Abstract 3048: Deep learning to predict survival prognosis for patients with non-small cell lung cancer using images and clinical data

Goal

We present a deep learning framework to predict survival prognosis of patients with non-small cell lung cancer using CT images and clinical data developed and validated in multi-institutional cohorts.

Methods

We developed 3D convolutional neural networks (CNNs) using CT images, tumor nodule segmentation, clinical data (i.e. age, sex, histology, and stage) and survival times as input to predict survival of non-small cell lung cancer. Both the CT images and the segmentations were used as inputs for the CNN, the latter to localize the volumes-of-interests (VOIs) in the CT scans. We used three cohorts to evaluate our framework: Moffitt (n=186), Maastro (n=311), and Stanford (n=130). We investigated pre-training with the LIDC-IDRI dataset (n=1010) and compared deep learning performance with 1674 manually-curated radiomics features and clinical data. We used two ways to evaluate the models. First, we split the three cohorts into training and test sets by selecting two of the cohorts for training and the remaining for testing. Secondly, we combined all cohorts and performed a 10-fold stratified cross-validation. We used the Concordance Index (CI) to estimate model performance. Next, we evaluated priming whereby we use 10% of the validation cohort during training and test on the remaining patients in the test cohort. To ensure that the test set was identical for both priming and non-priming experiments, we discarded these samples for the non-priming tests.

Results

Table 1 shows the results of the two evaluation strategies. We observed consistently high CIs for all three cohorts. LIDC pretraining and priming yielded the highest CIs as compared to radiomic and clinical-only pipelines. For the 10-fold cross-validation experiment, our pipeline achieved an average CI=0.64.

Conclusion

Our study highlights the promising results of our pipeline to predict survival prognosis across three cohorts without the need to define radiomics features.

Train on two cohorts, test on 3rd cohortpretrain on LIDC-IDRItest w/o primingtest w/ primingtest w/o priming (only radiomics features / clinical)test w/ priming (only radiomics features / clinical)train on Moffitt + Maastro, test on Stanfordyes0.620.630.52 (radiomics only)0.53 (radiomics only)train on Moffitt + Maastro, test on Stanfordno0.610.620.57 (clinical only)0.57 (clinical only)train on Moffitt + Stanford, test on Maastroyes0.60.60.55 (radiomics only)0.55 (radiomics only)train on Moffitt + Stanford, test on Maastrono0.580.60.56 (clinical only)0.56 (clinical only)train on Stanford + Maastro, test on Moffittyes0.580.60.53 (radiomics only)0.53 (radiomics only)train on Stanford + Maastro, test on Moffittno0.60.570.56 (clinical only)0.56 (clinical only)

Citation Format: Edward H. Lee, Mu Zhou, Noah Gamboa, Kevin Brennan, Haruka Itakura, Viswam Nair, Sandy Napel, Simon Wong, Olivier Gevaert. Deep learning to predict survival prognosis for patients with non-small cell lung cancer using images and clinical data [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2018; 2018 Apr 14-18; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2018;78(13 Suppl):Abstract nr 3048.

GFPT2-Expressing Cancer-Associated Fibroblasts Mediate Metabolic Reprogramming in Human Lung Adenocarcinoma

Metabolic reprogramming of the tumor microenvironment is recognized as a cancer hallmark. To identify new molecular processes associated with tumor metabolism, we analyzed the transcriptome of bulk and flow-sorted human primary non-small cell lung cancer (NSCLC) together with ¹⁸FDG-PET scans, which provide a clinical measure of glucose uptake. Tumors with higher glucose uptake were functionally enriched for molecular processes associated with invasion in adenocarcinoma and cell growth in squamous cell carcinoma (SCC). Next, we identified genes correlated to glucose uptake that were predominately overexpressed in a single cell-type comprising the tumor microenvironment. For SCC, most of these genes were expressed by malignant cells, whereas in adenocarcinoma, they were predominately expressed by stromal cells, particularly cancer-associated fibroblasts (CAF). Among these adenocarcinoma genes correlated to glucose uptake, we focused on glutamine-fructose-6-phosphate transaminase 2 (GFPT2), which codes for the glutamine-fructose-6-phosphate aminotransferase 2 (GFAT2), a rate-limiting enzyme of the hexosamine biosynthesis pathway (HBP), which is responsible for glycosylation. GFPT2 was predictive of glucose uptake independent of GLUT1, the primary glucose transporter, and was prognostically significant at both gene and protein level. We confirmed that normal fibroblasts transformed to CAF-like cells, following TGFβ treatment, upregulated HBP genes, including GFPT2, with less change in genes driving glycolysis, pentose phosphate pathway, and TCA cycle. Our work provides new evidence of histology-specific tumor stromal properties associated with glucose uptake in NSCLC and identifies GFPT2 as a critical regulator of tumor metabolic reprogramming in adenocarcinoma.Significance: These findings implicate the hexosamine biosynthesis pathway as a potential new therapeutic target in lung adenocarcinoma. Cancer Res; 78(13); 3445-57. ©2018 AACR.

