Classification Of Cancer Tissues Research Articles

A Review of Convolutional Neural Networks in Cancer Image Classification

Abstract. Cancer image analysis is a crucial element in the field of modern oncology, offering insights into tumor characteristics that are vital for accurate diagnosis and treatment planning. Conventional diagnostic techniques are frequently constrained by discrepancies in interpretation and delays in processing. In recent years, deep learning techniques, in particular convolutional neural networks (CNNs), have demonstrated remarkable capabilities in processing large-scale medical images, enabling the rapid detection and classification of cancerous tissue with high accuracy and precision. This paper reviews the application of CNNs in cancer image classification, introduces common CNN models, and discusses advances such as transfer learning and multimodal data fusion. Furthermore, this paper examines the current challenges confronting convolutional neural networks (CNNs) in the domain of cancer classification. It proposes promising avenues for future research and underscores the potential of CNNs in enhancing cancer diagnosis and even revolutionizing the healthcare system.

Saliva Based Diagnostic Prediction of Oral Squamous Cell Carcinoma using FTIR Spectroscopy.

Oral cancer ranks as the sixth most prevalent form of cancer worldwide, presenting a significant public health concern. According to the World Health Organization (WHO), within a 5-year period following diagnosis, the mortality rate among oral cancer patients of all stages stands at 45%. In this study, a total of 60 patients divided into 2 groups were recruited. Group A included 30 histo-pathologically confirmed OSCC patients and Group B included 30 healthy controls. A standardized procedure was followed to collect saliva samples. FTIR spectroscopy was done for all the saliva samples collected from both Group A and B. An IR Prestige-21 (Shimadzu Corp, Japan) spectrometer was used to record IR spectra in the 40-4000cm-1 range SVM classifier was applied in the classification of disease state from normal subjects using FTIR data. The peaks were identified at wave no 1180cm-1, 1230cm-1, 1340cm-1, 1360cm-1, 1420cm-1, 1460cm-1, 1500cm-1, 1540cm-1, 1560cm-1, and 1637cm-1. The observed results of SVM demonstrated the accuracy of 91.66% in the classification of Cancer tissues from the normal subjects with sensitivity of 83.33% while specificity and precision of 100.0%. The development of oral cancer leads to noticeable alterations in the secondary structure of proteins. These findings emphasize the promising use of ATR-FTIR platforms in conjunction with machine learning as a reliable, non-invasive, reagent-free, and highly sensitive method for screening and monitoring individuals with oral cancer.

Transfer Contrastive Learning for Raman Spectroscopy Skin Cancer Tissue Classification.

Using Raman spectroscopy (RS) signals for skin cancer tissue classification has recently drawn significant attention, because of its non-invasive optical technique nature using molecular structures and conformations within biological tissue for diagnosis. In reality, RS signals are noisy and unstable for training machine learning models. The scarcity of tissue samples also makes it challenging to learn reliable deep-learning networks for clinical usages. In this paper, we advocate a Transfer Contrasting Learning Paradigm (TCLP) to address the scarcity and noisy characteristics of the RS for skin cancer tissue classification. To overcome the challenge of limited samples, TCLPleverages transfer learning to pre-train deep learning models using RS data from similar domains (but collected from different RS equipments for other tasks). To tackle the noisy nature of the RS signals, TCLPuses contrastive learning to augment RS signals to learn reliable feature representation to represent RS signals for final classification. Experiments and comparisons, including statistical tests, demonstrate that TCLPoutperforms existing deep learning baselines for RS signal-based skin cancer tissue classification. Both code and data used in our study are available at: https://github.com/yeyimilk/tclp.

Prostate cancer tissue classification by multiphoton imaging, automated image analysis and machine learning.

Prostate carcinoma, a slow-growing and often indolent tumour, is the second most commonly diagnosed cancer among men worldwide. The prognosis is mainly based on the Gleason system through prostate biopsy analysis. However, new treatment and monitoring strategies depend on a more precise diagnosis. Here, we present results by multiphoton imaging for prostate tumour samples from 120 patients that allow to obtain quantitative parameters leading to specific tumour aggressiveness signatures. An automated image analysis was developed to recognise and quantify stromal fibre and neoplastic cell regions in each image. The set of metrics was able to distinguish between non-neoplastic tissue and carcinoma areas by linear discriminant analysis and random forest with accuracy of 89% ± 3%, but between Gleason groups of only 46% ± 6%. The reactive stroma analysis improved the accuracy to 65% ± 5%, clearly demonstrating that stromal parameters should be considered as additional criteria for a more accurate diagnosis.

