Publications

Analyzing the Inference Process in Deep Convolutional Neural Networks using Principal Eigenfeatures, Saturation and Logistic Regression Probes

Published in Journal of Applied Research in Electrical Engineering, 2023

The predictive performance of a neural network depends on the one hand on the difficulty of a problem, defined by the number of classes and complexity of the visual domain, and on the other hand on the capacity of the model, determined by the number of parameters and its structure. By applying layer saturation and logistic regression probes, we confirm that these factors influence the inference process in an antagonistic manner. This analysis allows the detection of over- and under-parameterization of convolutional neural networks. We show that the observed effects are independent of previously reported pathological patterns, like the “tail pattern”. In addition, we study the emergence of saturation patterns during training, showing that saturation patterns emerge early in the optimization process. This allows for quick detection of problems and potentially decreased cycle time during experiments. We also demonstrate that the emergence of tail patterns is independent of the capacity of the networks. Finally, we show that information processing within a tail of unproductive layers is different, depending on the topology of the neural network architecture.

Recommended citation: Richter M.L, Malihi. L, Windler, A.K.P., Krumnack. U. (2023). "Analyzing the Inference Process in Deep Convolutional Neural Networks using Principal Eigenfeatures, Saturation and Logistic Regression Probes." Journal of Applied Research in Electrical Engineering. 1(1). http://leilamalihi.github.io/files/papermat.pdf

Can Synthetic Images Improve CNN Performance in Wound Image Classification?

Published in Medical Informatic Europe (MIE)., 2023

Machine Learning (ML) based AI systems require a large amount of labelled training data to achieve this level. In cases of a shortage of such large amounts, Generative Adversarial Networks (GAN) are a standard tool for synthesising artificial training images that can be used to augment the data set. We investigated the quality of synthetic wound images regarding two aspects: (i) improvement of wound-type classification by a Convolutional Neural Network (CNN) and (ii) how realistic such images look to clinical experts (n = 217). Concerning (i), results show a slight classification improvement. However, the connection between classification performance and the size of the artificial data set is still unclear. Regarding (ii), although the GAN could produce highly realistic images, the clinical experts took them for real in only 31% of the cases. It can be concluded that image quality may play a more significant role than data size in improving the CNN-based classification result.

Recommended citation: MALIHI.L., HÜBNER. U.. (2023). "Can Synthetic Images Improve CNN Performance in Wound Image Classification?." Medical Informatic Europe (MIE).. 1(1). https://ebooks.iospress.nl/doi/10.3233/SHTI230311

An Image Based Object Recognition System for Wound Detection and Classification of Diabetic Foot and Venous Leg Ulcer

Published in Medical Informatic Europe (MIE), 2022

Venous leg ulcers and diabetic foot ulcers are the most common chronic wounds. Their prevalence has been increasing significantly over the last years, consuming scarce care resources. This study aimed to explore the performance of detection and classification algorithms for these types of wounds in images. To this end, algorithms of the YoloV5 family of pre-trained models were applied to 885 images containing at least one of the two wound types. The YoloV5m6 model provided the highest precision (0.942) and a high recall value (0.837). Its mAP_0.5:0.95 was 0.642. While the latter value is comparable to the ones reported in the literature, precision and recall were considerably higher. In conclusion, our results on good wound detection and classification may reveal a path towards (semi-) automated entry of wound information in patient records. To strengthen the trust of clinicians, we are currently incorporating a dashboard where clinicians can check the validity of the predictions against their expertise.

Recommended citation: Hüsers. J., Moeleken. M., Richter. M. , Przysucha. M., Malihi. L.,. "An Image Based Object Recognition System for Wound Detection and Classification of Diabetic Foot and Venous Leg Ulcer." Medical Informatic Europe (MIE). 1(2). https://ebooks.iospress.nl/doi/10.3233/SHTI220397

Automatic Wound Type Classification with Convolutional Neural Networks

Published in International Conference on Informatics, Management, and Technology in Healthcare (ICIMTH), 2022

Chronic wounds are ulcerations of the skin that fail to heal because of an underlying condition such as diabetes mellitus or venous insufficiency. The timely identification of this condition is crucial for healing. However, this identification requires expert knowledge unavailable in some care situations. Here, artificial intelligence technology may support clinicians. In this study, we explore the performance of a deep convolutional neural network to classify diabetic foot and venous leg ulcers using wound images. We trained a convolutional neural network on 863 cropped wound images. Using a hold-out test set with 80 images, the model yielded an F1-score of 0.85 on the cropped and 0.70 on the full images. This study shows promising results. However, the model must be extended in terms of wound images and wound types for application in clinical practice.

