New insights into spectral histopathology: infrared-based scoring of tumour aggressiveness of squamous cell lung carcinomas

Deep learning for the prediction of the chemotherapy response of metastatic colorectal cancer: comparing and combining H&E staining histopathology and infrared spectral histopathology

Colorectal cancer is a global public health problem with one of the highest death rates. It is the second most deadly type of cancer and the third most frequently diagnosed in the world. The present study focused on metastatic colorectal cancer (mCRC) patients who had been treated with chemotherapy-based regimen for which it remains uncertainty about the efficacy for all eligible patients. This is a major problem, as it is not yet possible to test different therapies in view of the consequences on the health of the patients and the risk of progression. Here, we propose a method to predict the efficacy of an anticancer treatment in an individualized way, using a deep learning model constructed on the retrospective analysis of the primary tumor of several patients. Histological sections from tumors were imaged by standard hematoxylin and eosin (HE) staining and infrared spectroscopy (IR). Images obtained were then processed by a convolutional neural network (CNN) to extract features and correlate them with the subsequent progression-free survival (PFS) of each patient. Separately, HE and IR imaging resulted in a PFS prediction with an error of 6.6 and 6.3 months respectively (28% and 26% of the average PFS). Combining both modalities allowed to decrease the error to 5.0 months (21%). The inflammatory state of the stroma seemed to be one of the main features detected by the CNN. Our pilot study suggests that multimodal imaging analyzed with deep learning methods allow to give an indication of the effectiveness of a treatment when choosing.

Mid-Infrared Imaging Characterization to Differentiate Lung Cancer Subtypes

Introduction: Lung cancer is the most common malignancy worldwide. Squamous cell carcinoma (SQ) and adenocarcinoma (LUAD) are the two most frequent histological subtypes. Small cell carcinoma (SCLC) subtype has the worst prognosis. Differential diagnosis is essential for proper oncological treatment. Life science associated mid- and near-infrared based microscopic techniques have been developed exponentially, especially in the past decade. Vibrational spectroscopy is a potential non-destructive approach to investigate malignancies.

Aims: Our goal was to differentiate lung cancer subtypes by their label-free mid-infrared spectra using supervised multivariate analyses.

Material and Methods: Formalin-fixed paraffin-embedded (FFPE) samples were selected from the archives. Three subtypes were selected for each group: 10-10 cases SQ, LUAD and SCLC. 2 μm thick sections were cut and laid on aluminium coated glass slides. Transflection optical setup was applied on Perkin-Elmer infrared microscope. 250 × 600 μm areas were imaged and the so-called mid-infrared fingerprint region (1800-648cm⁻¹) was further analysed with linear discriminant analysis (LDA) and support vector machine (SVM) methods.

Results: Both “patient-based” and “pixel-based” approaches were examined. Patient-based analysis by using 3 LDA models and 2 SVM models resulted in different separations. The higher the cut-off value the lower is the accuracy. The linear C-support vector classification (C-SVC) SVM resulted in the best (100%) accuracy for the three subtypes using a 50% cut-off value. The pixel-based analysis gave, similarly, the linear C-SVC SVM model to be the most efficient in the statistical indicators (SQ sensitivity 81.65%, LUAD sensitivity 82.89% and SCLC sensitivity 88.89%). The spectra cut-off, the kernel function and the algorithm function influence the accuracy.

Conclusion: Mid-Infrared imaging could be used to differentiate FFPE lung cancer subtypes. Supervised multivariate tools are promising to accurately separate lung tumor subtypes. The long-term perspective is to develop a spectroscopy-based diagnostic tool, revolutionizing medical differential diagnostics, especially cancer identification.

Infrared Spectroscopic Analysis in the Differentiation of Epithelial Misplacement From Adenocarcinoma in Sigmoid Colonic Adenomatous Polyps

Purpose

The differential diagnosis of epithelial misplacement from invasive cancer in the colon is a challenging endeavour, augmented by the introduction of bowel cancer population screening. The main aim of the work is to test, as a proof-of concept study, the ability of the infrared spectroscopic imaging approach to differentiate epithelial misplacement from adenocarcinoma in sigmoid colonic adenomatous polyps.

Methods

Ten samples from each of the four diagnostic groups, normal colonic mucosa, adenomatous polyps with low grade dysplasia, epithelial misplacement in adenomatous polyps and adenocarcinoma, were analysed using IR spectroscopic imaging and data processing methods. IR spectral images were subjected to data pre-processing and cluster analysis based segmentation to identify epithelial, connective tissue and stromal regions. Statistical analysis was carried out using principal component analysis and linear discriminant analysis based cross validation, to classify spectral features according to the pathology, and the diagnostic attributes were compared.

Results

The combined 4-group classification model on an average showed a sensitivity of 64%, a specificity of 88% and an accuracy of 76% for prediction based on a 'single spectrum', whilst a 'majority-vote' prediction on an average showed a sensitivity of 73%, a specificity of 90% and an accuracy of 82%. The 2-group model, for the differential diagnosis of epithelial misplacement versus adenocarcinoma, showed an average sensitivity and specificity of 82.5% for prediction based on a 'single spectrum' whilst a 'majority-vote' classification showed an average sensitivity and specificity of 90%. A 92% area under the curve (AUC) value was obtained when evaluating the classifier using the Receiver Operating Characteristics (ROC) curves.