Human Hepatocellular Carcinoma Classification from H&E Stained Histopathology Images with 3D Convolutional Neural Networks and Focal Loss Function.

This paper proposes a new Hepatocellular Carcinoma (HCC) classification method utilizing a hyperspectral imaging system (HSI) integrated with a light microscope. Using our custom imaging system, we have captured 270 bands of hyperspectral images of healthy and cancer tissue samples with HCC diagnosis from a liver microarray slide. Convolutional Neural Networks with 3D convolutions (3D-CNN) have been used to build an accurate classification model. With the help of 3D convolutions, spectral and spatial features within the hyperspectral cube are incorporated to train a strong classifier. Unlike 2D convolutions, 3D convolutions take the spectral dimension into account while automatically collecting distinctive features during the CNN training stage. As a result, we have avoided manual feature engineering on hyperspectral data and proposed a compact method for HSI medical applications. Moreover, the focal loss function, utilized as a CNN cost function, enables our model to tackle the class imbalance problem residing in the dataset effectively. The focal loss function emphasizes the hard examples to learn and prevents overfitting due to the lack of inter-class balancing. Our empirical results demonstrate the superiority of hyperspectral data over RGB data for liver cancer tissue classification. We have observed that increased spectral dimension results in higher classification accuracy. Both spectral and spatial features are essential in training an accurate learner for cancer tissue classification.

Deep Gaussian Processes for Classification With Multiple Noisy Annotators. Application to Breast Cancer Tissue Classification

Machine learning (ML) methods often require large volumes of labeled data to achieve meaningful performance. The expertise necessary for labeling data in medical applications like pathology presents a significant challenge in developing clinical-grade tools. Crowdsourcing approaches address this challenge by collecting labels from multiple annotators with varying degrees of expertise. In recent years, multiple methods have been adapted to learn from noisy crowdsourced labels. Among them, Gaussian Processes (GPs) have achieved excellent performance due to their ability to model uncertainty. Deep Gaussian Processes (DGPs) address the limitations of GPs using multiple layers to enable the learning of more complex representations. In this work, we develop Deep Gaussian Processes for Crowdsourcing (DGPCR) to model the crowdsourcing problem with DGPs for the first time. DGPCR models the (unknown) underlying true labels, and the behavior of each annotator is modeled with a confusion matrix among classes. We use end-to-end variational inference to estimate both DGPCR parameters and annotator biases. Using annotations from 25 pathologists and medical trainees, we show that DGPCR is competitive or superior to Scalable Gaussian Processes for Crowdsourcing (SVGPCR) and other state-of-the-art deep-learning crowdsourcing methods for breast cancer classification. Also, we observe that DGPCR with noisy labels obtains better results ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\text{F}1=81.91$ </tex-math></inline-formula> %) than GPs ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\text{F}1=81.57$ </tex-math></inline-formula> %) and deep learning methods ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\text{F}1=80.88$ </tex-math></inline-formula> %) with true labels curated by experts. Finally, we show an improved estimation of annotators’ behavior.

Aggregation of Multimodal ICE-MS Data into Joint Classifier Increases Quality of Brain Cancer Tissue Classification

Mass spectrometry fingerprinting combined with multidimensional data analysis has been proposed in surgery to determine if a biopsy sample is a tumor. In the specific case of brain tumors, it is complicated to obtain control samples, leading to model overfitting due to unbalanced sample cohorts. Usually, classifiers are trained using a single measurement regime, most notably single ion polarity, but mass range and spectral resolution could also be varied. It is known that lipid groups differ significantly in their ability to produce positive or negative ions; hence, using only one polarity significantly restricts the chemical space available for sample discrimination purposes. In this work, we have developed an approach employing mass spectrometry data obtained by eight different regimes of measurement simultaneously. Regime-specific classifiers are trained, then a mixture of experts techniques based on voting or mean probability is used to aggregate predictions of all trained classifiers and assign a class to the whole sample. The aggregated classifiers have shown a much better performance than any of the single-regime classifiers and help significantly reduce the effect of an unbalanced dataset without any augmentation.

