| 1. |
- Bengtsson, Ewert, 1948-, et al.
(författare)
-
Detection of Malignancy-Associated Changes Due to Precancerous and Oral Cancer Lesions: A Pilot Study Using Deep Learning
- 2018
-
Ingår i: CYTO2018.
-
Konferensbidrag (refereegranskat)abstract
- Background: The incidence of oral cancer is increasing and it is effecting younger individuals. PAP smear-based screening, visual, and automated, have been used for decades, to successfully decrease the incidence of cervical cancer. Can similar methods be used for oral cancer screening? We have carried out a pilot study using neural networks for classifying cells, both from cervical cancer and oral cancer patients. The results which were reported from a technical point of view at the 2017 IEEE International Conference on Computer Vision Workshop (ICCVW), were particularly interesting for the oral cancer cases, and we are currently collecting and analyzing samples from more patients. Methods: Samples were collected with a brush in the oral cavity and smeared on glass slides, stained, and prepared, according to standard PAP procedures. Images from the slides were digitized with a 0.35 micron pixel size, using focus stacks with 15 levels 0.4 micron apart. Between 245 and 2,123 cell nuclei were manually selected for analysis for each of 14 datasets, usually 2 datasets for each of the 6 cases, in total around 15,000 cells. A small region was cropped around each nucleus, and the best 2 adjacent focus layers in each direction were automatically found, thus creating images of 100x100x5 pixels. Nuclei were chosen with an aim to select well preserved free-lying cells, with no effort to specifically select diagnostic cells. We therefore had no ground truth on the cellular level, only on the patient level. Subsets of these images were used for training 2 sets of neural networks, created according to the ResNet and VGG architectures described in literature, to distinguish between cells from healthy persons, and those with precancerous lesions. The datasets were augmented through mirroring and 90 degrees rotations. The resulting networks were used to classify subsets of cells from different persons, than those in the training sets. This was repeated for a total of 5 folds. Results: The results were expressed as the percentage of cell nuclei that the neural networks indicated as positive. The percentage of positive cells from healthy persons was in the range 8% to 38%. The percentage of positive cells collected near the lesions was in the range 31% to 96%. The percentages from the healthy side of the oral cavity of patients with lesions ranged 37% to 89%. For each fold, it was possible to find a threshold for the number of positive cells that would correctly classify all patients as normal or positive, even for the samples taken from the healthy side of the oral cavity. The network based on the ResNet architecture showed slightly better performance than the VGG-based one. Conclusion: Our small pilot study indicates that malignancyassociated changes that can be detected by neural networks may exist among cells in the oral cavity of patients with precancerous lesions. We are currently collecting samples from more patients, and will present those results as well, with our poster at CYTO 2018.
|
|
| 2. |
- Bengtsson, Ewert, 1948-
(författare)
-
Image processing and its hardware support : Analysis vs synthesis - historical trends
- 2017
-
Ingår i: Image Analysis, SCIA 2017, Pt I. - Switzerland : Springer International Publishing. - 9783319591261 - 9783319591254 ; , s. 3-14
-
Konferensbidrag (refereegranskat)abstract
- Computers can be used to handle images in two fundamen-tally dierent ways. They can be used to analyse images to obtain quan-titative data or some image understanding. And they can be used tocreate images that can be displayed through computer graphics and vi-sualization. For both of these purposes it is of interest to develop ecientways of representing, compressing and storing the images. While SCIA,the Scandinavia Conference of Image Analysis, according to its name ismainly concerned with the former aspect of images, it is interesting tonote that image analysis throughout its history has been strongly in u-enced also by developments on the visualization side. When the confer-ence series now has reached its 20th milestone it may be worth re ectingon what factors have been important in forming the development of theeld. To understand where you are it is good to know where you comefrom and it may even help you understand where you are going.
