Improving differential diagnosis of pulmonary large cell neuroendocrine carcinoma and small cell lung cancer via a transcriptomic, biological pathway-based machine learning model

Hierarchical identification of a transcriptional panel for the histological diagnosis of lung neuroendocrine tumors

Background: Lung cancer is a complex disease composed of neuroendocrine (NE) and non-NE tumors. Accurate diagnosis of lung cancer is essential in guiding therapeutic management. Several transcriptional signatures have been reported to distinguish between adenocarcinoma (ADC) and squamous cell carcinoma (SCC) belonging to non-NE tumors. This study aims to identify a transcriptional panel that could distinguish the histological subtypes of NE tumors to complement the morphology-based classification of an individual.

Methods: A public dataset with NE subtypes, including 21 small-cell lung cancer (SCLC), 56 large-cell NE carcinomas (LCNECs), and 24 carcinoids (CARCIs), and non-NE subtypes, including 85 ADC and 61 SCC, was used as a training set. In the training set, consensus clustering was first used to filter out the samples whose expression patterns disagreed with their histological subtypes. Then, a rank-based method was proposed to develop a panel of transcriptional signatures for determining the NE subtype for an individual, based on the within-sample relative gene expression orderings of gene pairs. Twenty-three public datasets with a total of 3,454 samples, which were derived from fresh-frozen, formalin-fixed paraffin-embedded, biopsies, and single cells, were used for validation. Clinical feasibility was tested in 10 SCLC biopsy specimens collected from cancer hospitals via bronchoscopy.

Results: The NEsubtype-panel was composed of three signatures that could distinguish NE from non-NE, CARCI from non-CARCI, and SCLC from LCNEC step by step and ultimately determine the histological subtype for each NE sample. The three signatures achieved high average concordance rates with 97.31%, 98.11%, and 90.63%, respectively, in the 23 public validation datasets. It is worth noting that the 10 clinic-derived SCLC samples diagnosed via immunohistochemical staining were also accurately predicted by the NEsubtype-panel. Furthermore, the subtype-specific gene expression patterns and survival analyses provided evidence for the rationality of the reclassification by the NEsubtype-panel.

Conclusion: The rank-based NEsubtype-panel could accurately distinguish lung NE from non-NE tumors and determine NE subtypes even in clinically challenging samples (such as biopsy). The panel together with our previously reported signature (KRT5-AGR2) for SCC and ADC would be an auxiliary test for the histological diagnosis of lung cancer.

Classifying Pulmonary and Urinary High-grade Neuroendocrine Carcinoma by CK7 Immunohistochemistry

High-grade neuroendocrine carcinoma (HGNEC) is subclassified into small cell carcinoma (SmCC) and large cell neuroendocrine carcinoma (LCNEC). Although both are clinically aggressive, the SmCC and LCNEC need to have different treatment strategies, and accurate pathologic diagnosis is challenging. We studied a large retrospective cohort (186 cases) of HGNEC of bladder and lung to investigate the abundance of cytokeratin (CK) 7 expression and staining pattern in SmCC and LCNEC. Overall, the pulmonary and urinary HGNEC exhibited several different CK7 staining patterns, including negative staining (n=28), dot-like staining (n=73), partial membranous staining (n=26), and complete membranous staining (n=60). Overall, 88.9% (44/49) of pulmonary SmCC and 88.0% (44/50) of urinary SmCC showed negative or dot-like patterns for CK7, while 90.8% (59/65) of pulmonary LCNEC and 72.7% (16/22) of urinary LCNEC showed partial or complete membranous patterns for CK7 (χ 2 =105.05, P <0.0001). The distinct staining patterns were also present in those mixed SmCC and LCNEC. In addition, the specimen types or fixation did not affect CK7 staining patterns. In conclusion, CK7 has a high differential value for SmCC and LCNEC and could help guide personalized treatment for patients.

Deep Learning Facilitates Distinguishing Histologic Subtypes of Pulmonary Neuroendocrine Tumors on Digital Whole-Slide Images

Simple Summary

Challenges persist in diagnosing pulmonary neuroendocrine tumors. Our case study shows that deep learning combined with convolutional neural networks has the potential to assist in the diagnosis of pulmonary neuroendocrine tumors from digital whole-slide images.

Abstract

The histological distinction of lung neuroendocrine carcinoma, including small cell lung carcinoma (SCLC), large cell neuroendocrine carcinoma (LCNEC) and atypical carcinoid (AC), can be challenging in some cases, while bearing prognostic and therapeutic significance. To assist pathologists with the differentiation of histologic subtyping, we applied a deep learning classifier equipped with a convolutional neural network (CNN) to recognize lung neuroendocrine neoplasms. Slides of primary lung SCLC, LCNEC and AC were obtained from the Laboratory of Clinical and Experimental Pathology (University Hospital Nice, France). Three thoracic pathologists blindly established gold standard diagnoses. The HALO-AI module (Indica Labs, UK) trained with 18,752 image tiles extracted from 60 slides (SCLC = 20, LCNEC = 20, AC = 20 cases) was then tested on 90 slides (SCLC = 26, LCNEC = 22, AC = 13 and combined SCLC with LCNEC = 4 cases; NSCLC = 25 cases) by F1-score and accuracy. A HALO-AI correct area distribution (AD) cutoff of 50% or more was required to credit the CNN with the correct diagnosis. The tumor maps were false colored and displayed side by side to original hematoxylin and eosin slides with superimposed pathologist annotations. The trained HALO-AI yielded a mean F1-score of 0.99 (95% CI, 0.939–0.999) on the testing set. Our CNN model, providing further larger validation, has the potential to work side by side with the pathologist to accurately differentiate between the different lung neuroendocrine carcinoma in challenging cases.