Diabetic retinopathy is an eye complication caused by a widespread disease named diabetes mellitus. The examination of retinal fundus images procured by retinography is the most commonly used method for diagnosing diabetic retinopathy. Strategies based on deep learning have shown promising results in detecting diabetic retinopathy, achieving performance similar to that of the human eye regarding image inspection. However, the performance of these strategies heavily depends on fine-tuning the algorithm hyper-parameters and big datasets. In this work, we propose training a Deep Learning model with Low-Rank Adaptation (LoRA) approach to classify three stages of Diabetic Retinopathy: i) no sign of diabetic retinopathy, ii) Non-proliferative diabetic retinopathy, and iii) proliferative diabetic retinopathy. We propose using a low-rank representation to reduce significantly the number of trainable parameters. The experiment shows that the LoRA approach for image classification of the three stages of diabetic retinopathy manages to obtain state-of-the-art results even with a small dataset.

Diabetic retinopathy is an eye complication of a widespread disease named diabetes mellitus. The most widely used method for diagnosing diabetic retinopathy is the analysis of retinal fundus images obtained by retinography. Deep Learning-based methods have shown promising results as a diagnostic tool for diabetic retinopathy, achieving, in some cases, performance close to the human inspection of images. However, the performance of these methods relies heavily on fine-tuning the algorithm hyperparameters and big data sets. In this work, we propose training a Deep Learning network with evolutionary algorithms to classify three stages of Diabetic Retinopathy: i) no sign of diabetic retinopathy, ii) Non-proliferative dia-betic retinopathy, and iii) proliferative diabetic retinopathy. We propose a neuroevolution methodology for selecting the most efficient Deep Learning model. The results of the neuroevolution methodology were improved by including Simulated Annealing strategies, Population Reinitialization, and ensembles. With high accuracy, sensitivity, specificity, and kappa index rates of 0.889, 0.889, 0.951, and 0.822, respectively, in the best case found, the experiments show that our neuroevolution methodology for selecting the Deep Learning model hyperparameters is a competitive alternative for training deep neural networks to classify three stages of diabetic retinopathy even with a small data set.

Image segmentation is one of the first steps in most image processing procedures. The segmentation aims to obtain a more meaningful or simplified image representation by grouping pixels with common characteristics, which allows regions or features of interest to be uniquely identified. The result of the segmentation has a significant impact on the subsequent steps. Segmentation is part of several superior applications such as artificial vision, medical, topographic, and astronomical image analysis. No single or universal segmentation process gets optimal performance for all image types. Hence, determining a function that fits specific image types or applications becomes a detailed, complex, and not trivial task requiring much time and effort. In this paper, we propose using Multi-Objective Evolutionary Algorithms (MOEAs) as a training tool that combines operations that represent the techniques and strategies commonly used for generating image segmentation. As a result, sequences of operations are suitable for specific applications or image types. The objective functions used to guide the evolutionary process are sensitivity maximization (TPR) and specificity maximization (TNR), the basic components of ROC analysis. Sensitivity and specificity are commonly used as classification metrics to evaluate the quality of a proposed segmentation compared to an ideal segmentation. We used sensitivity and specificity as objective functions rather than accuracy because, as stated in [1], the dependence on prevalence makes accuracy less effective than a simultaneous consideration of sensitivity and specificity. Experiments were conducted on multiple images that share common characteristics obtained from image databases, specifically: i) benign and malignant melanoma images, ii) ophthalmoscopic retinal images, and iii) binary cell form images, where the segmentation generated by the proposed algorithm was compared with ideal segmentation. The results are quite promising and show that using MOEAs to generate sequences of segmentation operations valid for specific applications is feasible.

Panoramic radiographs of teeth, also called orthopantomograms, play an important role in different tasks, from the diagnosis of diseases to the identification of persons in forensic dentistry. The segmentation of teeth from these radiographs can be used as a preliminary step for tasks such as counting, measure the similarity between two orthopantomograms and recognition of a person's teeth. These types of radiographs suffer from low contrast which can make it difficult to segment the teeth. In this paper we propose a methodology in which a contrast enhancement technique is used on orthopantomograms, and then used in the segmentation task using U-Nets. In the experiments, different image contrast enhancement methods were evaluated, obtaining better values in the F1-score metric compared to using no pre-processing before segmentation using U-nets. Finally, a correlation analysis of different metrics of image quality enhancement evaluation was performed, showing that contrast is positively correlated with the precision and accuracy of the segmentation obtained by U-nets.