|
|
| 3. |
- Bengtsson, Ewert, 1948-
(författare)
-
Recognizing signs of malignancy : The quest for computer assisted cancer screening and diagnosis systems
- 2010
-
Ingår i: International Conference on Computational Intelligence and Computing Research (ICCIC), 2010 IEEE. - Coimbatore, India : IEEE Digital Library. - 9781424459650 ; , s. 1-6
-
Konferensbidrag (refereegranskat)abstract
- Almost all cancers are diagnosed through visual examination of microscopic tissue samples. Visual screening of cell samples, so called PAP-smears, has drastically reduced the incidence of cervical cancers in countries that have implemented population wide screening programs. But the visual examination is tedious, subjective and expensive. There has therefore been much research aiming for computer assisted or automated cell image analysis systems for cancer detection and diagnosis. Progress has been made but still most of cytology and pathology is done visually. In this presentation I will discuss some of the major issues involved, examine some of the proposed solutions and give some comments about the state of the art.
|
|
| 4. |
- Gavrilovic, Milan, 1981-, et al.
(författare)
-
Dimensionality Reduction for Colour Based Pixel Classification
- 2009
-
Ingår i: Proceedings SSBA 2009. - Halmstad : Halmstad University. - 9789163339240 ; , s. 65-68
-
Konferensbidrag (övrigt vetenskapligt/konstnärligt)abstract
- In digital images, providing classification based on colour, hue or spectral angle is a problem usually solved by combining a variety of pre-processing steps, as well as object wise classifiers. We have developed a method for transforming colour or multispectral image data to a 1D colour histogram with respect to the digital characteristics of intensity measurements. Classification is then reduced to 1D histogram segmentation which is a simpler problem. The proposed method, based on ideas of spectral decomposition, was previously applied in dual-colour fluorescence microscopy for quantification and detection of colocalization insensitive to cross-talk. In this paper the principle is expanded to unsupervised colour based pixel classification algorithms in hue-saturation-lightness or luminance-chrominance colour spaces.
|
|
| 5. |
- Gavrilovic, Milan, 1981-, et al.
(författare)
-
Spectral Angle Histogram : a Novel Image Analysis Tool for Quantification of Colocalization and Cross-talk
- 2009
-
Ingår i: 9th International ELMI Meeting on Advanced Light Microscopy. - Glasgow, UK. ; , s. 66-67
-
Konferensbidrag (övrigt vetenskapligt/konstnärligt)abstract
- In fluorescence microscopy, when analyzing spectral components, it is common to record two (or more) greyscale images. Each greyscale image, referred to as a channel, corresponds to intensities in different wavelength intervals. If each pixel of a two-channel image is plotted in a space spanned by the two intensity channels a conventional scatter-plot is obtained. Single-coloured pixels are distributed along the axes, while colocalized pixels are distributed closer to the diagonal of the scatter-plot, and cross-talk (as well as noise) is observed as deviations of the single-coloured vectors from the axes. Detection of colocalized pixels is often based on a division of this 2D space into different regions by intensity thresholding. We have developed a method for reducing the scatter-plot to a 1D spectral angle histogram through a series of steps that compensate for the quantization noise which is always present in digital image data.Using the spectral angle histogram, we can quantify colocalization in a fully automated and robust manner. As compared to previous methods for quantification of colocalization, this approach is insensitive to cross-talk. In fact, it can also be employed to quantify and compensate for cross-talk, using either linear unmixing or fuzzy classification by spectral angle, ensuring complete suppression of cross-talk with minimal loss of information. Recently we started investigating how the method can deal with autofluorescence. Initial tests on real image data show that the method may be useful for improved background suppression and amplification of the true signals.The article “Quantification of colocalization and cross-talk based on spectral angles”, describing the method, is about to be published in the Journal of Microscopy. Authors have also filed a patent application “Pixel classification in image analysis” in 2008.
|
|
| 6. |
- Issac Niwas, Swamidoss, et al.