A Convolutional Neural Networks Model for Breast Tissue Classification

Diagnosis of breast cancer and the determination of cancer type are essential information for cancer research in monitoring and managing the disease. Artificial intelligence techniques developed in recent years have led to many developments in medicine, as any information about the patient has become more valuable. Especially, artificial intelligence methods used in the detection and classification of cancer tissues directly assist physicians and contribute to the management of the treatment process. This study aims to classify breast tissues with ten different tissue characteristics utilizing the breast tissue data set, which has 106 electrical impedance spectroscopies taken from 64 patients in the UCI Machine Learning Repository database. Various machine learning algorithms such as k-nearest neighbors, support vector machine, decision tree, self-organizing fuzzy logic, and convolutional neural networks are used to classify these tissues with the accuracy of 81%, 78%, 82%, 92%, and 96%, respectively. This study demonstrated the benefit of the usage of convolutional neural networks in cancer detection and tissue classification. Compared to traditional methods, convolutional neural networks provided a more reliable and better results.

Hyperspectral terahertz imaging and optical clearance for cancer classification in breast tumor surgical specimen.

Purpose: We investigate the enhancement in terahertz (THz) images of freshly excised breast tumors upon treatment with an optical clearance agent. The hyperspectral imaging and spectral classifications are used to quantitatively demonstrate the image enhancement. Glycerol solution with 60% concentration is applied to excised breast tumor specimens for various time durations to investigate the effectiveness on image enhancement. Approach: THz reflection spectroscopy is utilized to obtain the absorption coefficient and the index of refraction of untreated and glycerol-treated tissues at each frequency up to 3THz. Two classifiers, spectral angular mapping (SAM) based on several kernels and Euclidean minimum distance (EMD) are implemented to evaluate the effectiveness of the treatment. The testing raw data is obtained from five breast cancer specimens: two untreated specimens and three specimens treated with glycerol solution for 20, 40, or 60min. All tumors used in the testing data have healthy tissues adjacent to cancerous ones consistent with the challenge faced in lumpectomy surgeries. Results: The glycerol-treated tissues showed a decrease in the absorption coefficients compared with untreated tissues, especially as the period of treatment increased. Although the sensitivity metric of the classifier presented higher values in the untreated tissues compared with the treated ones, the specificity and accuracy metrics demonstrated higher values for the treated tissues compared with the untreated ones. Conclusions: The biocompatible glycerol solution is a potential optical clearance agent in THz imaging while keeping the histopathology imaging intact. The SAM technique provided a good classification of cancerous tissues despite the small amount of cancer in the training data (only 7%). The SAM exponential kernel and EMD presented classification accuracy of to 85% compared with linear and polynomial kernels that provided accuracy ranging from 70% to 80%. Overall, glycerol treatment provides a potential improvement in cancer classification in freshly excised breast tumors.

Hybrid Feature Extraction Based Ensemble Classification Model to Diagnose Oral Carcinoma Using Histopathological Images

Detection and classification of cancerous tissue from histopathologic images is quite a challenging task for pathologists and computer assisted medical diagnosis systems because of the complexity of the histopathology image. For a good diagnostic system, feature extraction from the medical images plays a crucial role for better classification of images. Using inappropriate or redundant features leads to poor classification results because classification algorithm learns a lot of unimportant information from the images. We propose hybrid feature extractor using different feature extraction algorithms that can extract various types of features from histopathological image. For this study, feature fused Convolution Neural Network, Gray Level Cooccurrence Matrix, and Local Binary Pattern algorithms are used. The texture and deep features obtained from these methods are used as input vector to classifiers: Support Vector Machine, KNearest Neighbor, Naïve Bayes and Boosted Tree. Prediction results of these classifiers are combined using soft majority voting algorithm to predict final output. Proposed method achieved an accuracy of 98.71%, which is quite high as compared to previous similar research works. Proposed method was capable of identifying most of cancerous histopathology images. The combination of deep and textural features can be potentially used for creating computer assisted medical imaging diagnosis system that can detect cancer from histopathology images timely and accurately.

Deep learning data augmentation for Raman spectroscopy cancer tissue classification

Recently, Raman Spectroscopy (RS) was demonstrated to be a non-destructive way of cancer diagnosis, due to the uniqueness of RS measurements in revealing molecular biochemical changes between cancerous vs. normal tissues and cells. In order to design computational approaches for cancer detection, the quality and quantity of tissue samples for RS are important for accurate prediction. In reality, however, obtaining skin cancer samples is difficult and expensive due to privacy and other constraints. With a small number of samples, the training of the classifier is difficult, and often results in overfitting. Therefore, it is important to have more samples to better train classifiers for accurate cancer tissue classification. To overcome these limitations, this paper presents a novel generative adversarial network based skin cancer tissue classification framework. Specifically, we design a data augmentation module that employs a Generative Adversarial Network (GAN) to generate synthetic RS data resembling the training data classes. The original tissue samples and the generated data are concatenated to train classification modules. Experiments on real-world RS data demonstrate that (1) data augmentation can help improve skin cancer tissue classification accuracy, and (2) generative adversarial network can be used to generate reliable synthetic Raman spectroscopic data.

