smart.lab.tunisia@gmail.com +216 71 588 443
Connexion

Hana Mechria

Informations générales

Hana Mechria
Grade

Post Doc

Axes de recherche

Deep Learning
Health care

Publications

  • 2025
    Hana Mechria, Khaled Hassine, Mohamed Salah Gouider

    Mammogram images denoising based on deep convolutional neural network

    Imapct Factor 2024: 3.6, 2025

    Résumé

    Mammogram images are subject to various types of noise, which restricts the analysis of images and diagnosis. Mammogram image denoising is very important to improve image quality and to make the segmentation and classification results more correct. In this work, we propose a Deep Convolutional Neural Network (DCNN) to denoise the mammogram images in order to improve the image quality by handling Gaussian, Speckle, Poisson, and Salt and Pepper noise. The main objective of this study is to remove different types of noises from mammogram images and to maximize the quantity of information content in the enhanced images. We first add noise models to mammogram images and then enhance the image by removing the noise using DCNN. Furthermore, we compare our results with state-of-the-art denoising methods, such as the Adaptive Median filter, Wiener filter, Gaussian filter, Median filter, and Mean. Three datasets have been used, including Digital Database for Screening Mammography (DDSM), mini-Mammographic Image Analysis Society (mini-MIAS), and a local Tunisian dataset. The experimental results show that DCNN has a better denoising performance than the other methods, with an average PSNR range of 46.0-51.83 dB and an average SSIM range of 0.988-99.83, which may suggest its adaptability to different models of noise.

  • Hana Mechria, Khaled Hassine, Mohamed Salah Gouider

    Effect of Denoising on Performance of Deep Convolutional Neural Network For Mammogram Images Classification

    KES, 2022

    Résumé

    Digital mammograms are an important imaging modality for breast cancer screening and diagnosis. Several types of noise appear in mammograms and make the job of detecting breast cancer even more challenging due to missing details in the information of the image.
    In this study, we analyze the effect of mammogram images quality on the performance of the Deep Convolutional Neural Network on a mammogram images classification task. Thus, our objective is to show how the classification accuracy varies with the application of a denoising step.
    Indeed, we investigated two different approaches to breast cancer detection. The first is the classification of the original mammo-gram images without being denoised, and the second is the classification of mammogram images that are denoised using a Deep Convolutional Neural Network, Wiener filter and Median filter. Therefore, the mammogram images are first denoised using each of the three denoising methods and then, classified into two classes: cancer and normal, using AlexNet, a pre-trained Deep Convo-lutional Neural Network in order to show whether the denoising method used is effective when grafted onto a Deep Convolutional Neural Network by measuring accuracy, sensitivity, and specificity.
    Interesting results are achieved where the DCNN denoising step improved the Deep Convolutional Neural Network classification task with an increase of 3.47% for overall accuracy, 5.34% for overall specificity, and 0.56% for overall sensitivity.
  • Hana Mechria, Mohamed Salah Gouider, Khaled Hassine

    Breast Cancer Detection using Deep Convolutional Neural Network

    ICAART, 2019

    Résumé

    Deep Convolutional Neural Network (DCNN) is considered as a popular and powerful deep learning algorithm in image classification. However, there are not many DCNN applications used in medical imaging, because large dataset for medical images is not always available. In this paper, we present two DCNN architectures, a shallow DCNN and a pre-trained DCNN model: AlexNet, to detect breast cancer from 8000 mammographic images extracted from the Digital Database for Screening Mammography. In order to validate the performance of DCNN in breast cancer detection using a big data , we carried out a comparative study with a second deep learning algorithm Stacked AutoEncoders (SAE) in terms accuracy, sensitivity and specificity. The DCNN method achieved the best results with 89.23% of accuracy, 91.11% of sensitivity and 87.75% of specificity.