(författare)
-
Analysis of nuclei textures of fine needle aspirated cytology images for breast cancer diagnosis using complex Daubechies wavelets
- 2013
-
Ingår i: Signal Processing. - : Elsevier BV. - 0165-1684 .- 1872-7557. ; 93:10, s. 2828-2837
-
Tidskriftsartikel (refereegranskat)abstract
- Breast cancer is the most frequent cause of cancer induced death among women in the world. Diagnosis of this cancer can be done through radiological, surgical, and pathological assessments of breast tissue samples. A common test for detection of this cancer involves visual microscopic inspection of Fine Needle Aspiration Cytology (FNAC) samples of breast tissue. The result of analysis on this sample by a cytopathologist is crucial for the breast cancer patient. For the assessment of malignancy, the chromatin texture patterns of the cell nuclei are essential. Wavelet transforms have been shown to be good tools for extracting information about texture. In this paper, it has been investigated whether complex wavelets can provide better performance than the more common real valued wavelet transform. The features extracted through the wavelets are used as input to a k-nn classifier. The correct classification results are obtained as 93.9% for the complex wavelets and 70.3% for the real wavelets.
|
|
| 7. |
- Koriakina, Nadezhda, 1991-, et al.
(författare)
-
Visualization of convolutional neural network class activations in automated oral cancer detection for interpretation of malignancy associated changes
- 2019
-
Ingår i: 3rd NEUBIAS Conference, Luxembourg, 2-8 February 2019.
-
Konferensbidrag (övrigt vetenskapligt/konstnärligt)abstract
- Introduction: Cancer of the oral cavity is one of the most common malignancies in the world. The incidence of oral cavity and oropharyngeal cancer is increasing among young people. It is noteworthy that the oral cavity can be relatively easily accessed for routine screening tests that could potentially decrease the incidence of oral cancer. Automated deep learning computer aided methods show promising ability for detection of subtle precancerous changes at a very early stage, also when visual examination is less effective. Although the biological nature of these malignancy associated changes is not fully understood, the consistency of morphology and textural changes within a cell dataset could shed light on the premalignant state. In this study, we are aiming to increase understanding of this phenomenon by exploring and visualizing what parts of cell images are considered as most important when trained deep convolutional neural networks (DCNNs) are used to differentiate cytological images into normal and abnormal classes.Materials and methods: Cell samples are collected with a brush at areas of interest in the oral cavity and stained according to standard PAP procedures. Digital images from the slides are acquired with a 0.32 micron pixel size in greyscale format (570 nm bandpass filter). Cell nuclei are manually selected in the images and a small region is cropped around each nucleus resulting in images of 80x80 pixels. Medical knowledge is not used for choosing the cells but they are just randomly selected from the glass; for the learning process we are only providing ground truth on the patient level and not on the cell level. Overall, 10274 images of cell nuclei and the surrounding region are used to train state-of-the-art DCNNs to distinguish between cells from healthy persons and persons with precancerous lesions. Data augmentation through 90 degrees rotations and mirroring is applied to the datasets. Different approaches for class activation mapping and related methods are utilized to determine what image regions and feature maps are responsible for the relevant class differentiation.Results and Discussion:The best performing of the observed deep learning architectures reaches a per cell classification accuracy surpassing 80% on the observed material. Visualizing the class activation maps confirms our expectation that the network is able to learn to focus on specific relevant parts of the sample regions. We compare and evaluate our findings related to detected discriminative regions with the subjective judgements of a trained cytotechnologist. We believe that this effort on improving understanding of decision criteria used by machine and human leads to increased understanding of malignancy associated changes and also improves robustness and reliability of the automated malignancy detection procedure.
|
|
| 8. |
- Mehnert, Andrew, 1967, et al.
