Reliable entity subtyping is paramount for therapy stratification in lung cancer. Morphological evaluation remains the basis for entity subtyping and directs the application of additional methods such as immunohistochemistry (IHC). The decision of whether to perform IHC for subtyping is subjective, and access to IHC is not available worldwide. Thus, the application of additional methods to support morphological entity subtyping is desirable. Therefore, the ability of convolutional neuronal networks (CNNs) to classify the most common lung cancer subtypes, pulmonary adenocarcinoma (ADC), pulmonary squamous cell carcinoma (SqCC), and small-cell lung cancer (SCLC), was evaluated. A cohort of 80 ADC, 80 SqCC, 80 SCLC, and 30 skeletal muscle specimens was assembled; slides were scanned; tumor areas were annotated; image patches were extracted; and cases were randomly assigned to a training, validation or test set. Multiple CNN architectures (VGG16, InceptionV3, and InceptionResNetV2) were trained and optimized to classify the four entities. A quality control (QC) metric was established. An optimized InceptionV3 CNN architecture yielded the highest classification accuracy and was used for the classification of the test set. Image patch and patient-based CNN classification results were 95% and 100% in the test set after the application of strict QC. Misclassified cases mainly included ADC and SqCC. The QC metric identified cases that needed further IHC for definite entity subtyping. The study highlights the potential and limitations of CNN image classification models for tumor differentiation.

Objective: Recognition of neuroendocrine differentiation is important for tumor classification and treatment stratification. To detect and confirm neuroendocrine differentiation, a combination of morphology and immunohistochemistry is often required. In this regard, synaptophysin, chromogranin A, and CD56 are established immunohistochemical markers. Insulinoma-associated protein 1 (INSM1) has been suggested as a novel stand-alone marker with the potential to replace the current standard panel. In this study, we compared the sensitivity and specificity of INSM1 and established markers. Materials and Methods: A cohort of 493 lung tumors including 112 typical, 39 atypical carcinoids, 77 large cell neuroendocrine carcinomas, 144 small cell lung cancers, 30 thoracic paragangliomas, 47 adenocarcinomas, and 44 squamous cell carcinomas were selected and tissue microarrays were constructed. Synaptophysin, chromogranin A, CD56, and INSM1 were stained on all cases and evaluated manually as well as with an analysis software. Positivity was defined as ≥1% stained tumor cells in at least 1 of 2 cores per patient. Results: INSM1 was positive in 305 of 402 tumors with expected neuroendocrine differentiation (typical and atypical carcinoids, large cell neuroendocrine carcinomas, small cell lung cancers, and paraganglioma; sensitivity: 76%). INSM1 was negative in all but 1 of 91 analyzed non-neuroendocrine tumors (adenocarcinomas, squamous cell carcinomas; specificity: 99%). All conventional markers, as well as their combination, had a higher sensitivity (97%) and a lower specificity (78%) for neuroendocrine differentiation compared with INSM1. Conclusions: Although INSM1 might be a meaningful adjunct in the differential diagnosis of neuroendocrine neoplasias, a general uncritical vote for replacing the traditional markers by INSM1 may not be justified.

The diagnosis and the subtyping of non-Hodgkin lymphoma (NHL) are challenging and require expert knowledge, great experience, thorough morphological analysis, and often additional expensive immunohistological and molecular methods. As these requirements are not always available, supplemental methods supporting morphological-based decision making and potentially entity subtyping are required. Deep learning methods have been shown to classify histopathological images with high accuracy, but data on NHL subtyping are limited. After annotation of histopathological whole-slide images and image patch extraction, we trained and optimized an EfficientNet convolutional neuronal network algorithm on 84,139 image patches from 629 patients and evaluated its potential to classify tumor-free reference lymph nodes, nodal small lymphocytic lymphoma/chronic lymphocytic leukemia, and nodal diffuse large B-cell lymphoma. The optimized algorithm achieved an accuracy of 95.56% on an independent test set including 16,960 image patches from 125 patients after the application of quality controls. Automatic classification of NHL is possible with high accuracy using deep learning on histopathological images and routine diagnostic applications should be pursued.

Purpose Identification of proteolytic peptides from matrix‐assisted laser desorption/ionization (MALDI) imaging remains a challenge. The low fragmentation yields obtained using in situ post source decay impairs identification. Liquid chromatography‐tandem mass spectrometry (LC‐MS/MS) is an alternative to in situ MS/MS, but leads to multiple identification candidates for a given mass. The authors propose to use LC‐MS/MS‐based biomarker discovery results to reliably identify proteolytic peptides from MALDI imaging. Experimental design The authors defined m/z values of interest for high grade squamous intraepithelial lesion (HSIL) by MALDI imaging. In parallel the authors used data from a biomarker discovery study to correlate m/z from MALDI imaging with masses of peptides identified by LC‐MS/MS in HSIL. The authors neglected candidates that were not significantly more abundant in HSIL according to the biomarker discovery investigation. Results The authors assigned identifications to three m/z of interest. The number of possible identifiers for MALDI imaging m/z peaks using LC‐MS/MS‐based biomarker discovery studies was reduced by about tenfold compared using a single LC‐MS/MS experiment. One peptide identification candidate was validated by immunohistochemistry. Conclusion and clinical relevance This concept combines LC‐MS/MS‐based quantitative proteomics with MALDI imaging and allows reliable peptide identification. Public datasets from LC‐MS/MS biomarker discovery experiments will be useful to identify MALDI imaging m/z peaks.