For the training of blood pressure predictive models, it is necessary to determine the optimal number of predictors when the data set is of high dimensionality. Applying the appropriate dimensionality reduction technique according to the dataset will reduce the number of components and improve the performance of the predictive models. This work proposes the dimensionality reduction of the data set through the explorations of linear and nonlinear relationships of photoplethysmography signals by applying principal component analysis through the decomposition of Pearson and Spearman correlation matrices. The differences between the explained and cumulative variances of the principal components are minimal by applying Pearson and Spearman correlations. The predictive models trained with the first 5 principal components obtained better results for the estimation of blood pressure, with minimal loss of information with respect to the original data set.

Special devices capable of transforming continuous images into digital formats generate digital images. Image digitization enables treating, visualizing and storing images in a computer system. However, the high-intensity light captured by a device of this type can cause unwanted bright areas in the generated images. Since these bright areas can lead to image processing errors, manifested as a false appearance, their deletion and subsequent reconstruction can produce a more faithful image. A conventional approach, based on mathematical morphology, consists of reducing or eliminating unwanted image brightness by an erosion process. The image obtained through this method undergoes a reconstruction process based on successive geodesic dilations. In this paper, a new brightness reduction method is proposed. This method identifies bright areas by applying a defined brightness limit value to remove pixels with brightness values above it, but leaving the remaining pixels unmodified. Then, the image is reconstructed by the usual image reconstruction approach based on mathematical morphology. Compared to the conventional approach, the proposed method in this work enhances brightness, generating an image more faithful to the actual object.

The acquisition of microscopic images of minerals with good contrast is critical for the identification and analysis of their properties. However, in many cases, the microscopic images of minerals obtained are unclear due to the image environment, imperfect adjustment of the microscopy operators or improper collection of samples. In this paper, we present an algorithm to enhance the microscopic images of minerals by multiscale Top-Hat transform using contrast adjustment weights. First, the multiple dark and bright features of the mineral image are extracted using the top-hat transform. Secondly, bright scale differences and dark scale differences obtained in the previous step are calculated. Third, all the intensities of the multiple dark and bright features from the previous steps are summed separately. Finally, the bright features adjusted for a contrast weight are then added to the image and dark features adjusted for the same weight are subtracted from the image. Experimental results on various kinds of microscopic mineral images verified the effective performance of this proposed enhancing the contrast, improving the detail and spatial information about the images

This work investigated and implemented the integration and effective use of Computer Aided Diagnostic and Detection systems (CAD) with a Picture Archiving and Communication System (PACS) and how this could provide standardized data and expand the coverage of specialized professionals. Likewise, standardized data allowed interoperability between systems. The designed CAD is focused on Diabetic Retinopathy diagnosis with three main functions: (I) digital image processing for detection, segmentation and extraction of retina image anomaly features, (II) binary classification, and finally (III) converter of the results to the DICOM standard (Digital Imagine and Communication Systems). It was decided to use an open source and free license PACS called Orthanc. The results show the successful integration of the systems, since the studies entered in the CAD are accessed directly from the PACS. At the same time, usability tests show a high degree of efficiency, effectiveness and satisfaction of clinical users in the use of the system.

Infrared images (IR) help us to detect hidden targets in the environment, according to the radiation they emit. These work well on the day, at night and in weather conditions such as rain or fog. At the same time, visible images (VIS) provide us with good details of the scenes, which are better perceived by the human eye. Therefore, the fusion of an infrared image and a visible image of the same scene is very useful. In this paper, we propose a fusion method of visible and infrared images using a multiscale morphological approach. First, the base image is generated through the fusion of the two source images. Second, multiple bright and dark features are extracted from each source image by the top-hat transform. Third, the multiple bright and dark scales of the source images are fused. Finally, the fusion of the visible and infrared images is obtained by adding to the base image the maximum values obtained in the previous step. The results show that the proposed method is competitive compared to state-of-the-art methods in terms of contrast, brightness, texture and spatial information.

Extracting useful image features and preserving details effectively is a crucial part of fusion of images. Infrared images can distinguish objects from their background based on the difference in radiation. In the other hand, visible images can provide textured details consistent with the human visual system. The fusion of these two types of images can combine the advantages of thermal radiation information in infrared images and detailed texture information in visible images. In this work, we propose an algorithm of fusion of infrared and visible images using the multiscale top-hat transformation. The extraction of bright and dark regions from the images is done using two structuring elements. This algorithm provides significantly better results in contrast, brightness and texture than other state-of-theart algorithms.