A Rapid Segmentation-Insensitive "Digital Biopsy" Method for Radiomic Feature Extraction: Method and Pilot Study Using CT Images of Non-Small Cell Lung Cancer

Quantitative imaging approaches compute features within images' regions of interest. Segmentation is rarely completely automatic, requiring time-consuming editing by experts. We propose a new paradigm, called “digital biopsy,” that allows for the collection of intensity- and texture-based features from these regions at least 1 order of magnitude faster than the current manual or semiautomated methods. A radiologist reviewed automated segmentations of lung nodules from 100 preoperative volume computed tomography scans of patients with non–small cell lung cancer, and manually adjusted the nodule boundaries in each section, to be used as a reference standard, requiring up to 45 minutes per nodule. We also asked a different expert to generate a digital biopsy for each patient using a paintbrush tool to paint a contiguous region of each tumor over multiple cross-sections, a procedure that required an average of <3 minutes per nodule. We simulated additional digital biopsies using morphological procedures. Finally, we compared the features extracted from these digital biopsies with our reference standard using intraclass correlation coefficient (ICC) to characterize robustness. Comparing the reference standard segmentations to our digital biopsies, we found that 84/94 features had an ICC >0.7; comparing erosions and dilations, using a sphere of 1.5-mm radius, of our digital biopsies to the reference standard segmentations resulted in 41/94 and 53/94 features, respectively, with ICCs >0.7. We conclude that many intensity- and texture-based features remain consistent between the reference standard and our method while substantially reducing the amount of operator time required.

Belatacept Conversion in an HIV-Positive Kidney Transplant Recipient With Prolonged Delayed Graft Function

We report an HIV-positive renal transplant recipient with delayed graft function who was converted from tacrolimus to belatacept in an attempt to improve renal function. The patient had kidney biopsies at 4 and 8 weeks posttransplant that revealed acute tubular necrosis and mild fibrosis. After 14 weeks of delayed function, belatacept was initiated and tacrolimus was weaned off. Shortly after discontinuing tacrolimus, renal function began to improve. The patient was able to discontinue dialysis 21 weeks posttransplant. HIV viral load was undetectable at last follow-up. To our knowledge, this is the first report of belatacept use in a patient with HIV.

Chronobiologic Considerations in Pharmacologic and Toxicologic Studies: Fact or Fancy

Previous studies showing circadian variation in pharmacologic response to drugs in man and experimental animals have been reviewed. The chronobiologic response is discussed in terms of the pharmacologic factors that determine drug action. The importance of internal markers in chronopharmacologic studies is emphasized and is illustrated by our studies with the antihypersive drugs. Identification of chronobiologic variations in the therapeutic and toxic response to drugs, when present, along with an understanding of their mechanisms, would permit the design of optimal therapeutic regimens for these drugs.

Cryopreservation of specialized chicken lines using cultured primordial germ cells

Biosecurity and sustainability in poultry production requires reliable germplasm conservation. Germplasm conservation in poultry is more challenging in comparison to other livestock species. Embryo cryopreservation is not feasible for egg-laying animals, and chicken semen conservation has variable success for different chicken breeds. A potential solution is the cryopreservation of the committed diploid stem cell precursors to the gametes, the primordial germ cells ( PGCS: ). Primordial germ cells are the lineage-restricted cells found at early embryonic stages in birds and form the sperm and eggs. We demonstrate here, using flocks of partially inbred, lower-fertility, major histocompatibility complex- ( MHC-: ) restricted lines of chicken, that we can easily derive and cryopreserve a sufficient number of independent lines of male and female PGCs that would be sufficient to reconstitute a poultry breed. We demonstrate that germ-line transmission can be attained from these PGCs using a commercial layer line of chickens as a surrogate host. This research is a major step in developing and demonstrating that cryopreserved PGCs could be used for the biobanking of specialized flocks of birds used in research settings. The prospective application of this technology to poultry production will further increase sustainability to meet current and future production needs.

A dominant adenocarcinoma with multifocal ground glass lesions does not behave as advanced disease

BACKGROUND

Invasive lung adenocarcinomas increasingly present with synchronous, multifocal, in situ lesions that appear as ground glass opacities (GGOs). The optimal approach in this circumstance (often nonsmokers) remains unclear. We evaluated a general strategy of anatomic resection of the dominant tumor (DT) and wedge resection of accessible ipsilateral GGOs.

METHODS

This is a retrospective review of 39 patients with suspected multifocal in situ adenocarcinomas and 1 DT in a predominantly Caucasian population. Mean follow-up is 30.7 months.

RESULTS

Forty-nine percent of patients had no or minimal smoking history; 21% were Asian. The resected DT was pathologically "bronchioloalveolar carcinoma" (26%), minimally invasive adenocarcinoma (5%), adenocarcinoma with bronchioloalveolar features (41%), or moderate well-differentiated adenocarcinoma (28%). The p stage of the DT was IA in 20, IB in 15, and IIA in 4, with mean diameter of 2.6 cm. Thirty-two patients (82%) underwent anatomic resection of the DT; 7 (18%) underwent wedge resection. The mean number of GGOs present initially was 2.7 (range, 1 to 7) with a 5.2-mm mean diameter. An unresected nodule increased in size during follow-up in only 9 patients (23%). The mean diameter growth among these was 3.2 mm, with mean doubling time of 49 months. New GGOs (range, 1 to 8) developed in 16 patients (41%), all of which remained at 7 mm or less. Distant metastasis developed in 2 patients (5.2%); only 1 patient has required intervention for progression of a GGO. The overall survival is 100%.

CONCLUSIONS

Patients with limited, multifocal, in situ adenocarcinomas and a clinical N0 DT enjoy prolonged survival with generally anatomic resection of the DT and wedge resection of accessible GGOs. These patients should not be considered to harbor T4 or M1a disease.