath MB
Cancer Imaging. 2019 Jun 28; Volume 19 (Issue 1): Pages 45 PUBMED
Quantitative Imaging features Improve Discrimination of Malignancy in Pulmonary nodules.
Balagurunathan Y, Schabath MB, Wang H, Liu Y, Gillies RJ
Sci Rep. 2019 Jun 12; Volume 9 (Issue 1): Pages 8528 PUBMED
Automated pulmonary nodule CT image characterization in lung cancer screening.
Reeves AP, Xie Y, Jirapatnakul A
Int J Comput Assist Radiol Surg. 2015 Jun; Volume [Epub ahead of print]: Pages [Epub ahead of print] PUBMED