Faculty Bibliography

Visual inspection of histopathology slides is one of the main methods used by pathologists to assess the stage, type and subtype of lung tumors. Adenocarcinoma (LUAD) and squamous cell carcinoma (LUSC) are the most prevalent subtypes of lung cancer, and their distinction requires visual inspection by an experienced pathologist. In this study, we trained a deep convolutional neural network (inception v3) on whole-slide images obtained from The Cancer Genome Atlas to accurately and automatically classify them into LUAD, LUSC or normal lung tissue. The performance of our method is comparable to that of pathologists, with an average area under the curve (AUC) of 0.97. Our model was validated on independent datasets of frozen tissues, formalin-fixed paraffin-embedded tissues and biopsies. Furthermore, we trained the network to predict the ten most commonly mutated genes in LUAD. We found that six of them-STK11, EGFR, FAT1, SETBP1, KRAS and TP53-can be predicted from pathology images, with AUCs from 0.733 to 0.856 as measured on a held-out population. These findings suggest that deep-learning models can assist pathologists in the detection of cancer subtype or gene mutations. Our approach can be applied to any cancer type, and the code is available at https://github.com/ncoudray/DeepPATH .

Visual inspection of histopathology slides is one of the main methods used by pathologists to assess the stage, type and subtype of lung tumors. Adenocarcinoma (LUAD) and squamous cell carcinoma (LUSC) are the most prevalent subtypes of lung cancer, and their distinction requires visual inspection by an experienced pathologist. In this study, we trained a deep convolutional neural network (inception v3) on whole-slide images obtained from The Cancer Genome Atlas to accurately and automatically classify them into LUAD, LUSC or normal lung tissue. The performance of our method is comparable to that of pathologists, with an average area under the curve (AUC) of 0.97. Our model was validated on independent datasets of frozen tissues, formalin-fixed paraffin-embedded tissues and biopsies. Furthermore, we trained the network to predict the ten most commonly mutated genes in LUAD. We found that six of them-STK11, EGFR, FAT1, SETBP1, KRAS and TP53-can be predicted from pathology images, with AUCs from 0.733 to 0.856 as measured on a held-out population. These findings suggest that deep-learning models can assist pathologists in the detection of cancer subtype or gene mutations. Our approach can be applied to any cancer type, and the code is available at https://github.com/ncoudray/DeepPATH .

OBJECTIVES/OBJECTIVE:The Blueprint (BP) Programmed Death Ligand 1 (PD-L1) Immunohistochemistry Comparability Project is a pivotal academic/professional society and industrial collaboration to assess the feasibility of harmonizing the clinical use of five independently developed commercial PD-L1 immunohistochemistry assays. The goal of BP phase 2 (BP2) was to validate the results obtained in BP phase 1 by using real-world clinical lung cancer samples. METHODS:BP2 were conducted using 81 lung cancer specimens of various histological and sample types, stained with all five trial-validated PD-L1 assays (22C3, 28-8, SP142, SP263, and 73-10); the slides were evaluated by an international panel of pathologists. BP2 also assessed the reliability of PD-L1 scoring by using digital images, and samples prepared for cytological examination. PD-L1 expression was assessed for percentage (tumor proportional score) of tumor cell (TC) and immune cell areas showing PD-L1 staining, with TCs scored continuously or categorically with the cutoffs used in checkpoint inhibitor trials. RESULTS:The BP2 results showed highly comparable staining by the 22C3, 28-8 and SP263 assays; less sensitivity with the SP142 assay; and higher sensitivity with the 73-10 assay to detect PD-L1 expression on TCs. Glass slide and digital image scorings were highly concordant (Pearson correlation >0.96). There was very strong reliability among pathologists in TC PD-L1 scoring with all assays (overall intraclass correlation coefficient [ICC] = 0.86-0.93), poor reliability in IC PD-L1 scoring (overall ICC = 0.18-0.19), and good agreement in assessing PD-L1 status on cytological cell block materials (ICC = 0.78-0.85). CONCLUSION/CONCLUSIONS:BP2 consolidates the analytical evidence for interchangeability of the 22C3, 28-8, and SP263 assays and lower sensitivity of the SP142 assay for determining tumor proportion score on TCs and demonstrates greater sensitivity of the 73-10 assay compared with that of the other assays.

The landscape of care for early stage non-small cell lung cancer continues to evolve. While some of the developments do not seem as dramatic as what has occurred in advanced disease in recent years, there is a continuous improvement in our ability to diagnose disease earlier and more accurately. We have an increased understanding of the diversity of early stage disease and how to better tailor treatments to make them more tolerable without impacting efficacy. The International Association for the Study of Lung Cancer and the Journal of Thoracic Oncology publishes this annual update to help readers keep pace with these important developments. Experts in the care of early stage lung cancer patients have provided focused updates across multiple areas including screening, pathology, staging, surgical techniques and novel technologies, adjuvant therapy, radiotherapy, surveillance, disparities, and quality of life. The source for information includes large academic meetings, the published literature, or novel unpublished data from other international oncology assemblies.

