Publikasjoner
-
Bakken, Marianne; Kvam, Johannes; Stepanov, Alexey & Berge, Asbjørn (2020). Principal Feature Visualisation in Convolutional Neural Networks. Lecture Notes in Computer Science (LNCS).
ISSN 0302-9743.
12368, s 18- 31 . doi:
10.1007/978-3-030-58592-1_2
Vis sammendrag
We introduce a new visualisation technique for CNNs called Principal Feature Visualisation (PFV). It uses a single forward pass of the original network to map principal features from the final convolutional layer to the original image space as RGB channels. By working on a batch of images we can extract contrasting features, not just the most dominant ones with respect to the classification. This allows us to differentiate between several features in one image in an unsupervised manner. This enables us to assess the feasibility of transfer learning and to debug a pre-trained classifier by localising misleading or missing features.
-
Kvam, Johannes; Holm, Sverre & Angelsen, Bjørn Atle Johan (2019). Exploiting Ballou's rule for better tissue classification. Journal of the Acoustical Society of America.
ISSN 0001-4966.
145(4), s 2103- 2112 . doi:
10.1121/1.5096533
-
Kvam, Johannes; Solberg, Stian; Myhre, Ola Finneng; Rodriguez-Molares, Alfonso & Angelsen, Bjørn Atle Johan (2019). Nonlinear bulk elasticity imaging using dual frequency ultrasound. Journal of the Acoustical Society of America.
ISSN 0001-4966.
146(4), s 2492- 2500 . doi:
10.1121/1.5129120
-
Kvam, Johannes; Gangsei, Lars Erik; Kongsro, Jørgen & Solberg, Anne H Schistad (2018). The use of deep learning to automate the segmentation of the skeleton from CT volumes of pigs. Translational Animal Science.
ISSN 2573-2102.
2(3), s 324- 335 . doi:
10.1093/tas/txy060
-
Flørenæs, Even; Solberg, Stian; Myhre, Ola Finneng; Kvam, Johannes; Brende, Ole Martin & Angelsen, Bjørn Atle Johan (2017). In-vitro detection of micro calcifications using dual band ultrasound. Proceedings - IEEE Ultrasonics Symposium.
ISSN 1948-5719.
. doi:
10.1109/ULTSYM.2017.8092857
-
Kvam, Johannes & Kongsro, Jørgen (2017). In vivo prediction of intramuscular fat using ultrasound and deep learning. Computers and Electronics in Agriculture.
ISSN 0168-1699.
142, s 521- 523 . doi:
10.1016/j.compag.2017.11.020
-
Myhre, Ola Finneng; Kvam, Johannes & Angelsen, Bjørn Atle J. (2016). Dual frequency transducer design for suppression of multiple scattering. Proceedings - IEEE Ultrasonics Symposium.
ISSN 1948-5719.
2016-November . doi:
10.1109/ULTSYM.2016.7728429
-
Kvam, Johannes; Angelsen, Bjørn Atle J. & Elster, Anne C. (2015). GPU simulation of nonlinear propagation of dual band ultrasound pulse complexes. AIP Conference Proceedings.
ISSN 0094-243X.
1685:070003 . doi:
10.1063/1.4934440
Se alle arbeider i Cristin
-
Bakken, Marianne; Kvam, Johannes & Berge, Asbjørn (2020). Fast reasoning visualization for deep convolutional networks.
-
Bakken, Marianne; Kvam, Johannes; Stepanov, Alexey & Berge, Asbjørn (2020). Principal Feature Visualisation in Convolutional Neural Networks.
-
Flørenæs, Even; Solberg, Stian; Kvam, Johannes; Myhre, Ola Finneng; Brende, Ole Martin & Angelsen, Bjørn Atle Johan (2017). In vitro detection of microcalcifications using dual band ultrasound.
-
Solberg, Stian; Hansen, Rune; Berg, Sigrid; Kvam, Johannes & Angelsen, Bjørn Atle Johan (2017). In-vitro contrast agent detection combining pulse inversion and SURF imaging. Proceedings - IEEE Ultrasonics Symposium.
ISSN 1948-5719.
. doi:
10.1109/ULTSYM.2017.8092574
Vis sammendrag
Microbubbles as an ultrasound contrast agent have great diagnostic value, and can be used to visualize the vascularization in a variety of organs, e.g. imaging the neo-angiogenesis in tumors. Pulse inversion (PI) is a common method for detection of these bubbles, implemented in many ultrasound scanners. A dual band imaging technique named SURF has shown the ability to further enhance contrast-to-tissue ratio (CTR) in imaging of microbubbles. A new method is proposed here that combines the two techniques, potentially resulting in even better CTR.
-
Kvam, Johannes; Solberg, Stian; Brende, Ole Martin; Myhre, Ola Finneng; Rodriguez-Molares, Alfonso; Kongsro, Jørgen & Angelsen, Bjørn Atle J. (2016). Tissue Characterization with SURF Imaging.
