Showing papers in "Medical & Biological Engineering & Computing in 1994"
TL;DR: In this procedure, essentially nothing is assumed about the source currents, except that they are spatially restricted to a certain region, and the results can describe the structure of the current flow fairly well.
Abstract: The authors have applied estimation theory to the problem of determining primary current distributions from measured neuromagnetic fields. In this procedure, essentially nothing is assumed about the source currents, except that they are spatially restricted to a certain region. Simulation experiments show that the results can describe the structure of the current flow fairly well. By increasing the number of measurements, the estimate can be made more localised. The current distributions may be also used as an interpolation and an extrapolation for the measured field patterns.
1,796 citations
TL;DR: A model which assesses the closed-loop interaction between heart period and arterial pressure variabilities and the influence of respiration on both is applied to evaluate the sources of low frequency and high frequency sources in conscious dogs and humans, finding a strong rhythmic modulation of HP during CAO.
Abstract: A model which assesses the closed-loop interaction between heart period (HP) and arterial pressure (AP) variabilities and the influence of respiration on both is applied to evaluate the sources of low frequency (LF approximately 0.1 Hz) and high frequency (HF, respiratory rate approximately 0.25 Hz) in conscious dogs (n = 18) and humans (n = 5). A resonance of AP closed-loop regulation is found to amplify LF oscillations. In dogs, the resonance gain increases slightly during baroreceptor unloading (mild hypotension obtained with nitroglycerine (NTG) i.v. infusion, n = 8) and coronary artery occlusion ((CAO), n = 6), and it is abolished by ganglionic transmission blockade ((ARF), Arfonad i.v. infusion, n = 3). In humans, this gain is considerably increased by passive tilt. Different, possibly central, sources of LF oscillations are also evaluated, finding a strong rhythmic modulation of HP during CAO. At HF, a direct respiratory arrhythmia is dominant in dogs at control, while it is considerably reduced during CAO. On the contrary, in humans, a strong influence of respiration on AP is shown which induces a reflex respiratory arrhythmia. An index of the gain of baroreceptive response, alpha cl, was decreased by NTG and CAO, and virtually abolished by chronic arterial baroreceptive denervation (TABD, n = 4) and ARF.
192 citations
TL;DR: Computer simulation of pulsatile non-Newtonian blood flow has been carried out in different human carotid artery bifurcation models and demonstrates that the wall distensibility alters the flow feld and the wall shear stress during the systolic phase.
Abstract: Computer simulation of pulsatile non-Newtonian blood flow has been carried out in different human carotid artery bifurcation models In the first part of the investigation, two rigid walled models are analysed, differing in the bifurcation angle (wide angle and acute angle bifurcation) and in the shape of both the sinus (narrow and larger sinus width) and the bifurcation region (small and larger rounding of the flow divider), in order to contribute to the study of the geometric factor in atherosclerosis The results show a significant difference in the wall shear stress and in the flow separation Flow recirculation in the sinus is much more pronounced in the acute angle carotid An important factor in flow separation is the sinus width In the second part of the study, flow velocity and wall shear stress distribution have been analysed in a compliant carotid artery bifurcation model In the mathematical model, the non-Newtonian flow field and the idealized elastic wall displacement are coupled and calculated iteratively at each time step Maximum displacement of approximately 6% of the diastolic vessel diameter occurs at the side wall of the bifurcation region The investigation demonstrates that the wall distensibility alters the flow field and the wall shear stress during the systolic phase Comparison with corresponding rigid wall results shows that flow separation and wall shear stress are reduced in the distensible wall model
118 citations
TL;DR: A new method is demonstrated to visualise the microcirculation map of the human retina using a dynamic laser speckle effect and the results were displayed in colour graphics showing the spatial distribution of the blood flow in the retina.
Abstract: A new method is demonstrated to visualise the microcirculation map of the human retina using a dynamic laser speckle effect. The retina is illuminated with a diode laser spot through a retinal camera, and its speckle image is detected by an area sensor. The output signal from the sensor is digitised, and the data for more than a hundred scannings of the speckle image are stored in a mass image memory. The difference between a pair of successive image data was calculated and integrated for each pixel. The results were displayed in colour graphics showing the spatial distribution of the blood flow in the retina.
99 citations
TL;DR: A design for an amplifier for bioelectric events is presented that has fewer parts, with reduced dimensions and with a lower power consumption, that allows the construction of amplifiers of a high quality in terms of noise and common mode rejection.
Abstract: A design for an amplifier for bioelectric events is presented that has fewer parts than conventional designs. The design allows the construction of amplifiers of a high quality in terms of noise and common mode rejection, with reduced dimensions and with a lower power consumption. Gain, bandwidth and number of channels are easily adapted to a wide range of biomedical applications. An application example is given in the form of a multichannel EEG amplifier (gain is 20000), in which each channel consists of three operational amplifiers (one single and one dual), six resistors and two capacitors. The equivalent input noise voltage and current are 0.15 μVrms and 1 pArms, respectively, in a bandwidth of 0.2–40 Hz, and a common mode rejection ratio of 136 dB is achieved without trimming.
95 citations
TL;DR: Although many VF detection techniques have been developed and claim good performance, independent evaluation has shown that some techniques are not optimal and one approach with potential lies in the area of neural computing.