Deep Learning on Oral Squamous Cell Carcinoma Ex Vivo Fluorescent Confocal Microscopy Data: A Feasibility Study.

Background: Ex vivo fluorescent confocal microscopy (FCM) is a novel and effective method for a fast-automatized histological tissue examination. In contrast, conventional diagnostic methods are primarily based on the skills of the histopathologist. In this study, we investigated the potential of convolutional neural networks (CNNs) for automatized classification of oral squamous cell carcinoma via ex vivo FCM imaging for the first time. Material and Methods: Tissue samples from 20 patients were collected, scanned with an ex vivo confocal microscope immediately after resection, and investigated histopathologically. A CNN architecture (MobileNet) was trained and tested for accuracy. Results: The model achieved a sensitivity of 0.47 and specificity of 0.96 in the automated classification of cancerous tissue in our study. Conclusion: In this preliminary work, we trained a CNN model on a limited number of ex vivo FCM images and obtained promising results in the automated classification of cancerous tissue. Further studies using large sample sizes are warranted to introduce this technology into clinics.

Cancer Tissue Classification Using Supervised Machine Learning Applied to MALDI Mass Spectrometry Imaging.

Simple SummaryClassic histopathological examination of tissues remains the mainstay for cancer diagnosis and staging. However, in some cases histopathologic analysis yields ambiguous results, leading to inconclusive disease classification. We set out to explore the diagnostic potential of mass spectrometry-based imaging for tumour classification based on proteomic fingerprints. Combining mass spectrometry with supervised machine learning, we were able to distinguish colorectal tumor from normal tissue with an overall accuracy of 98%. In addition, this approach was able to predict the presence of lymph node metastasis in primary tumour of endometrial cancer with an overall accuracy of 80%. These results highlight the potential of this technology to determine the optimal treatment for cancer patients to reduce morbidity and improve patients’ outcomes.Matrix assisted laser desorption/ionization mass spectrometry imaging (MALDI MSI) can determine the spatial distribution of analytes such as protein distributions in a tissue section according to their mass-to-charge ratio. Here, we explored the clinical potential of machine learning (ML) applied to MALDI MSI data for cancer diagnostic classification using tissue microarrays (TMAs) on 302 colorectal (CRC) and 257 endometrial cancer (EC)) patients. ML based on deep neural networks discriminated colorectal tumour from normal tissue with an overall accuracy of 98% in balanced cross-validation (98.2% sensitivity and 98.6% specificity). Moreover, our machine learning approach predicted the presence of lymph node metastasis (LNM) for primary tumours of EC with an accuracy of 80% (90% sensitivity and 69% specificity). Our results demonstrate the capability of MALDI MSI for complementing classic histopathological examination for cancer diagnostic applications.

Label-free microfluidic enrichment of cancer cells from non-cancer cells in ascites

The isolation of a patient's metastatic cancer cells is the first, enabling step toward treatment of that patient using modern personalized medicine techniques. Whereas traditional standard-of-care approaches select treatments for cancer patients based on the histological classification of cancerous tissue at the time of diagnosis, personalized medicine techniques leverage molecular and functional analysis of a patient's own cancer cells to select treatments with the highest likelihood of being effective. Unfortunately, the pure populations of cancer cells required for these analyses can be difficult to acquire, given that metastatic cancer cells typically reside in fluid containing many different cell populations. Detection and analyses of cancer cells therefore require separation from these contaminating cells. Conventional cell sorting approaches such as Fluorescence Activated Cell Sorting or Magnetic Activated Cell Sorting rely on the presence of distinct surface markers on cells of interest which may not be known nor exist for cancer applications. In this work, we present a microfluidic platform capable of label-free enrichment of tumor cells from the ascites fluid of ovarian cancer patients. This approach sorts cells based on differences in biomechanical properties, and therefore does not require any labeling or other pre-sort interference with the cells. The method is also useful in the cases when specific surface markers do not exist for cells of interest. In model ovarian cancer cell lines, the method was used to separate invasive subtypes from less invasive subtypes with an enrichment of ~ sixfold. In ascites specimens from ovarian cancer patients, we found the enrichment protocol resulted in an improved purity of P53 mutant cells indicative of the presence of ovarian cancer cells. We believe that this technology could enable the application of personalized medicine based on analysis of liquid biopsy patient specimens, such as ascites from ovarian cancer patients, for quick evaluation of metastatic disease progression and determination of patient-specific treatment.

