Polyp Segmentation Using a Hybrid Vision Transformer and a Hybrid Loss Function.

Accurate and early detection of precursor adenomatous polyps and their removal at the early stage can significantly decrease the mortality rate and the occurrence of the disease since most colorectal cancer evolve from adenomatous polyps. However, accurate detection and segmentation of the polyps by doctors are difficult mainly these factors: (i) quality of the screening of the polyps with colonoscopy depends on the imaging quality and the experience of the doctors; (ii) visual inspection by doctors is time-consuming, burdensome, and tiring; (iii) prolonged visual inspections can lead to polyps being missed even when the physician is experienced. To overcome these problems, computer-aided methods have been proposed. However, they have some disadvantages or limitations. Therefore, in this work, a new architecture based on residual transformer layers has been designed and used for polyp segmentation. In the proposed segmentation, both high-level semantic features and low-level spatial features have been utilized. Also, a novel hybrid loss function has been proposed. The loss function designed with focal Tversky loss, binary cross-entropy, and Jaccard index reduces image-wise and pixel-wise differences as well as improves regional consistencies. Experimental works have indicated the effectiveness of the proposed approach in terms of dice similarity (0.9048), recall (0.9041), precision (0.9057), and F2 score (0.8993). Comparisons with the state-of-the-art methods have shown its better performance.

Medical image data augmentation: techniques, comparisons and interpretations.

Designing deep learning based methods with medical images has always been an attractive area of research to assist clinicians in rapid examination and accurate diagnosis. Those methods need a large number of datasets including all variations in their training stages. On the other hand, medical images are always scarce due to several reasons, such as not enough patients for some diseases, patients do not want to allow their images to be used, lack of medical equipment or equipment, inability to obtain images that meet the desired criteria. This issue leads to bias in datasets, overfitting, and inaccurate results. Data augmentation is a common solution to overcome this issue and various augmentation techniques have been applied to different types of images in the literature. However, it is not clear which data augmentation technique provides more efficient results for which image type since different diseases are handled, different network architectures are used, and these architectures are trained and tested with different numbers of data sets in the literature. Therefore, in this work, the augmentation techniques used to improve performances of deep learning based diagnosis of the diseases in different organs (brain, lung, breast, and eye) from different imaging modalities (MR, CT, mammography, and fundoscopy) have been examined. Also, the most commonly used augmentation methods have been implemented, and their effectiveness in classifications with a deep network has been discussed based on quantitative performance evaluations. Experiments indicated that augmentation techniques should be chosen carefully according to image types.

Evaluation of denoising techniques to remove speckle and Gaussian noise from dermoscopy images.

Computerized methods provide analyses of skin lesions from dermoscopy images automatically. However, the images acquired from dermoscopy devices are noisy and cause low accuracy in automated methods. Therefore, various methods have been applied for denoising in the literature. There are some review-type papers about these methods. However, their authors have focused on either denoising with a specific approach or denoising from other images rather than dermoscopy images, which have a different characteristic. It is not possible to determine which method is the most suitable for denoising from dermoscopy images according to the results presented in them. Therefore, a review on the denoising approaches applied with dermoscopy images is required and, according to our knowledge, there is no such a review-type paper. To fill this gap in the literature, the required review has been performed in this work. Also, in this work, the methods in the literature have been implemented using the same data sets containing images with speckle or Gaussian types of noise. The results have been analyzed not only visually but also quantitatively to compare capabilities of the techniques. Our experiments indicated that each denoising technique has its own disadvantages and advantages. The main contributions of this paper are three-fold: (i) A comprehensive review on the denoising approaches applied with dermoscopy images has been presented. (ii) The denoising techniques have been implemented with the same images for meaningful comparisons. (iii) Both visual and quantitative analyses with different metrics have been performed and comparative performance evaluations have been presented.

Diagnosis of skin diseases in the era of deep learning and mobile technology.