Recommended citation: Malihi. L., Hüsers. J.. (2022). "Automatic Wound Type Classification with Convolutional Neural Networks." International Conference on Informatics, Management, and Technology in Healthcare (ICIMTH). 1(3). https://pubmed.ncbi.nlm.nih.gov/35773863/

Exploring the Properties and Evolution of Neural Network Eigenspaces during Training

Published in Machine Vision and Image Processing (MVIP), 2021

We investigate properties and the evolution of the emergent inference process inside neural networks using layer saturation. and logistic regression probes. We demonstrate that the difficulty of a problem, defined by the number of classes and complexity of the visual domain, as well as the number of parameters in neural network layers affect the predictive performance in an antagonistic manner. We further show that this relationship can be measured using saturation. This opens the possibility of detecting over- and under-parameterization of neural networks. We further show that the observed effects are independent of previously reported pathological patterns like the “tail pattern” described in [1]. Finally, we study the emergence of saturation patterns during training, showing that saturation patterns emerge early during training. This allows for early analysis and potentially increased cycle-time during experiments.

Recommended citation: Your Name, You. (2009). "Paper Title Number 1." Machine Vision and Image Processing (MVIP). 1(1). https://ieeexplore.ieee.org/document/9738741

Single stuck-at-faults detection using test generation vector and deep stacked-sparse-autoencoder

Published in SN applied science Springer Journal, 2020

This paper proposed a new method for testing digital circuits without hardware implementation. This data-based method detects hundreds of single stuck-at faults in the ALU circuits, utilizing deep stacked-sparse-autoencoder (SSAE). ATALANTA software is one of the free automatic test pattern generation tools which cover faults in high accuracy. Test vectors which are extracted from bench circuits via ATALANTA software are the key point of the paper. Fault detection is introduced as a two-class problem. SSAE network is trained using the test vectors. Dimension reduction is done automatically in SSAE. Network performance is tested by changing sparse coefficients, number of stacked autoencoder and data augmentation. The results of this step are compared with the traditional multilayer perceptron classification. In this method, unlike SSAE, a manual method of reducing the dimension and extracting the feature is used. Fault coverage of ATALANTA software is over than 94%. Finally, the results obtained from the deep neural network show its significant performance in the circuit faults detection automatically.

Recommended citation: Malihi. L., Malihi. R. . (2020). "Paper Title Number 1." SN applied science Springer Journal. 1(1). https://link.springer.com/article/10.1007/s42452-020-03460-0

Design of wide band Microstrip-line-fed antenna with circular polarization at ISM band for biomedical applications

Published in 10th International Conference on Information Technology, 2020

In this paper, a new wide band Micro strip antenna for implantable biomedical application in the frequency of ISM (industrial, scientific and medical) band (2.4–2.48 GHz) is presented. The total size of the proposed antenna is 20 × 20 mm2 with the thickness of 1 mm , and this antenna is embedded in FR4 substrate with dielectric constant of 4.4. The antenna parameters show good results such as lower return loss, perfect impedance matching, and better gain. The 3-dB axial ratio bandwidth is 2.2GHZ( from 1.2 -3.4 GHz) that is broader than the other conventional antennas. The proposed antenna possesses the return loss of -29 dB at 2.4GHz. Thus the proposed antenna can be employed for several implantable applications.

Recommended citation: Your Name, You. (2009). "Design of wide band Microstrip-line-fed antenna with circular polarization at ISM band for biomedical applications" 10th International Conference on Information Technology. 1(1). http://leilamalihi.github.io/files/paper.pdf

New monopole planar implant antenna for medical application

Published in 5th international conference on information technology, 2018

The design and characterization of a microstrip-fed planar monopole antenna with circular polarization is presented. The antenna operates in the Industrial, Scientific and Medical (ISM) (5.725–5.875GHz),and WLAN (5150–5350 MHz) with return loss lower than -10 dB. The antenna has a compact aperture size 20 × 20 mm2, fabricated on FR4 substrate with dielectric constant of 4.4, thickness of 1 mm. The advantages of the proposed antenna are the simple yet efficient design of the radiator, a 3-dB axial-ratio operating band with a compact size.

Recommended citation: Malihi. R.,Malihi. L., Noorinia.J.. (2018). "New monopole planar implant antenna for medical application." 5th international conference on information technology. 1(1). https://civilica.com/doc/843751/

Airport extraction from satellite images

Published in International conference on fundamental research in electrical engineering, 2017

In accurate remote sensing, automatic airport extraction through satellite images can be the best approach to prepare, edit, and update an efficient database.Therefore, in this article, a method for automatic airport extraction from satellite images is proposed. In the proposed method, the pre-processing of the input images was done first. For this purpose, a Bilateral filter has been used to reduce noise. This filter is more efficient than traditional Haas noise reduction methods such as Gaussian and preserves edges well. Also, in order to make the histogram uniform, histogram adjustment has been used.In addition, the Atsu method has been used to complete this method. Then, the holes of the extraction mask are filled and the extra parts are removed using the morphology closure operator with the appropriate element. Finally, a mask was obtained that showed the area of ​​the airport well. In order to check the proposed method in more detail, various categories were used in this field. For this purpose, various features were extracted from the image and given to 7 different classifications and the performance of each one was checked. The best performance was the KNN classification with 45.95% accuracy.

