| 1051. |
- Mahbod, Amirreza, et al.
(författare)
-
Investigating and Exploiting Image Resolution for Transfer Learning-based Skin Lesion Classification
- 2021
-
Ingår i: 2020 25TH INTERNATIONAL CONFERENCE ON PATTERN RECOGNITION (ICPR). - : IEEE Computer Society. ; , s. 4047-4053
-
Konferensbidrag (refereegranskat)abstract
- Skin cancer is among the most common cancer types. Dermoscopic image analysis improves the diagnostic accuracy for detection of malignant melanoma and other pigmented skin lesions when compared to unaided visual inspection. Hence, computer-based methods to support medical experts in the diagnostic procedure are of great interest. Fine-tuning pre-trained convolutional neural networks (CNNs) has been shown to work well for skin lesion classification. Pre-trained CNNs are typically trained with natural images of a fixed image size significantly smaller than captured skin lesion images and consequently dermoscopic images are downsampled for fine-tuning. However, useful medical information may be lost during this transformation. In this paper, we explore the effect of input image size on skin lesion classification performance of fine-tuned CNNs. For this, we resize dermoscopic images to different resolutions, ranging from 64 x 64 to 768 x 768 pixels and investigate the resulting classification performance of three well-established CNNs, namely DenseNet-121, ResNet-18, and ResNet-50. Our results show that using very small images (of size 64 x 64 pixels) degrades the classification performance, while images of size 128 x 128 pixels and above support good performance with larger image sizes leading to slightly improved classification. We further propose a novel fusion approach based on a three-level ensemble strategy that exploits multiple fine-tuned networks trained with dermoscopic images at various sizes. When applied on the ISIC 2017 skin lesion classification challenge, our fusion approach yields an area under the receiver operating characteristic curve of 89.2% and 96.6% for melanoma classification and seborrheic keratosis classification, respectively, outperforming state-of-the-art algorithms.
|
|
| 1052. |
|
|
| 1053. |
- Mahmud, Tanjim, et al.
(författare)
-
Exploring Deep Transfer Learning Ensemble for Improved Diagnosis and Classification of Alzheimer’s Disease
- 2023
-
Ingår i: BI 2023: Brain Informatics, Proceedings. - : Springer. ; , s. 109-120
-
Konferensbidrag (refereegranskat)abstract
- Alzheimer’s disease (AD) is a progressive and irreversible neurological disorder that affects millions of people worldwide. Early detection and accurate diagnosis of AD are crucial for effective treatment and management of the disease. In this paper, we propose a transfer learning-based approach for the diagnosis of AD using magnetic resonance imaging (MRI) data. Our approach involves extracting relevant features from the MRI data using transfer learning by alter the weights and then using these features to train pre-trained models and combined ensemble classifier. We evaluated our approach on a dataset of MRI scans from patients with AD and healthy controls, achieving an accuracy of 95% for combined ensemble models. Our results demonstrate the potential of transfer learning-based approaches for the early and accurate diagnosis of AD, which could lead to improved patient outcomes and more effective management of the disease.
|
|
| 1054. |
- Majtner, Tomáš, et al.
(författare)
-
On the Effectiveness of Generative Adversarial Networks as HEp-2 Image Augmentation Tool
- 2019
-
Ingår i: Scandinavian Conference on Image Analysis. - Cham : Springer International Publishing. - 9783030202040 ; , s. 439-451
-
Konferensbidrag (refereegranskat)abstract
- One of the big challenges in the recognition of biomedical samples is the lack of large annotated datasets. Their relatively small size, when compared to datasets like ImageNet, typically leads to problems with efficient training of current machine learning algorithms. However, the recent development of generative adversarial networks (GANs) appears to be a step towards addressing this issue. In this study, we focus on one instance of GANs, which is known as deep convolutio nal generative adversarial network (DCGAN). It gained a lot of attention recently because of its stability in generating realistic artificial images. Our article explores the possibilities of using DCGANs for generating HEp-2 images. We trained multiple DCGANs and generated several datasets of HEp-2 images. Subsequently, we combined them with traditional augmentation and evaluated over three different deep learning configurations. Our article demonstrates high visual quality of generated images, which is also supported by state-of-the-art classification results.
|
|
| 1055. |
- Maksuti, Elira, et al.
