Combining deep learning and coherent anti-Stokes Raman scattering imaging for automated differential diagnosis of lung cancer

Abstract

Abstract. Lung cancer is the most prevalent type of cancer and the leading cause of cancer-related deaths worldwide. Coherent anti-Stokes Raman scattering (CARS) is capable of providing cellular-level images and resolving pathologically related features on human lung tissues. However, conventional means of analyzing CARS images requires extensive image processing, feature engineering, and human intervention. This study demonstrates the feasibility of applying a deep learning algorithm to automatically differentiate normal and cancerous lung tissue images acquired by CARS. We leverage the features learned by pretrained deep neural networks and retrain the model using CARS images as the input. We achieve 89.2% accuracy in classifying normal, small-cell carcinoma, adenocarcinoma, and squamous cell carcinoma lung images. This computational method is a step toward on-the-spot diagnosis of lung cancer and can be further strengthened by the efforts aimed at miniaturizing the CARS technique for fiber-based microendoscopic imaging. © The Authors. Published by SPIE under a Creative Commons Attribution 3.0 Unported License. Distribution or reproduction of this work in whole or in part requires full attribution of the original publication, including its DOI.

Automated classification of hepatocellular carcinoma differentiation using multiphoton microscopy and deep learning

“…In early research, CNNs were mainly applied to white‐light images . In recent years, CNN methods have achieved many breakthroughs in the field of biomedical image processing ; for example, in classification tasks for different diseases , including the automatic diagnosis of lung cancer . Although a variety of CNNs have emerged recently, their basic structure consists of an input layer, convolution layer, activation function, pooling layer, and fully connected layer.…”

“…Alternatively, other pre‐trained networks can be selected to improve the accuracy of the CNN . Furthermore, considering the abundance of available imaging pathways, such as third‐harmonic generation, coherent anti‐Stokes Raman scattering , and other MPM techniques, there are many other potential routes to obtain label‐free imaging and automated classification of HCC samples.…”

“…An artificial method was used to increase the number of training samples. As shown in , the augmentation of experimental data sets by symmetry operations (rotation by an angle, horizontal or vertical flipping) has become a standard method to expand sample sizes for deep learning. All the collected HCC images were horizontally flipped to obtain a new set of images.…”

Ultra-fast fit-free analysis of complex fluorescence lifetime imaging via deep learning

“…Of particular note, Machine Learning (ML) and Deep Learning (DL) methods have recently profoundly impacted the image processing field. For example, deepneural networks (DNNs) are currently providing high level, robust performances in numerous biomedical applicationssuch as in pathology through multiple imaging modalities, [17][18][19] natural language processing, 20 image reconstruction via M direct mapping from the sensor-image domain 21 and reinforcement learning applied to drug discovery. 22 DL methods are increasingly employed in molecular optical imaging applications from resolution enhancement in histopathology 23 , super resolution microscopy 24 , fluorescence signal prediction from label-free images 25 , single molecular localization 26 , fluorescence microscopy image restoration 27 and hyperspectral single pixel lifetime imaging for instance 28 .…”

Image processing methods for computer-aided screening for disease

“…GoogLeNet was successfully used in the detection of lymph node metastases in women with breast cancer (Bejnordi et al 2017), in the classification of normal and cancerous lung tissues from CARS (Coherent anti-Stokes Raman scattering) images (Weng et al 2017) or retinal pathologies using optical coherence tomography (OCT) images (Karri, Chakraborty, and Chatterjee 2017). The input in VGG19 and GoogleNet networks were the preprocessed color fundus images from the datasets of the study with a final size of 224x224x3 and centered at the optic disc.…”