Conclusions

IR spectroscopy shows promise in its ability to differentiate epithelial misplacement from adenocarcinoma in tissue sections, considered as one of the most challenging endeavours in population-wide diagnostic histopathology. Further studies with larger series, including cases with challenging diagnostic features are required to ascertain the capability of this modern digital pathology approach. In the long-term, IR spectroscopy based pathology which is relatively low-cost and rapid, could be a promising endeavour to consider for integration into the existing histopathology pathway, in particular for population based screening programmes where large number of samples are scrutinised.

Identification of Neoadjuvant Chemotherapy Response in Muscle-Invasive Bladder Cancer by Fourier-Transform Infrared Micro-Imaging

Simple Summary

Assessing the tumor response to chemotherapy is a paramount predictive step to improve patient care. Infrared spectroscopy probes the chemical composition of samples, and in combination with statistical multivariate processing, presents the capacity to highlight subtle molecular alterations associated with malignancy characteristics. Microscopic infrared imaging of tissue samples reveals spectral heterogeneity within histological structures, providing a new approach to characterize tumoral heterogeneity. We have taken advantage of the analytical capabilities of mid-infrared spectral imaging to implement a classification model to predict the response of a tumor to chemotherapy. Our development was demonstrated in muscle-invasive bladder cancer (MIBC) by comparing samples from responders and non-responders to neoadjuvant chemotherapy.

Abstract

Background: Neoadjuvant chemotherapy (NAC) improves survival in responder patients. However, for non-responders, the treatment represents an ineffective exposure to chemotherapy and its potential adverse events. Predicting the response to treatment is a major issue in the therapeutic management of patients, particularly for patients with muscle-invasive bladder cancer. Methods: Tissue samples of trans-urethral resection of bladder tumor collected at the diagnosis time, were analyzed by mid-infrared imaging. A sequence of spectral data processing was implemented for automatic recognition of informative pixels and scoring each pixel according to a continuous scale (from 0 to 10) associated with the response to NAC. The ground truth status of the responder or non-responder was based on histopathological examination of the samples. Results: Although the TMA spots of tumors appeared histologically homogeneous, the infrared approach highlighted spectral heterogeneity. Both the quantification of this heterogeneity and the scoring of the NAC response at the pixel level were used to construct sensitivity and specificity maps from which decision criteria can be extracted to classify cancerous samples. Conclusions: This proof-of-concept appears as the first to evaluate the potential of the mid-infrared approach for the prediction of response to neoadjuvant chemotherapy in MIBC tissues.

In search of the correlation between nanomechanical and biomolecular properties of prostate cancer cells with different metastatic potential

Nanomechanical properties of living cells, as measured with atomic force microscopy (AFM), are increasingly recognized as criteria that differentiate normal and pathologically altered cells. Locally measured cell elastic properties, described by the parameter known as Young's modulus, are currently proposed as a new diagnostic parameter that can be used at the early stage of cancer detection. In this study, local mechanical properties of normal human prostate (RWPE-1) cells and a range of malignant (22Rv1) and metastatic prostate cells (LNCaP, Du145 and PC3) were investigated. It was found that non-malignant prostate cells are stiffer than cancer cells while the metastatic cells are much softer than malignant cells from the primary tumor site. Next, the biochemical properties of the cells were measured using confocal Raman (RS) and Fourier-transform infrared (FT-IR) spectroscopies to reveal these cells' biochemical composition as malignant transformation proceeds. Nanomechanical and biochemical profiles of five different prostate cell lines were subsequently analyzed using partial least squares regression (PLSR) in order to identify which spectral features of the RS and FT-IR spectra correlate with the cell's elastic properties. The PLSR-based model could predict Young's modulus values based on both RS and FT-IR spectral information. These outcomes show not only that AFM, RS and FT-IR techniques can be used for discrimination between normal and cancer cells, but also that a linear correlation between mechanical response and biomolecular composition of the cells that undergo malignant transformation can be found. This knowledge broadens our understanding of how prostate cancer cells evolve thorough the multistep process of tumor pathogenesis.

Label-Free Infrared Spectral Histology of Skin Tissue Part II: Impact of a Lumican-Derived Peptide on Melanoma Growth

Melanoma is the most aggressive type of cutaneous malignancies. In addition to its role as a regulator of extracellular matrix (ECM) integrity, lumican, a small leucine-rich proteoglycan, also exhibits anti-tumor properties in melanoma. This work focuses on the use of infrared spectral imaging (IRSI) and histopathology (IRSH) to study the effect of lumican-derived peptide (L9Mc) on B16F1 melanoma primary tumor growth. Female C57BL/6 mice were injected with B16F1 cells treated with L9Mc (n = 10) or its scrambled peptide (n = 8), and without peptide (control, n = 9). The melanoma primary tumors were subjected to histological and IR imaging analysis. In addition, immunohistochemical staining was performed using anti-Ki-67 and anti-cleaved caspase-3 antibodies. The IR images were analyzed by common K-means clustering to obtain high-contrast IRSH that allowed identifying different ECM tissue regions from the epidermis to the tumor area, which correlated well with H&E staining. Furthermore, IRSH showed good correlation with immunostaining data obtained with anti-Ki-67 and anti-cleaved caspase-3 antibodies, whereby the L9Mc peptide inhibited cell proliferation and increased strongly apoptosis of B16F1 cells in this mouse model of melanoma primary tumors.