Fitting methods for intravoxel incoherent motion imaging of prostate cancer on region of interest level: Repeatability and gleason score prediction

Merisaari, Harri; Movahedi, Parisa; Perez, Ileana Montoya; Toivonen, Jussi; Pesola, M.; Taimen, Pekka; Boström, Peter J.; Pahikkala, Tapio; Kiviniemi, Aida; Aronen, Hannu J.; Jambor, Ivan

doi:10.1002/mrm.26169

Cited by 49 publications

(77 citation statements)

References 43 publications

(120 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This could in part be due to the somewhat different ranges and combinations of D and f in the simulations and in vivo data, where from a parameter-estimation perspective, least favorable combinations of D and f potentially were missing or less frequently represented in the in vivo data. Still, the overall trend from both simulations and in vivo evaluation is that only minor differences in bias and variability can be seen between estimation approaches, which is in concordance with previous findings [16]. The same trend is seen for the ability to differentiate between tumor and healthy liver tissue, which was similar for all estimation approaches, including the substantially biased Bayesian posterior mean of f.…”

Section: Discussionsupporting

confidence: 87%

“…[15,16,26]) and more recently for estimation limited to D and f [27,28]. However, comparison of estimation approaches for D and f only is, to our knowledge, limited to a single study in which the segmented approach was compared with least-squares fitting of the sIVIM model [16].…”

Section: Discussionmentioning

confidence: 99%

“…1), both assuming b values in certain ranges and that D* ≫ D. In the former approach, often referred to as segmented fitting, estimation is done in two steps [15][16][17]. In the first step, data from b values below a certain threshold (b thr ) are omitted.…”

Section: Parameter Estimationmentioning

confidence: 99%

“…Multiple studies concerning approaches for fitting the IVIM model have been conducted, e.g., [16,17,20]. However, evaluation of approaches for estimating D and f only is, to our knowledge, limited to a subanalysis in a single study [16].…”

mentioning

confidence: 99%

“…However, evaluation of approaches for estimating D and f only is, to our knowledge, limited to a subanalysis in a single study [16]. Furthermore, while Bayesian estimation techniques have shown great promise for the full IVIM model, evaluation of the potential improvement in estimating D and f only is lacking [21][22][23].…”

mentioning

confidence: 99%

See 4 more Smart Citations

Comparison of methods for estimation of the intravoxel incoherent motion (IVIM) diffusion coefficient (D) and perfusion fraction (f)

Jalnefjord

Andersson

Montelius

et al. 2018

Magn Reson Mater Phy

View full text Add to dashboard Cite

Objective Intravoxel incoherent motion (IVIM) shows great potential in many applications, e.g., tumor tissue characterization. To reduce image-quality demands, various IVIM analysis approaches restricted to the diffusion coefficient (D) and the perfusion fraction (f) are increasingly being employed. In this work, the impact of estimation approach for D and f is studied. Materials and methods Four approaches for estimating D and f were studied: segmented IVIM fitting, least-squares fitting of a simplified IVIM model (sIVIM), and Bayesian fitting of the sIVIM model using marginal posterior modes or posterior means. The estimation approaches were evaluated in terms of bias and variability as well as ability for differentiation between tumor and healthy liver tissue using simulated and in vivo data. Results All estimation approaches had similar variability and ability for differentiation and negligible bias, except for the Bayesian posterior mean of f, which was substantially biased. Combined use of D and f improved tumor-to-liver tissue differentiation compared with using D or f separately. Discussion The similar performance between estimation approaches renders the segmented one preferable due to lower numerical complexity and shorter computational time. Superior tissue differentiation when combining D and f suggests complementary biologically relevant information.

show abstract

Section: Discussionsupporting

confidence: 87%