(författare)
-
ARTERIAL STIFFNESS ESTIMATION BY SHEAR WAVE ELASTOGRAPHY : VALIDATION IN PHANTOMS WITH MECHANICAL TESTING
- 2016
-
Ingår i: Ultrasound in Medicine and Biology. - : Elsevier BV. - 0301-5629 .- 1879-291X. ; 42:1, s. 308-321
-
Tidskriftsartikel (refereegranskat)abstract
- Arterial stiffness is an independent risk factor found to correlate with a wide range of cardiovascular diseases. It has been suggested that shear wave elastography (SWE) can be used to quantitatively measure local arterial shear modulus, but an accuracy assessment of the technique for arterial applications has not yet been performed. In this study, the influence of confined geometry on shear modulus estimation, by both group and phase velocity analysis, was assessed, and the accuracy of SWE in comparison with mechanical testing was measured in nine pressurized arterial phantoms. The results indicated that group velocity with an infinite medium assumption estimated shear modulus values incorrectly in comparison with mechanical testing in arterial phantoms (6.7 +/- 0.0 kPa from group velocity and 30.5 +/- 0.4 kPa from mechanical testing). To the contrary, SWE measurements based on phase velocity analysis (30.6 +/- 3.2 kPa) were in good agreement with mechanical testing, with a relative error between the two techniques of 8.8 +/- 6.0% in the shear modulus range evaluated (40-100 kPa). SWE by phase velocity analysis was validated to accurately measure stiffness in arterial phantoms.
|
|
| 1056. |
- Maksuti, Elira, et al.
(författare)
-
Contribution of the Arterial System and the Heart to Blood Pressure during Normal Aging : A Simulation Study
- 2016
-
Ingår i: PLOS ONE. - : Public Library of Science (PLoS). - 1932-6203. ; 11:6
-
Tidskriftsartikel (refereegranskat)abstract
- During aging, systolic blood pressure continuously increases over time, whereas diastolic pressure first increases and then slightly decreases after middle age. These pressure changes are usually explained by changes of the arterial system alone (increase in arterial stiffness and vascular resistance). However, we hypothesise that the heart contributes to the age-related blood pressure progression as well. In the present study we quantified the blood pressure changes in normal aging by using a Windkessel model for the arterial system and the time-varying elastance model for the heart, and compared the simulation results with data from the Framingham Heart Study. Parameters representing arterial changes (resistance and stiffness) during aging were based on literature values, whereas parameters representing cardiac changes were computed through physiological rules (compensated hypertrophy and preservation of end-diastolic volume). When taking into account arterial changes only, the systolic and diastolic pressure did not agree well with the population data. Between 20 and 80 years, systolic pressure increased from 100 to 122 mmHg, and diastolic pressure decreased from 76 to 55 mmHg. When taking cardiac adaptations into account as well, systolic and diastolic pressure increased from 100 to 151 mmHg and decreased from 76 to 69 mmHg, respectively. Our results show that not only the arterial system, but also the heart, contributes to the changes in blood pressure during aging. The changes in arterial properties initiate a systolic pressure increase, which in turn initiates a cardiac remodelling process that further augments systolic pressure and mitigates the decrease in diastolic pressure.
|
|
| 1057. |
- Malhi, Avleen, et al.
(författare)
-
Explaining Machine Learning-based Classifications of in-vivo Gastral Images
- 2019
-
Ingår i: 2019 Digital Image Computing. - : IEEE. - 9781728138572 ; , s. 530-536
-
Konferensbidrag (refereegranskat)abstract
- This paper proposes an explainable machine learning tool that can potentially be used for decision support in medical image analysis scenarios. For a decision-support system it is important to be able to reverse-engineer the impact of features on the final decision outcome. In the medical domain, such functionality is typically required to allow applying machine learning to clinical decision making. In this paper, we present initial experiments that have been performed on in-vivo gastral images obtained from capsule endoscopy. Quantitative analysis has been performed to evaluate the utility of the proposed method. Convolutional neural networks have been used for training the validating of the image data set to provide the bleeding classifications. The visual explanations have been provided in the images to help health professionals trust the black box predictions. While the paper focuses on the in-vivo gastral image use case, most findings are generalizable.
|
|
| 1058. |
- Malm, Patrik, et al.
