Model-based cardiac diagnosis of pulmonary embolism

Abstract

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 8 7 ( 2 0 0 7 ) [46][47][48][49][50][51][52][53][54][55][56][57][58][59][60] j o u r n a l h o m e p a g e : w w w . i n t l . e l s e v i e r h e a l t h . c o m / j o u r n a l s / c m p b IntroductionCardiac disease state is highly patient specific and difficult to accurately diagnose due to the limited measurements available. In addition, the body's natural reflex responses try to restore circulatory equilibrium, which can often mask the underlying symptoms [1,2]. Successful diagnosis and treatment often rely on the experience and intuition of clinical staff. Thus, a physiological, identifiable and validated computer model offers several potential advantages in diagnosis and therapy selection, by aggregating diverse patient data into a compact, patient specific, clinically relevant and potentially real-time assessment of circulatory status. * Corresponding author. Tel.: +64 3 366 7001; fax: +64 3 364 2078.E-mail address: chris.hann@canterbury.ac.nz (C.E. Hann).There are many CVS models in the literature ranging from very complex finite element models [3-6] to relatively simpler pressure volume approaches [7][8][9]. However, the focus is often on only specific areas of CVS dysfunction. Although there are full CVS models, patient-specific parameter optimization is either not considered or restricted to small subsets of the overall much larger parameter set (e.g. [10,11]). This restriction to specific CVS aspects can dramatically limit the range of CVS disturbances that can be detected, thus prohibiting use as a broader diagnostic tool for patients with unknown condition. For relatively larger, more complex system models computational cost and feasibility can also become a major issue.

Unique parameter identification for cardiac diagnosis in critical care using minimal data sets

“…(1) and (2) do not significantly effect parameter identification [26]. The parameter Po.iu/ has also been shown to have a limited effect [26] and for discrete data is typically identified to be close to 0 [21]. Therefore L mt , L av and P 0,lυf are set to 0.…”

“…The parameter identification method presented is an extension of the concept developed in [21,26]. The idea in [21,26] is to set up an iteration between a linear least squares optimization and a forward solution, which is partly based on a Picard iteration [34].…”

“…The idea in [21,26] is to set up an iteration between a linear least squares optimization and a forward solution, which is partly based on a Picard iteration [34]. This approach is therefore distinctively different from other integral formulations like the modulating function approach [35] which does not iterate and is therefore not directly suitable for the discrete data and high non-linearities present in this application.…”

“…Therefore, the standard method of non-linear regression [36], is not suitable, as it is too computationally intense and can often result in local minima. In this section the method of [21] is significantly improved by avoiding the requirement of a continuous volume profile, which is typically not known. In addition the number of forward simulations and the computational requirements for each iteration are dramatically reduced.…”

“…This approach has been well developed and validated in the literature [12][13][14][15][16] so it serves as a good starting point to test the methods and concepts in this paper. The specific baseline model structure considered is a six chamber cardiovascular model including ventricular interaction and inertial effects that has been previously developed [16] and validated in clinical animal trials [21][22][23][24]. However, note that the approach is general and could be applied to any cardiac model structure.…”