Wyniki wyszukiwania - BazTech

1

Segmentation of bone structures with the use of deep learning techniques

Krawczyk Zuzanna, Starzyński Jacek

Bulletin of the Polish Academy of Sciences. Technical Sciences

|

2021

|

Vol. 69, nr 3

art. no. e136751

EN

The paper is focused on automatic segmentation task of bone structures out of CT data series of pelvic region. The authors trained and compared four different models of deep neural networks (FCN, PSPNet, U-net and Segnet) to perform the segmentation task of three following classes: background, patient outline and bones. The mean and class-wise Intersection over Union (IoU), Dice coefficient and pixel accuracy measures were evaluated for each network outcome. In the initial phase all of the networks were trained for 10 epochs. The most exact segmentation results were obtained with the use of U-net model, with mean IoU value equal to 93.2%. The results where further outperformed with the U-net model modification with ResNet50 model used as the encoder, trained by 30 epochs, which obtained following result: mIoU measure – 96.92%, “bone” class IoU – 92.87%, mDice coefficient – 98.41%, mDice coefficient for “bone” – 96.31%, mAccuracy – 99.85% and Accuracy for “bone” class – 99.92%.

2

Multi-feature ensemble system in the renal tumour classification task

Osowska-Kurczab Aleksandra Maria, Markiewicz Tomasz, Dziekiewicz Miroslaw, Lorent Malgorzata

Bulletin of the Polish Academy of Sciences. Technical Sciences

|

2021

|

Vol. 69, nr 3

art. no. e136749

EN

Recently, the analysis of medical imaging is gaining substantial research interest, due to advancements in the computer vision field. Automation of medical image analysis can significantly improve the diagnosis process and lead to better prioritization of patients waiting for medical consultation. This research is dedicated to building a multi-feature ensemble model which associates two independent methods of image description: textural features and deep learning. Different algorithms of classification were applied to single-phase computed tomography images containing 8 subtypes of renal neoplastic lesions. The final ensemble includes a textural description combined with a support vector machine and various configurations of Convolutional Neural Networks. Results of experimental tests have proved that such a model can achieve 93.6% of weighted F1-score (tested in 10-fold cross validation mode). Improvement of performance of the best individual predictor totalled 3.5 percentage points.

3

Deep learning-based framework for tumour detection and semantic segmentation

Kot Estera, Krawczyk Zuzanna, Siwek Krzysztof, Królicki Leszek, Czwarnowski Piotr

Bulletin of the Polish Academy of Sciences. Technical Sciences

|

2021

|

Vol. 69, nr 3

art. no. e136750

EN

For brain tumour treatment plans, the diagnoses and predictions made by medical doctors and radiologists are dependent on medical imaging. Obtaining clinically meaningful information from various imaging modalities such as computerized tomography (CT), positron emission tomography (PET) and magnetic resonance (MR) scans are the core methods in software and advanced screening utilized by radiologists. In this paper, a universal and complex framework for two parts of the dose control process – tumours detection and tumours area segmentation from medical images is introduced. The framework formed the implementation of methods to detect glioma tumour from CT and PET scans. Two deep learning pre-trained models: VGG19 and VGG19-BN were investigated and utilized to fuse CT and PET examinations results. Mask R-CNN (region-based convolutional neural network) was used for tumour detection – output of the model is bounding box coordinates for each object in the image – tumour. U-Net was used to perform semantic segmentation – segment malignant cells and tumour area. Transfer learning technique was used to increase the accuracy of models while having a limited collection of the dataset. Data augmentation methods were applied to generate and increase the number of training samples. The implemented framework can be utilized for other use-cases that combine object detection and area segmentation from grayscale and RGB images, especially to shape computer-aided diagnosis (CADx) and computer-aided detection (CADe) systems in the healthcare industry to facilitate and assist doctors and medical care providers.

4

A new method of cardiac sympathetic index estimation using a 1D-convolutional neural network

Kołodziej Marcin, Majkowski Andrzej, Tarnowski Paweł, Rak Remigiusz Jan, Rysz Andrzej

Bulletin of the Polish Academy of Sciences. Technical Sciences

|

2021

|

Vol. 69, nr 3

art. no. e136921

EN

Epilepsy is a neurological disorder that causes seizures of many different types. The article presents an analysis of heart rate variability (HRV) for epileptic seizure prediction. Considering that HRV is nonstationary, our research focused on the quantitative analysis of a Poincare plot feature, i.e. cardiac sympathetic index (CSI). It is reported that the CSI value increases before the epileptic seizure. An algorithm using a 1D-convolutional neural network (1D-CNN) was proposed for CSI estimation. The usability of this method was checked for 40 epilepsy patients. Our algorithm was compared with the method proposed by Toichi et al. The mean squared error (MSE) for testing data was 0.046 and the mean absolute percentage error (MAPE) amounted to 0.097. The 1D-CNN algorithm was also compared with regression methods. For this purpose, a classical type of neural network (MLP), as well as linear regression and SVM regression, were tested. In the study, typical artifacts occurring in ECG signals before and during an epileptic seizure were simulated. The proposed 1D-CNN algorithm estimates CSI well and is resistant to noise and artifacts in the ECG signal.

5

Improving accuracy of detecting dangerous objects with deep learning

Zacniewski A.

Autobusy : technika, eksploatacja, systemy transportowe

|

2016

|

R. 17, nr 12

513--516

EN

In this article, the problem of detecting dangerous objects with deep learning is presented. Convolutional Neural Networks are created with Python language ecosystem (Theano and Keras libraries), and then trained with different number of layers and different parameters. Accuracy of detection dangerous objects for artificial Neural Network with smaller number of layers is computed and obtained result is improved with deep learning. CIFAR-10 dataset is used due to useful classes included.

PL

W artykule przedstawiono problem detekcji niebezpiecznych obiektów przy użyciu technik deep learning. Konwolucyjne sieci neuronowe tworzone są przy pomocy bibliotek języka Python takich jak Keras i Theano, a następnie trenowane są przy różnej liczbie warstw i z różnymi parametrami. Skuteczność detekcji niebezpiecznych obiektów dla małej liczby warstw sztucznej sieci neuronowej jest obliczana, a uzyskany wynik jest ulepszany przy użyciu technik deep learning. Zbiór danych CIFAR-10 został wykorzystany w badaniach z powodu dużej użyteczności występujących w nim klas.