(författare)
-
Debris removal in Pap-smear images
- 2013
-
Ingår i: Computer Methods and Programs in Biomedicine. - : Elsevier BV. - 0169-2607 .- 1872-7565. ; 111:1, s. 128-138
-
Tidskriftsartikel (refereegranskat)abstract
- Since its introduction in the 1940s the Pap-smear test has helped reduce the incidence of cervical cancer dramatically in countries where regular screening is standard. The automation of this procedure is an open problem that has been ongoing for over fifty years without reaching satisfactory results. Existing systems are discouragingly expensive and yet they are only able to make a correct distinction between normal and abnormal samples in a fraction of cases. Therefore, they are limited to acting as support for the cytotechnicians as they perform their manual screening. The main reason for the current limitations is that the automated systems struggle to overcome the complexity of the cell structures. Samples are covered in artefacts such as blood cells, overlapping and folded cells, and bacteria, that hamper the segmentation processes and generate large number of suspicious objects. The classifiers designed to differentiate between normal cells and pre-cancerous cells produce unpredictable results when classifying artefacts. In this paper, we propose a sequential classification scheme focused on removing unwanted objects, debris, from an initial segmentation result, intended to be run before the actual normal/abnormal classifier. The method has been evaluated using three separate datasets obtained from cervical samples prepared using both the standard Pap-smear approach as well as the more recent liquid based cytology sample preparation technique. We show success in removing more than 99% of the debris without loosing more than around one percent of the epithelial cells detected by the segmentation process.
|
|
| 1059. |
- Malm, Patrik, 1982-
(författare)
-
Image Analysis in Support of Computer-Assisted Cervical Cancer Screening
- 2013
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Cervical cancer is a disease that annually claims the lives of over a quarter of a million women. A substantial number of these deaths could be prevented if population wide cancer screening, based on the Papanicolaou test, were globally available. The Papanicolaou test involves a visual review of cellular material obtained from the uterine cervix. While being relatively inexpensive from a material standpoint, the test requires highly trained cytology specialists to conduct the analysis. There is a great shortage of such specialists in developing countries, causing these to be grossly overrepresented in the mortality statistics. For the last 60 years, numerous attempts at constructing an automated system, able to perform the screening, have been made. Unfortunately, a cost-effective, automated system has yet to be produced.In this thesis, a set of methods, aimed to be used in the development of an automated screening system, are presented. These have been produced as part of an international cooperative effort to create a low-cost cervical cancer screening system. The contributions are linked to a number of key problems associated with the screening: Deciding which areas of a specimen that warrant analysis, delineating cervical cell nuclei, rejecting artefacts to make sure that only cells of diagnostic value are included when drawing conclusions regarding the final diagnosis of the specimen. Also, to facilitate efficient method development, two methods for creating synthetic images that mimic images acquired from specimen are described.
|
|
| 1060. |
- Malm, Patrik, et al.
(författare)
-
Simulation of bright-field microscopy images depicting pap-smear specimen
- 2015
-
Ingår i: Cytometry Part A. - : Wiley. - 1552-4922 .- 1552-4930. ; 87:3, s. 212-226
-
Tidskriftsartikel (refereegranskat)abstract
- As digital imaging is becoming a fundamental part of medical and biomedical research, the demand for computer-based evaluation using advanced image analysis is becoming an integral part of many research projects. A common problem when developing new image analysis algorithms is the need of large datasets with ground truth on which the algorithms can be tested and optimized. Generating such datasets is often tedious and introduces subjectivity and interindividual and intraindividual variations. An alternative to manually created ground-truth data is to generate synthetic images where the ground truth is known. The challenge then is to make the images sufficiently similar to the real ones to be useful in algorithm development. One of the first and most widely studied medical image analysis tasks is to automate screening for cervical cancer through Pap-smear analysis. As part of an effort to develop a new generation cervical cancer screening system, we have developed a framework for the creation of realistic synthetic bright-field microscopy images that can be used for algorithm development and benchmarking. The resulting framework has been assessed through a visual evaluation by experts with extensive experience of Pap-smear images. The results show that images produced using our described methods are realistic enough to be mistaken for real microscopy images. The developed simulation framework is very flexible and can be modified to mimic many other types of bright-field microscopy images.
|
|
