Unbiased Stereology. Three‐Dimensional Measurement in Microscopy.

Establishing an accurate count and classification of leukocytes commonly known as WBC (white blood cells) is crucial in the assessment and detection of illness of an individual, which involves complications on the immune system that leads to various types of diseases including infections, anemia, leukemia, cancer, AIDS (Acquired Immune Deficiency Syndrome) etc. The two widely used methods to count WBC is with the use of hematology analyzer and manual counting. Currently, in the age of modernization there has been numerous research in the field of image processing incorporated with various segmentation and classification techniques to be able to generate alternatives for WBC classification and counting. However, the accuracy of these existing methods could still be improved. Thus, in this paper we proposed a new method that could segment various types of WBCs: monocytes, lymphocytes, eosinophils, basophils, and neutrophils from a microscopic blood image using HSV (Hue, Saturation, Value) saturation component with blob analysis for segmentation and incorporate CNN (Convolutional Neural Network) for counting which in turn generates more accurate results.

White blood cell classification and counting using convolutional neural network

Problem statement: Various images are low quality and difficultly to detect and extract information. Therefore, the image has to get under a process called image enhancement which contains an aggregation of techniques that look for improving the visual aspect of an image. Medical images are one of the fundamental images, because they are used in more sensitive field which is a medical field. The raw data obtained straight from devices of medical acquisition may afford a comparatively poor image quality representation and may destroy by several types of noises. Image Enhancement (IE) and denoising algorithms for executing the requirements of digital medical image enhancement is introduced. The main goal of this study is to improve features and gain better characteristics of medical images for a right diagnosis. Approach: The proposed techniques start by the median filter for removing noise on images followed by unsharp mask filter which is believable the usual type of sharpening. Medical images were usually poor quality especially in contrast. For solving this problem, we proposed Contrast Limited Adaptive Histogram Equalization (CLAHE) which is one of the techniques in a computer image processing domain. It was used to amend contrast in images. Results: For testing purposes, different sizes and various types of medical images were used and more than 60 images in different parts of the body. From the experts’ evaluation, they noted that the enhanced images improved up to 80% from the original images depends on medical images modalities. Conclusion: The proposed algorithms results were significant for increasing the visibleness of relatively details without distorting the images.

/pdf/improving-diagnostic-viewing-of-medical-images-using-546t9xqjqy.pdf

Improving Diagnostic Viewing of Medical Images using Enhancement Algorithms

White Blood Cells play an important role in observing the health condition of an individual. The opinion related to blood disease involves the identification and characterization of a patient's blood sample. Recent approaches employ Convolutional Neural Network (CNN), Recurrent Neural Network (RNN) and merging of CNN and RNN models to enrich the understanding of image content. From beginning to end, training of big data in medical image analysis has encouraged us to discover prominent features from sample images. A single cell patch extraction from blood sample techniques for blood cell classification has resulted in the good performance rate. However, these approaches are unable to address the issues of multiple cells overlap. To address this problem, the Canonical Correlation Analysis (CCA) method is used in this paper. CCA method views the effects of overlapping nuclei where multiple nuclei patches are extracted, learned and trained at a time. Due to overlapping of blood cell images, the classification time is reduced, the dimension of input images gets compressed and the network converges faster with more accurate weight parameters. Experimental results evaluated using publicly available database show that the proposed CNN and RNN merging model with canonical correlation analysis determines higher accuracy compared to other state-of-the-art blood cell classification techniques.

White Blood Cells Image Classification Using Deep Learning with Canonical Correlation Analysis

Density prediction for petroleum and derivatives by gamma-ray attenuation and artificial neural networks

Haar-Cascade classifier method has been applied to detect the presence of a human on the thermal image. The evaluation was done on the performance of detection, represented by its precision and recall values. The thermal camera images were varied to obtain comprehensive results, which covered the distance of the object from the camera, the angle of the camera to the object, the number of objects, and the environmental conditions during image acquisition. The results showed that the greater the camera-object distance, the precision and recall of human detection results declined. Human objects would also be hard to detect if his/her pose was not facing frontally. The method was able to detect more than one human in the image with positions of in front of each other, side by side, or overlapped to one another. However, if there was any other object in the image that had characteristics similar to a human, the object would also be detected as a human being, resulting in a false detection. These other objects could be an infrared shadow formed from the reflection on glass or painted walls.

Thermal image human detection using Haar-cascade classifier

An X-ray Compton scatter method for density measurement at a point within an object

Information about counts and percentages of each type of leukocytes in blood is much needed to diagnose patients' illness. To gain that information, some functional enhancements had been applied to the optical microscopes so that they could produce digital images. Output images from these engineered microscopes were then extracted to get feature values from each image. These feature values then became input sets to the method of K-Means Clustering so that the leukocyte images could be classified according to each own cluster. Generally, the leukocyte classification process is conducted through four phases, which are image pre-processing, leukocyte segmentation, feature extraction, and leukocyte classification. Leukocyte types which were classified in this research were neutrophil, lymphocyte, monocyte, and eosinophil. Experiments were conducted using five kinds of features, which are normalized area, circularity, eccentricity, normalized parameter, and solidity, and by varying their types and their significant influences. The purpose of these trials were to determine which feature types would result in the highest value of accuracy and the effects of adding these respective features to the resulted accuracy. Based on the conducted classification results, it was found that the highest accuracy value was reached by circularity feature, which was 67%, meanwhile the lowest accuracy value was produced by the eccentricity feature, which was 43%. In this research, it was concluded that the accuracy value is ultimately determined by selecting the correct feature type rather than adding more features.

Classification of leukocyte images using K-Means Clustering based on geometry features

Haar Cascade Classifier is a method for detection of objects within an image, which is widely applied on face detection. This paper discusses the utilization of Haar Cascade Classifier in locating the positions of white blood cells in an image. The results showed that this method is able to localize white blood cells with precision and recall values of 95% and 74% respectively. This method is also able to distinguish white blood cells from other objects that have color resembling white blood cells.

Localization of white blood cell images using Haar Cascade classifiers

The research aimed at developing plant growth monitoring system based on image processing technique from images captured using multi-cameras (webcams). The output of the research was a monitoring system equipped with graphical user interface (GUI) to monitor the plant growth and nondestructively predict age and fresh weight of mustard (Brassica rapa var. parachinensis L.). Five webcams on five different positions were installed on the image-capturing box to capture the plant images every 3 days up to 48 days. Then, 1050 images of 70 plants on vegetative period resulted. Of these, 700 images were used for training and 350 images for validation. An artificial neural network (ANN) was used to predict age and weight of the plants. The network architecture consisted of four layers with five input neurons, first hidden layer with five neurons, second hidden layer with five neurons, and output layers with one neuron. Training function of ANN used was trainlm and learning rate value was 0.001. The activation function for predicting plant age and plant weight was logsig-logsig-tansig and logsig-tansig-tansig, respectively. ANN models could predict plant age with an R2 value of 0.96 and fresh weight with an R2 value of 0.95.

Balza Achmad

Papers

Thermal image human detection using Haar-cascade classifier

An X-ray Compton scatter method for density measurement at a point within an object

Classification of leukocyte images using K-Means Clustering based on geometry features

Localization of white blood cell images using Haar Cascade classifiers

Development of Plant Growth Monitoring System Using Image Processing Techniques Based on Multiple Images