TISSUE CLASSIFICATION FOR COLORECTAL CANCER UTILIZING TECHNIQUES OF DEEP LEARNING AND MACHINE LEARNING

Since the introduction of image pattern recognition and computer vision processing, the classification of cancer tissues has been a challenge at pixel-level, slide-level, and patient-level. Conventional machine learning techniques have given way to Deep Learning (DL), a contemporary, state-of-the-art approach to texture classification and localization of cancer tissues. Colorectal Cancer (CRC) is the third ranked cause of death from cancer worldwide. This paper proposes image-level texture classification of a CRC dataset by deep convolutional neural networks (CNN). Simple DL techniques consisting of transfer learning and fine-tuning were exploited. VGG-16, a Keras pre-trained model with initial weights by ImageNet, was applied. The transfer learning architecture and methods responding to VGG-16 are proposed. The training, validation, and testing sets included 5000 images of 150 × 150 pixels. The application set for detection and localization contained 10 large original images of 5000 × 5000 pixels. The model achieved F1-score and accuracy of 0.96 and 0.99, respectively, and produced a false positive rate of 0.01. AUC-based evaluation was also measured. The model classified ten large previously unseen images from the application set represented in false color maps. The reported results show the satisfactory performance of the model. The simplicity of the architecture, configuration, and implementation also contributes to the outcome this work.

AlexResNet+: A Deep Hybrid Featured Machine Learning Model for Breast Cancer Tissue Classification

The exponential rise in cancer diseases, primarily the breast cancer has alarmed academia-industry to achieve more efficient and reliable breast cancer tissue identification and classification. Unlike classical machine learning approaches which merely focus on enhancing classification efficiency, in this paper the emphasis was made on extracting multiple deep features towards breast cancer diagnosis. To achieve it, in this paper A Deep Hybrid Featured Machine Learning Model for Breast Cancer Tissue Classification named, AlexResNet+ was developed. We used two well known and most efficient deep learning models, AlexNet and shorted ResNet50 deep learning concepts for deep feature extraction. To retain high dimensional deep features while retaining optimal computational efficiency, we applied AlexNet with five convolutional layers, and three fully connected layers, while ResNet50 was applied with modified layered architectures. Retrieving the distinct deep features from AlexNet and ResNet deep learning models, we obtained the amalgamated feature set which were applied as input for support vector machine with radial basis function (SVM-RBF) for two-class classification. To assess efficacy of the different feature set, performances were obtained for AlexNet, shorted ResNet50 and hybrid features distinctly. The simulation results over DDMS mammogram breast cancer tissue images revealed that the proposed hybrid deep features (AlexResNet+) based model exhibits the highest classification accuracy of 95.87%, precision 0.9760, sensitivity 1.0, specificity 0.9621, F-Measure 0.9878 and AUC of 0.960.

Validation of Breast Cancer Margins by Tissue Spray Mass Spectrometry.