The current process of searching for terminology for medical coding in health standards in the Hospital of Clinics of Paraguay, is done through coding manuals or through the internet of cell phones. This process takes a long time to be done during the medical consultation. The optimization of the current process would allow the search of encoded terminologies during the medical consultation, avoiding that the resident doctors have to perform this extra work after the workday. This work implements a medical terminology server through web services and using the text search engine library: Apache Lucene that is specialized in storage agility and data recovery. The implemented server also allows to perform successful searches of terminologies encoded in standards through friendly or familiar terms. The search time of coded terminologies using the implemented server was compared with the search time of the current process in the hospital of clinics using the internet of cell phones. Under these criteria, the terminology server responds up to 18 times faster than the current search process. In addition, it was made a comparison of the server against a search engine, called Metamorphosys. The terminology server implemented turned out to be quite competitive against Metamorphosys by presenting quite similar response times. Although both tools have a similar average response time, Metamorphosys presents some high atypical values in some algorithms that it presents as search options. The degree of satisfaction of the user in the use of the system was evaluated through a usability questionnaire called “System Usability Scale” (SUS). The evaluation rated the system as “Good” according to the Bangor.

Searching for adequate input parameters of a Contrast Enhancement Algorithm is a fundamental task, aimed to get more suitable images, in terms of contrast, for decision making or further applications. Several Optimization approaches take a single image as the input of the process, so they get poor performance. In this proposal, a group of several images of the same type are taken as input of a Robust Multi-Objective Particle Swarm Optimization approach, in order to achieve more suitable input parameters for a kind of images, instead of a single image. The results are a set of input parameters for the Contrast Enhancement Algorithm, suitable for Contrast Enhancement of a group of images of the same type.

Locating objects or people in an indoor environment quickly, accurately and at low cost is a great need nowadays and in different scenarios. Some examples are the location of products in a warehouse or the quick location of patients, medical personnel or equipment in a hospital. A location system is necessary for health care, home care, stock control or inventory. In this context, this paper presents the design and implementation of a low-cost centralized indoor location system that uses BLE technology (Bluetooth Low Energy) together with the particle filter algorithm. Experimental results show that the system can achieve an accuracy of 1.8 m, at best, with an accuracy of 74% within 3 m. The cost of the infrastructure goes hand in hand with the number of objects or people to be located.

Morphological mathematics applied to color images requires a color space and an ordering method to be able to define a complete lattice. Among the drawbacks in the methods proposed in the current literature we find some that produce false colors, or are inefficient at computation time. False colors are undesirable in the process of transformation as they alter original colors of the image, turning them into noise. On the other hand, any vector sorting method must be efficient enough to meet the requirements of procedures that do not allow for lengthy response times. In this work we propose a method of order based on Loewners order, using a transformation of the colors associated with the pixels of an image in the RGB space to symmetric matrices. In order to bring about a comparison of the results obtained through the proposed method and the others, we tested on a consistent basis, measuring the computation time in the process of basic operations, such as dilation and erosion, checked the generation of no false colors, removal of noise (smoothing), brightness reduction and local contrast enhancement, all in the same computing environment. The results demonstrate the validity of the method, resulting in efficiency of time and quality results, as well as producing no false colors.

Melanoma is a type of skin cancer and is caused by the uncontrolled growth of atypical melanocytes. In recent decades, computer aided diagnosis is used to support medical professionals; however, there is still no globally accepted tool. In this context, similar to state-of-the-art we propose a system that receives a dermatoscopy image and provides a diagnostic if the lesion is benign or malignant. This tool is based on next modules: Preprocessing, Segmentation, Feature Extraction and Classification. Preprocessing involves the removal of hairs. Segmentation is to isolate the lesion. Feature extraction is considering the ABCD dermoscopy rule. The classification is performed by the Support Vector Machine. Experimental evidence indicates that the proposal has 90.63 % accuracy, 95 % sensitivity and 83.33 % specificity on a dataset of 104 dermatoscopy images. These results are favorable considering the performance of diagnosis by traditional progress in the area of dermatology.

Due to its versatility, the image processing area offers a very wide range of techniques to solve challenges in an effective way linked to fields such as medicine, agriculture, biology, industrial automation and document processing. Therefore a correct and advanced training of professionals in this area is an important task. In this sense, a new educational image processing tool is currently being developed at the Facultad Politecnica of the Universidad Nacional of Asuncion. The development focused on improving the interaction between students and teachers and also showing new advances in digital image processing area. To achieve this goal, a tool is being developed to be expanded in the future to be adapted to new challenges and different audiences. The first stage of development was completed, which allowed developing an extendable basic tool in the near future.

The hemoflagellate protozoan parasite Trypanosoma cruzi is the causative agent of the Chagas disease. The two drugs used clinically have a high level of toxicity and are active only during the acute phase of the disease, making it urgent the development of new safe and effective treatment. The first step in the screening of new compounds is to identify their effectiveness by manual microscopic counting of the intracellular parasite form (amastigotes), which is a slow and tedious methodology. This paper presents an approach for the automatic counting of intracellular parasites using watershed transform with internal and external markers as segmentation technique, and connected components labeling for subsequent counting. The contrast against the classic counting, conducted by experts in the field, validated the technique, showing that the proposal is very efficient and with a low error rate.