(författare)
-
Two non-linear parametric models of contrast enhancement for DCE-MRI of the breast amenable to fitting using linear least squares
- 2010
-
Ingår i: Proc. 2010 International Conference on Digital Image Computing: Techniques and Applications (DICTA). - IEEE Computer Society, Los Alamitos, CA, USA : IEEE Computer Society. - 9781424488162 ; , s. 611-616
-
Konferensbidrag (refereegranskat)abstract
- This paper proffers two non-linear empirical parametric models - linear slope and Ricker - for use in characterising contrast enhancement in dynamic contrast enhanced (DCE) MRI. The advantage of these models over existing empirical parametric and pharmacokinetic models is that they can be fitted using linear least squares (LS). This means that fitting is quick, there is no need to specify initial parameter estimates, and there are no convergence issues. Furthermore the LS fit can itself be used to provide initial parameter estimates for a subsequent NLS fit (self-starting models). The results of an empirical evaluation of the goodness of fit (GoF) of these two models, measured in terms of both MSE and R2, relative to a two-compartment pharmacokinetic model and the Hayton model are also presented. The GoF was evaluated using both routine clinical breast MRI data and a single high temporal resolution breast MRI data set. The results demonstrate that the linear slope model fits the routine clinical data better than any of the other models and that the two parameter self-starting Ricker model fits the data nearly as well as the three parameter Hayton model. This is also demonstrated by the results for the high temporal data and for several temporally sub-sampled versions of this data.
|
|
| 9. |
- Runow Stark, Christina, et al.
(författare)
-
Brush Biopsy For HR-HPV Detection With FTA Card And AI For Cytology Analysis - A Viable Non-invasive Alternative
- 2018
-
Ingår i: EAOM2018.
-
Konferensbidrag (refereegranskat)abstract
- Introduction: Oral cancer accounts for about 800-1,000 new cases each year in Sweden and the ratio of cancer related to high-risk human papillomavirus (HR-HPV) is increasing in the younger population due to changes in sexual habits. The most two frequent HR-HPV types 16 and 18 have both significant oncogenic potential.Objectives: In this pilot study we evaluate two non-invasive automated methods; 1) detection of HR-HPV using FTA cards, and 2) image scanning of cytology for detection of premalignant lesions as well as eradicate the early stage of neoplasia.Material and Methods: 160 patients with verified HR-HPV oropharyngeal cancer, previous ano-genital HR-HPV-infection or potentially malignant oral disorder were recruited for non-invasive brush sampling and analyzed with two validated automated methods both used in cervix cancer screening. For analysis of HR-HPV DNA the indicating FTA elute micro cardTM were used for dry collection, transportation and storage of the brush samples. For analysis of cell morphology changes an automated liquid base Cytology method (Preserve Cyt) combined with deep learning computer aided technique was used.Results: Preliminary results show that the FTA-method is reliable and indicates that healthy and malignant brush samples can be separated by image analysis. Conclusions: With further development of these fully automated methods, it is possible to implement a National Screening Program of the oral mucosa, and thereby select patients for further investigation in order to find lesions with potential malignancy in an early stage.
|
|
| 10. |
- Walhagen, Peter, et al.
(författare)
-
AI based prostate analysis system trained without human supervision to predict patient outcome from tissue samples
- 2022
-
Ingår i: Journal of Pathology Informatics. - : Elsevier BV. - 2229-5089 .- 2153-3539.
-
Tidskriftsartikel (refereegranskat)abstract
- In order to plan the best treatment for prostate cancer patients the aggressiveness of the tumor is graded based on visual assessment of tissue biopsies according to the Gleason scale. Recently a number of AI models have been developed that can be trained to do this grading as well as human pathologists. But the accuracy of the AI grading will be limited by the accuracy of the subjective “ground truth†Gleason grades used for the training. We have trained an AI to predict patient outcome directly based on image analysis of a large biobank of tissue samples with known outcome without input of any human knowledge about cancer grading. The model has shown similar and in some cases better ability to predict patient outcome on an independent test-set than expert pathologists doing the conventional grading.
|
|