Efficient methods developed with deep learning in the last ten years have provided objectivity and high accuracy in the diagnosis of skin diseases. They also support accurate, cost-effective and timely treatment. In addition, they provide diagnoses without the need to touch patients, which is very desirable when the disease is contagious or the patients have another contagious disease. On the other hand, it is not possible to run deep networks on resource-constrained devices (e.g., mobile phones). Therefore, lightweight network architectures have been proposed in the literature. However, merely a few mobile applications have been developed for the diagnosis of skin diseases from colored photographs using lightweight networks. Moreover, only a few types of skin diseases have been addressed in those applications. Additionally, they do not perform as well as the deep network models, particularly for pattern recognition. Therefore, in this study, a novel model has been constructed using MobileNet. Also, a novel loss function has been developed and used. The main contributions of this study are: (i) proposing a novel hybrid loss function; (ii) proposing a modified-MobileNet architecture; (iii) designing and implementing a mobile phone application with the modified-MobileNet and a user-friendly interface. Results indicated that the proposed technique can diagnose skin diseases with 94.76% accuracy.

Deep learning based classification of facial dermatological disorders.

Common properties of dermatological diseases are mostly lesions with abnormal pattern and skin color (usually redness). Therefore, dermatology is one of the most appropriate areas in medicine for automated diagnosis from images using pattern recognition techniques to provide accurate, objective, early diagnosis and interventions. Also, automated techniques provide diagnosis without depending on location and time. In addition, the number of patients in dermatology departments and costs of dermatologist visits can be reduced. Therefore, in this work, an automated method is proposed to classify dermatological diseases from color digital photographs. Efficiency of the proposed approach is provided by 2 stages. In the 1st stage, lesions are detected and extracted by using a variational level set technique after noise reduction and intensity normalization steps. In the 2nd stage, lesions are classified using a pre-trained DenseNet201 architecture with an efficient loss function. In this study, five common facial dermatological diseases are handled since they also cause anxiety, depression and even suicide death. The main contributions provided by this work can be identified as follows: (i) A comprehensive survey about the state-of-the-art works on classifications of dermatological diseases using deep learning; (ii) A new fully automated lesion detection and segmentation based on level sets; (iii) A new adaptive, hybrid and non-symmetric loss function; (iv) Using a pre-trained DenseNet201 structure with the new loss function to classify skin lesions; (v) Comparative evaluations of ten convolutional networks for skin lesion classification. Experimental results indicate that the proposed approach can classify lesions with high performance (95.24% accuracy).

Diagnosis of Alzheimer's disease with Sobolev gradient-based optimization and 3D convolutional neural network.

Alzheimer's disease is a neuropsychiatric, progressive, also an irreversible disease. There is not an effective cure for the disease. However, early diagnosis has an important role for treatment planning to delay its progression since the treatments have the most impact at the early stage of the disease. Neuroimages obtained by different imaging techniques (for example, diffusion tensor-based and magnetic resonance-based imaging) provide powerful information and help to diagnose the disease. In this work, a deeply supervised and robust method has been developed using three dimensional features to provide objective and accurate diagnosis from magnetic resonance images. The main contributions are (a) a new three dimensional convolutional neural network topology; (b) a new Sobolev gradient-based optimization with weight values for each decision parameters; (c) application of the proposed topology and optimizer to diagnose Alzheimer's disease; (d) comparisons of the results obtained from the recent techniques that have been implemented for Alzheimer's disease diagnosis. Experimental results and quantitative evaluations indicated that the proposed network model is able to achieve to extract desired features from images and provides automated diagnosis with 98.06% accuracy.

Automated fluorescent miscroscopic image analysis of PTBP1 expression in glioma.