Biospecimen quality can vary depending on many pre- and post-collection variables. In this study, we consider a natural disaster as a post-collection variable that may have compromised the quality of frozen tissue specimens. To investigate this possible link, we compared the quality of nucleic acids, the level of antigenicity, and the preservation of histology from frozen specimens collected before and after the power outage caused by Hurricane Sandy. To analyze nucleic acid quality, we extracted both DNA and RNA and performed capillary electrophoresis to compare the quality and concentrations of the nucleic acids. To compare antigenicity, frozen sections were cut and immunostained for thyroid transcription factor 1 (TTF-1), a nuclear transcription protein commonly used as a diagnostic biomarker for multiple cancer types, including thyroid and lung cancers. Positive expression of TTF-1, as noted by homogenous nuclear staining, would demonstrate that the TTF-1 proteins could still bind antibodies and, therefore, that these proteins were not significantly degraded. Furthermore, representative frozen sections stained with hematoxylin and eosin were also assessed qualitatively by a trained pathologist to examine any possible histologic aberrations. Due to the similar quality of the tissue samples collected before and after the storm, Hurricane Sandy had no discernable effect on the quality of frozen specimens, and these specimens exposed to the natural disaster are still valuable research tools.

Multiple tumor nodules (MTNs) are seen with increasing frequency in clinical practice. Based on the 2015 WHO classification of lung tumors, we assessed the reproducibility of the comprehensive histologic assessment (CHA) to distinguish second primary lung cancers (SPLC) from intrapulmonary metastases (IPM), looking for the most distinctive histological features. An international panel of lung pathologists reviewed a scanned sequential cohort of 126 tumors from 48 patients, recorded an agreed set of histologic features, including tumor typing and predominant pattern of adenocarcinoma, thereby opining whether the case was SPLC, IPM or a combination. Cohen's Kappa statistics of 0.60 on overall assessment of SPLC or IPM indicated a good agreement. Likewise, there was good agreement (0.64 Kappa score, p<0.0001) between WHO histological pattern in individual cases and SPLC or IPM status but proportions diversified for histology and SPLC or IPM status (McNemar's test, p<0.0001). The strongest associations for distinguishing between SPLC and IM were observed for nuclear pleomorphism, cell size, acinus formation, nucleolar size, mitotic rate, nuclear inclusions, intra-alveolar clusters and necrosis. Conversely, lymphocytosis, mucin content, lepidic growth, vascular invasion, macrophage response, clear cell change, acute inflammation keratinization and emperipolesis did not reach significance with tumor extent. CHA is recommended for distinguishing SPLC from IPM, with good reproducibility among lung pathologists. In addition to main histologic type and predominant patterns of histologic subtypes, nuclear pleomorphism, cell size, acinus formation, nucleolar size, and mitotic rate strongly correlate with p staging status.

CONTEXT/BACKGROUND:- Endobronchial ultrasound-guided transbronchial needle aspiration (EBUS-TBNA) has emerged as a very useful tool in the field of diagnostic respiratory cytology. Rapid on-site evaluation (ROSE) of EBUS-TBNA not only has the potential to improve diagnostic yield of the procedure but also to triage samples for predictive molecular testing to guide personalized treatments for lung cancer. OBJECTIVE:- To provide an overview of the current status of the literature regarding ROSE of EBUS-TBNA in the diagnosis of lung cancer. DATA SOURCES/METHODS:- An electronic literature search in PubMed and Google databases was performed using the following key words: cytology, lung cancer, on-site evaluation, rapid on-site evaluation, and ROSE EBUS-TBNA. Only articles published in English were included in this review. CONCLUSIONS:- Rapid on-site evaluation can ensure that the targeted lesion is being sampled and can enable appropriate specimen triage. If available, it should be used with EBUS-TBNA in the diagnosis of lung cancer because it can minimize repeat procedures for additional desired testing (ie, molecular studies). Some studies have shown that ROSE does not adversely affect the number of aspirations, total procedure time of EBUS-TBNA, or the rate of postprocedure complications; it is also helpful in providing a preliminary diagnosis that can reduce the number of additional invasive procedures, such as mediastinoscopy. As EBUS technology continues to evolve, our knowledge of the role of ROSE in EBUS-TBNA for the diagnosis of lung cancer will also continue to grow and evolve.