Recommended citation: Ebadi.S.M., PoorMohamad.A. ,Malihi.L. (2017). "Airport extraction from satellite images." International conference on fundamental research in electrical engineering. 1(1). https://civilica.com/doc/672785/

Published in , 1900

Improvement in Classification Accuracy Rate Using Multiple Classifier Fusion towards Computer Vision Detection of Malaria Parasite (Plasmodium vivax)

Published in Ajums Journals, 2015

Background: The main method used for the laboratory confirmation of malaria is the conventional light microscopy; however, microscopy has three main disadvantages: I) it is time-consuming and labor-intensive; II) its results depend heavily on good techniques, reagents and microscopes; III) in many cases decisions about treatment are often taken without using the result of microscopy because of long delays in providing the results to the clinician. Hence, an extreme necessity of the fast automatic detection of the disease is required to diagnose and treat promptly. Objectives: Through the improvement of classification accuracy rate, this work aims to present a computer-assisted diagnosis system for malaria parasite. Materials and Methods: This study was conducted using 400 confirmed images of blood slides infected with malaria parasite. The MATLAB software was used for the implementation of computation procedures. Using five extracted features (flat texture, saturation channel histogram, color histogram, gradient, and granulometry) and six classifiers (k-Nearest Neighbors (k-NN), 1-Nearest Neighbor (1-NN), decision tree (DT), Fisher, linear discriminant analysis (LDA), and quadratic discriminant analysis (QDA)), images were classified into two classes: parasitic and nonparasitic. Then, classifier fusion was done using several algorithms: mean, min, max, stack, median, Adaboost, and bagging.

Recommended citation: Malihi. L., Ansari-Asl. K., Behbahani. A (2015). "Improvement in Classification Accuracy Rate Using Multiple Classifier Fusion towards Computer Vision Detection of Malaria Parasite (Plasmodium vivax)." Ajums Journals. 1(1). https://brieflands.com/articles/jjhs-15009.pdf

Computer-Aided Diagnosis of Malaria Parsite using Patern Recognition Methods

Published in Ajums Journals, 2015

In many cases of paraitic identification by visual inspection is difficult, time consuming and depends heavily on the experience of microscopists. Computer-aided diagnosis can make a significant help in saving the time, reducing workforces and the possible operator errors. The aim of this study was to assess the performance of four classifiers for detection of malaria parasite was investigated. Subjects and Methods:A total of 400 images of malaria parasite-infected blood slides were used.Intially by masking the red blood cells, in order to match the stained extracted elements, only red blood cells were used for next stage of the study. Then, the color histogram, granulometry, texture, saturation level histogram, gradient and flat­ texture features were extracted. For discriminating parasitic images from non-parasitic images four classifiers have been used: K-Nearest Neighbors (KNN), Nearest Mean (NM), 1-Nearest Neighbors (1NN), and Fisher linear discriminator (Fisher). Results: The best classification accuracy of 92.5%, which was achieved by KNN classifier. The accuracies of 1-NN, Fisher and NM classifiers were 90.25%, 85%, and 60.25%, respectively.

Recommended citation: Malihi. L., Ansari-Asl. K., Behbahani. A.. (2015). "Computer-Aided Diagnosis of Malaria Parsite using Patern Recognition Methods." Ajums Journals. 1(1). https://jsmj.ajums.ac.ir/article_46919.html?lang=en

Malaria Parasite Detection in Giemsa-stained Blood Cell Image

Published in Machine Vision and Image Processing (MVIP), 2013

Malaria is one of the leading causes of death, especially in high-risk groups die infants, toddlers, and pregnant women. In the world of almost 1 million people die because of it every year. Malaria is transmitted by the bite of a female Anopheles mosquito vectors that have been infected by Plasmodium. Identification of Plasmodium in the blood is done by visual observation blood cells using a microscope. To aid this process, some research are carried out to develop identification plasmodium using methods based computer aided diagnosis (CAD) and digital image processing. The infected cells are extracted typically using image grayscale. In this study, image on the pigment color space and color space are compared so that the intensity will be obtained an optimum color channels in the process of the appearance of the parasite plasmodium features. This study used a sample of Giemsa-stained blood cell images which are infected with the malaria parasite. the results of this study stated after the histogram on the image that the results are graphed evenly and no dominant, and the image looks more clear.

Recommended citation: Malihi. L., Ansari-Asl. K., Behbahani. A. (2013). "Malaria Parasite Detection in Giemsa- stained Blood Cell Image." Machine Vision and Image Processing (MVIP). 1(1). https://ieeexplore.ieee.org/document/8089291