Multiplexed immunofluorescent testing has not entered into diagnostic neuropathology due to the presence of several technical barriers, amongst which includes autofluorescence. This study presents the implementation of a methodology capable of overcoming the visual challenges of fluorescent microscopy for diagnostic neuropathology by using automated digital image analysis, with long term goal of providing unbiased quantitative analyses of multiplexed biomarkers for solid tissue neuropathology. In this study, we validated PTBP1, a putative biomarker for glioma, and tested the extent to which immunofluorescent microscopy combined with automated and unbiased image analysis would permit the utility of PTBP1 as a biomarker to distinguish diagnostically challenging surgical biopsies. As a paradigm, we utilized second resections from patients diagnosed either with reactive brain changes (pseudoprogression) and recurrent glioblastoma (true progression). Our image analysis workflow was capable of removing background autofluorescence and permitted quantification of DAPI-PTBP1 positive cells. PTBP1-positive nuclei, and the mean intensity value of PTBP1 signal in cells. Traditional pathological interpretation was unable to distinguish between groups due to unacceptably high discordance rates amongst expert neuropathologists. Our data demonstrated that recurrent glioblastoma showed more DAPI-PTBP1 positive cells and a higher mean intensity value of PTBP1 signal compared to resections from second surgeries that showed only reactive gliosis. Our work demonstrates the potential of utilizing automated image analysis to overcome the challenges of implementing fluorescent microscopy in diagnostic neuropathology.

Vessel segmentation from abdominal magnetic resonance images: adaptive and reconstructive approach.

The liver vessels, which have low signal and run next to brighter bile ducts, are difficult to segment from MR images. This study presents a fully automated and adaptive method to segment portal and hepatic veins on magnetic resonance images. In the proposed approach, segmentation of these vessels is achieved in four stages: (i) initial segmentation, (ii) refinement, (iii) reconstruction, and (iv) post-processing. In the initial segmentation stage, k-means clustering is used, the results of which are refined iteratively with linear contrast stretching algorithm in the next stage, generating a mask image. In the reconstruction stage, vessel regions are reconstructed with the marker image from the first stage and the mask image from the second stage. Experimental data sets include slices that show fat tissues, which have the same gray level values with vessels, outside the margin of the liver. These structures are removed in the last stage. Results show that the proposed approach is more efficient than other thresholding-based methods. Copyright © 2016 John Wiley & Sons, Ltd.

Quantitative validation of anti-PTBP1 antibody for diagnostic neuropathology use: Image analysis approach.

Traditional diagnostic neuropathology relies on subjective interpretation of visual data obtained from a brightfield microscopy. This approach causes high variability, unsatisfactory reproducibility, and inability for multiplexing even among experts. These problems may affect patient outcomes and confound clinical decision-making. Also, standard histological processing of pathological specimens leads to auto-fluorescence and other artifacts, a reason why fluorescent microscopy is not routinely implemented in diagnostic pathology. To overcome these problems, objective and quantitative methods are required to help neuropathologists in their clinical decision-making. Therefore, we propose a computerized image analysis method to validate anti-PTBP1 antibody for its potential use in diagnostic neuropathology. Images were obtained from standard neuropathological specimens stained with anti-PTBP1 antibody. First, the noise characteristics of the images were modeled and images are de-noised according to the noise model. Next, images are filtered with sigma-adaptive Gaussian filtering for normalization, and cell nuclei are detected and segmented with a k-means-based deterministic approach. Experiments on 29 data sets from 3 cases of brain tumor and reactive gliosis show statistically significant differences between the number of positively stained nuclei in images stained with and without anti-PTBP1 antibody. The experimental analysis of specimens from 3 different brain tumor groups and 1 reactive gliosis group indicates the feasibility of using anti-PTBP1 antibody in diagnostic neuropathology, and computerized image analysis provides a systematic and quantitative approach to explore feasibility.

Automatic labeling of portal and hepatic veins from MR images prior to liver transplantation.

PURPOSE: Living donated liver transplantation is an important task since a person (healthy donor) donates some part of her/his liver to a person in this surgery operation. The success of this operation mainly depends on the sufficiency of vessels and volume of the liver. Accurate labeling of portal and hepatic veins of donors reduces the incidence of complications during and after transplantation. Therefore, prior to the hepatic surgery, automatic analysis and labeling of vasculature structures in the liver are vital to see whether liver is suitable or not for transplantation. However, automatic labeling of veins in the liver is challenging because of partial volume effects, noise and image resolution, which causes wrong connections between vessels. The goal of this paper is to propose an automatic labeling approach for vessels. METHODS: The proposed automated labeling method is based on gray-level values in the MR images and anatomical information. In this work, detection and segmentation of vascular structures in the liver is performed automatically with clustering-based segmentation and refinement stages. RESULTS: The accuracy of the automatic labeling approach is 85 %. Required processing time for the proposed method (average 6 s) is shorter than manual approach (average 295 s) for labeling of hepatic and portal veins from segmented vessels. CONCLUSION: The proposed approach is efficient in terms of both computational cost and accuracy of labeling and segmentation of hepatic and portal veins.