Vis sammendrag
Tissue characterization has become an essential part of diagnosing diseases. Elasticity imaging in particular has become an integral part as tissue elasticity is often associated with pathological condition. There are two types of elasticity in soft tissue, bulk (volume) elasticity and shear (deformation) elasticity. Many diseases affect both, making measurement of these diagnostically interesting. Imaging methods such as elastography and acoustic radiation force methods have shown promise, but imaging at deep depths is challenging for both. The bulk elasticity can be difficult to measure, but it has a nonlinear component which might be possible to measure in a pulse echo situation. By imaging the tissue with different manipulations of bulk elasticity the nonlinear component can be extracted. Second-order ultrasound field (SURF) imaging is a dual band imaging technique that utilizes a low frequency (LF) pulse to manipulate the nonlinear bulk elasticity of the medium observed by a co-propagating high frequency (HF) pulse. The manipulation of the material causes the HF to experience an increased (or decreased) propagation velocity which depend on the manipulation pressure and nonlinear elasticity of the medium. The offset in propagation velocity causes an accumulative delay or advancement compared to a single HF pulse. This effect is called the nonlinear propagation delay (NPD). By transmitting multiple pulse complexes with different LF polarities, SURF utilizes the NPD to suppress multiple scattering noise. However, as the NPD is given by the interaction between the LF manipulation pressure and the medium, it is possible to estimate the local nonlinear bulk elasticity of the medium using the same scheme. Simulations have shown that it is possible to distinguish between materials by estimating local variations in the NPD gradient. Some in-vitro and in-vivo experiments have also shown promising results. The NPD is a bi-product of the conventional SURF image processing. This makes it possible to obtain tissue characterization data without requiring a mode switch from SURF imaging. The SURF image and material parameters can be extracted from data from the same acquisition. The method does not rely on the generation of an acoustic radiation force, as the manipulation is introduced by the LF wave. Due to the low frequency the absorption is negligible, causing the depth limit to be determined by the HF maximum B-mode imaging depth.
-
Myhre, Ola Finneng; Kvam, Johannes & Angelsen, Bjørn Atle J. (2016). Dual Frequency Transducer Design for Suppression of Multiple Scattering.
Vis sammendrag
Reverberation noise reduces the contrast resolution in ultrasound images. In some cases the signal from the anatomy can be completely masked out, making diagnosis challenging. SURF (Second-order Ultrasound Field) imaging is a dual band imaging technique that has shown the capability to suppress reverberation noise and enhance the anatomical signal, providing a higher contrast resolution. SURF transducers radiate pulse complexes comprised of two widely separated frequencies through a partially common radiation surface. In order to achieve optimal noise suppression, careful design of the acoustic stack and radiation apertures is needed. This paper presents the optimisation criteria for SURF probes, and describes design solutions using a 9/0.5 MHz linear array for carotid imaging as an example. Simulated transfer functions are compared to those of a manufactured probe.
-
Yemane, Petros Tesfamichael; Angelsen, Bjørn Atle J.; Kvam, Johannes; Afadzi, Mercy; Myhre, Ola Finneng & Davies, Ruth Catharina de Lange (2016). Simulation of ultrasound radiation force: for transport of drugs and nanoparticles in tumors.. The 8th National PhD Conference in Medical Imaging.
-
Kvam, Johannes; Angelsen, Bjørn Atle J. & Elster, Anne C. (2015). GPU Simulation of Nonlinear Propagation of Dual Band Ultrasound Pulse Complexes.
Vis sammendrag
In a new method of ultrasound imaging, called SURF imaging, dual band pulse complexes composed of overlapping low frequency (LF) and high frequency (HF) pulses are transmitted, where the frequency ratio LF:HF ∼ 1 : 20, and the relative bandwidth of both pulses are ∼ 50 − 70%. The LF pulse length is hence ∼ 20 times the HF pulse length. The LF pulse is used to nonlinearly manipulate the material elasticity observed by the co-propagating HF pulse. This produces nonlinear interaction effects that give more information on the propagation of the pulse complex. Due to the large difference in frequency and pulse length between the LF and the HF pulses, we have developed a dual level simulation where the LF pulse propagation is first simulated independent of the HF pulse, using a temporal sampling frequency matched to the LF pulse. A separate equation for the HF pulse is developed, where the the presimulated LF pulse modifies the propagation velocity. The equations are adapted to parallel processing in a GPU, where nonlinear simulations of a typical HF beam of 10 MHz down to 40 mm is done in ∼ 2 secs in a standard GPU. This simulation is hence very useful for studying the manipulation effect of the LF pulse on the HF pulse.
-
Kvam, Johannes; Angelsen, Bjørn Atle J.; Brende, Ole Martin & Elster, Anne C. (2013). GPGPU Accelerated Solution of NonlinearWave Propagation and Heat Diffusion.
-
Elster, Anne C.; Jensen, Rune Erlend; Nikolaisen, Ivar Ursin; Falch, Thomas Løfsgaard; Pakdel, Samira; Bozorgi, Mohammadmehdi; Smistad, Erik; Kvam, Johannes; Brende, Ole Martin; Pedersen, Stian Aaraas; Melhus, Lars Kirkholt; Nordhus, Lars Espen; Skomedal, Andreas; Nordahl, Andreas & Knutsen, Henrik (2012). Stand: PCer med superdatakrefter.
Vis sammendrag
3D-simuleringer som tidligere trengte superdatamaskiner, kan nå omprogrammeres til å hente ut regnekraft i moderne skjermkort. Ved hjelp av superdatakreftene til moderne grafikkort på spreke PCer viser vi 3D-simuleringer av snø på storskjerm: Det føles nesten som om det snør på deg! (v/ Anne Cathrine Elster)
Se alle arbeider i Cristin
Publisert 10. aug. 2016 14:45
- Sist endret 10. aug. 2016 14:45