Current methods for the intraoperative determination of breast cancer margins commonly suffer from the insufficient accuracy, specificity and/or low speed of analysis, increasing the time and cost of operation as well the risk of cancer recurrence. The purpose of this study is to develop a method for the rapid and accurate determination of breast cancer margins using direct molecular profiling by mass spectrometry (MS). Direct molecular fingerprinting of tiny pieces of breast tissue (approximately 1 × 1 × 1 mm) is performed using a home-built tissue spray ionization source installed on a Maxis Impact quadrupole time-of-flight mass spectrometer (qTOF MS) (Bruker Daltonics, Hamburg, Germany). Statistical analysis of MS data from 50 samples of both normal and cancer tissue (from 25 patients) was performed using orthogonal projections onto latent structures discriminant analysis (OPLS-DA). Additionally, the results of OPLS classification of new 19 pieces of two tissue samples were compared with the results of histological analysis performed on the same tissues samples. The average time of analysis for one sample was about 5 min. Positive and negative ionization modes are used to provide complementary information and to find out the most informative method for a breast tissue classification. The analysis provides information on 11 lipid classes. OPLS-DA models are created for the classification of normal and cancer tissue based on the various datasets: All mass spectrometric peaks over 300 counts; peaks with a statistically significant difference of intensity determined by the Mann–Whitney U-test (p < 0.05); peaks identified as lipids; both identified and significantly different peaks. The highest values of Q2 have models built on all MS peaks and on significantly different peaks. While such models are useful for classification itself, they are of less value for building explanatory mechanisms of pathophysiology and providing a pathway analysis. Models based on identified peaks are preferable from this point of view. Results obtained by OPLS-DA classification of the tissue spray MS data of a new sample set (n = 19) revealed 100% sensitivity and specificity when compared to histological analysis, the “gold” standard for tissue classification. “All peaks” and “significantly different peaks” datasets in the positive ion mode were ideal for breast cancer tissue classification. Our results indicate the potential of tissue spray mass spectrometry for rapid, accurate and intraoperative diagnostics of breast cancer tissue as a means to reduce surgical intervention.

Automated classification of cancerous tissue of the face and head with confocal microscopy for treatment and diagnosis using deep learning

Background: Malignant dermal lesions of the face and head are one of the most frequent skin lesions and are associated with a number of diagnostic and therapeutic problems. This is particularly relevant in older, immunosuppressed patients, and those with multiple pre-operations. Resection of advanced neoplasia in the face and head region is associated with extensive, often functionally and aesthetically debilitating defects, reconstructures in general anesthesia and prolonged hospital stays. In many cases, even maximum radicality of the procedure cannot guarantee a tumor-free resection with clean margins and thus presents a high risk of reoccurrence. The histological examination takes place ex-vivo and lasts on average 5 working days. If the margins are not tumor-free, a further resection is necessary. After several days of initial pathological investigation, the orientation and accurate determination of the resection area is complicated for subjective reasons and physiological postoperative processes in the wound – even if done by the same surgeon.

Fusion of stability and multi-objective optimization for solving cancer tissue classification problem

The concept of stability is one of the commonly used physical phenomena. Current paper builds on the hypothesis that the optimal number of clusters present in the dataset corresponds to that partitioning which is most stable over some small changes in the dataset. In order to quantify the degree of stability, a new measure is also proposed in the paper. Thereafter an expert clustering approach is developed in the current paper which utilizes the properties of stability for automatically detecting the number of clusters from a given dataset. Initially, several different variants of the dataset are generated by introducing small perturbations. A multi-objective based expert clustering framework is developed to automatically partition different variants of the data. A new objective function, capturing stability property of clustering solution namely ‘Agreement-index’, along with two well-known objective functions are optimized simultaneously using a multi-objective simulated annealing based process, namely AMOSA for the purpose of clustering. Finally, the problem of cancer classification is addressed as the application domain of the proposed expert framework. Results of our newly developed stability based clustering namely Stab-clustering with respect to existing approaches are shown for twelve microarray cancer datasets in terms of different cluster quality measures. The obtained results confirm the robustness of our proposed technique over state-of-the-art. A thorough biological and statistical significance tests are also conducted to prove the effectiveness of the proposed approach.

Automatic Classification of Cancerous Tissue in Laserendomicroscopy Images of the Oral Cavity using Deep Learning

Oral Squamous Cell Carcinoma (OSCC) is a common type of cancer of the oral epithelium. Despite their high impact on mortality, sufficient screening methods for early diagnosis of OSCC often lack accuracy and thus OSCCs are mostly diagnosed at a late stage. Early detection and accurate outline estimation of OSCCs would lead to a better curative outcome and a reduction in recurrence rates after surgical treatment. Confocal Laser Endomicroscopy (CLE) records sub-surface micro-anatomical images for in vivo cell structure analysis. Recent CLE studies showed great prospects for a reliable, real-time ultrastructural imaging of OSCC in situ. We present and evaluate a novel automatic approach for OSCC diagnosis using deep learning technologies on CLE images. The method is compared against textural feature-based machine learning approaches that represent the current state of the art. For this work, CLE image sequences (7894 images) from patients diagnosed with OSCC were obtained from 4 specific locations in the oral cavity, including the OSCC lesion. The present approach is found to outperform the state of the art in CLE image recognition with an area under the curve (AUC) of 0.96 and a mean accuracy of 88.3% (sensitivity 86.6%, specificity 90%).