Quantification of liver fat: A comprehensive review.

Fat accumulation in the liver causes metabolic diseases such as obesity, hypertension, diabetes or dyslipidemia by affecting insulin resistance, and increasing the risk of cardiac complications and cardiovascular disease mortality. Fatty liver diseases are often reversible in their early stage; therefore, there is a recognized need to detect their presence and to assess its severity to recognize fat-related functional abnormalities in the liver. This is crucial in evaluating living liver donors prior to transplantation because fat content in the liver can change liver regeneration in the recipient and donor. There are several methods to diagnose fatty liver, measure the amount of fat, and to classify and stage liver diseases (e.g. hepatic steatosis, steatohepatitis, fibrosis and cirrhosis): biopsy (the gold-standard procedure), clinical (medical physics based) and image analysis (semi or fully automated approaches). Liver biopsy has many drawbacks: it is invasive, inappropriate for monitoring (i.e., repeated evaluation), and assessment of steatosis is somewhat subjective. Qualitative biomarkers are mostly insufficient for accurate detection since fat has to be quantified by a varying threshold to measure disease severity. Therefore, a quantitative biomarker is required for detection of steatosis, accurate measurement of severity of diseases, clinical decision-making, prognosis and longitudinal monitoring of therapy. This study presents a comprehensive review of both clinical and automated image analysis based approaches to quantify liver fat and evaluate fatty liver diseases from different medical imaging modalities.

Fully automated liver segmentation using Sobolev gradient-based level set evolution.

Quantitative analysis and precise measurements on the liver have vital importance for pre-evaluation of surgical operations and require high accuracy in liver segmentation from all slices in a data set. However, automated liver segmentation from medical image data sets is more challenging than segmentation of any other organ due to various reasons such as vascular structures in the liver, high variability of liver shapes, similar intensity values, and unclear edges between liver and its adjacent organs. In this study, a variational level set-based segmentation approach is proposed to be efficient in terms of processing time and accuracy. The efficiency of this method is achieved by (1) automated initialization of a large initial contour, (2) using an adaptive signed pressure force function, and also (3) evolution of the level set with Sobolev gradient. Experimental results show that the proposed fully automated segmentation technique avoids local minima and stops evolution of the active contour at desired liver boundaries with high speed and accuracy. Copyright © 2016 John Wiley & Sons, Ltd.

Fully automated liver segmentation from SPIR image series.

Accurate liver segmentation is an important component of surgery planning for liver transplantation, which enables patients with liver disease a chance to survive. Spectral pre-saturation inversion recovery (SPIR) image sequences are useful for liver vessel segmentation because vascular structures in the liver are clearly visible in these sequences. Although level-set based segmentation techniques are frequently used in liver segmentation due to their flexibility to adapt to different problems by incorporating prior knowledge, the need to initialize the contours on each slice is a common drawback of such techniques. In this paper, we present a fully automated variational level set approach for liver segmentation from SPIR image sequences. Our approach is designed to be efficient while achieving high accuracy. The efficiency is achieved by (1) automatically defining an initial contour for each slice, and (2) automatically computing weight values of each term in the applied energy functional at each iteration during evolution. Automated detection and exclusion of spurious structures (e.g. cysts and other bright white regions on the skin) in the pre-processing stage increases the accuracy and robustness. We also present a novel approach to reduce computational cost by employing binary regularization of level set function. A signed pressure force function controls the evolution of the active contour. The method was applied to ten data sets. In each image, the performance of the algorithm was measured using the receiver operating characteristics method in terms of accuracy, sensitivity and specificity. The accuracy of the proposed method was 96%. Quantitative analyses of results indicate that the proposed method can accurately, efficiently and consistently segment liver images.