Abstract: VENTRICULAR FIBRILLATION (VF) is a lethal cardiac arrhythmia which can be stopped by prompt application of a DC shock (defibrillation). On the electrocardiogram (ECG), VF appears as an irregular, undulating waveform. (Fig. 1) ECG monitoring systems and automatic defibrillators attempt to reduce delays in recognising VF by incorporating automatic detection algorithms. However, electrode artifact and other cardiac arrhythmias can produce VF-like ECG signals which lead to false positive detections. In addition, the ECG waveform during VF is poorly characterised and this can lead to false negative detections. Development and testing of VF detection algorithms has relied on a limited database of VF recordings because VF is unpredictable, and hence rarely recorded in a form suitable for analysis. Although many VF detection techniques have been developed and claim good performance (JAKOBSSEN e t al., 1990), independent evaluation has shown that some techniques are not optimal (CLAYTON et al., 1993). There is clearly room for further development. One approach with potential lies in the area of neural computing. Neural computing is a rapidly growing field (MILLER et al., 1992; BEALE and JACKSON, 1990). Simple processing units (neurons) are linked together by weighted connections. Each neuron processes its weighted inputs according to its activation function, and the output is then passed on to the inputs of the next layer of neurons. The remarkable feature of a neural network lies in its flexibility. By allocating appropriate values to the weights, a network can perform specific and complicated operations on its inputs. A network can hence be trained to perform a particular operation, using a set of training data comprising a series of input patterns for which the correct output is known. Each training pattern is repeatedly presented to the inputs. The network weights, orginally set to random values, are progressively optimised using a training algorithm which attempts to produce the correct output for the training patterns. Training continues until the errors associated with
87 citations
TL;DR: It is argued that although limitations exist, pulse compression techniques can still be used to improve resolution or increase penetrational depth and the real advantage over conventional short-pulse imaging comes from the possibility that these improvements can be accomplished without increasing the peak intensity of the interrogated pulse above any threshold levels set by possible bio-effect considerations.
Abstract: Pulse compression techniques that are capable of producing a large signal-to-noise (SNR) enhancement, have been used successfully in many different fields. For medical applications, frequency-dependent attenuation in soft tissue can limit the usefulness of this method. In the paper, this issue is examined through model-simulation studies. Frequency-modulation (FM) chirp, considered in the study, is just one form of pulse coding technique. Pulse propagation effects in soft tissue are modelled as a linear zero phase filter. A method to perform simulations and estimate the effective time-bandwidth product K is outlined. K describes the SNR enhancement attainable under limitations imposed by the soft-tissue medium. An effective time-bandwidth product is evaluated as a function of soft-tissue linear attenuation coefficient αo, scatterer depth z and the bandwidth of the interrogating FM pulse, under realistic conditions. Results indicate that, under certain conditions, K can be significantly lower than its expected value in a non-attenuating medium. It is argued that although limitations exist, pulse compression techniques can still be used to improve resolution or increase penetrational depth. The real advantage over conventional short-pulse imaging comes from the possibility that these improvements can be accomplished without increasing the peak intensity of the interrogating pulse above any threshold levels set by possible bio-effect considerations.
78 citations
TL;DR: The results show that the use of an artificial neural network can improve the sensitivity of reporting AF from 88.5% using the deterministic approach to 92%, without sacrificing specificity (92.3%).
Abstract: Different forms of artificial intelligence have been applied to pattern recognition in medicine. Recently, however, a relatively new technique involving software-based neural networks has become more readily available. Deterministic logic is currently applied to rhythm analysis in computer-assisted ECG interpretation method developed in the University of Glasgow. The aim of the present study is to compare an artificial neural network with deterministic logic for separating sinus rhythm (SR) with supraventricular extrasystoles (SVEs) and/or ventricular extrasystoles (VEs) from atrial fibrillation (AF) at a particular point in the diagnostic logic of the Glasgow Program. A total of 2363 ECGs with 1495 AF and 868 SR+(SVEs and/or VEs) are used for training and testing a variety of neural networks, and the optimum design is selected. Methods for combining the results of the neural-network classification and the deterministic interpretation are also developed. A further 717 ECGs are used to test the selected network. The results show that the use of an artificial neural network can improve the sensitivity of reporting AF from 88.5% using the deterministic approach to 92%, without sacrificing specificity (92.3%).
69 citations
TL;DR: It is concluded that single-node models (although widely used) cannot replace this multiple nonlinear node model, as the stimulus repetition rates that can be followed by the simulated nerve fibre are limited by impulse conduction properties, rather than by the frequency following behaviour of a single node.
Abstract: Starting with the spatially extended non-linear node model (Reilly et al., 1985), which incorporates Frankenhaeuser-Huxley non-linearities at each of several nodes in a row, a model is developed to describe many aspects of the behaviour of mammalian nerve fibres in a quantitative way. By taking into account the effects of temperature and by introducing a realistic nerve morphology, a good fit is obtained between the shape, duration and conduction velocity of simulated and in vivo action potentials in mammals. The resulting model correctly predicts the influence of physiological variations of body temperature on various aspects of nerve behaviour. It is shown that the absolute refractory period predicted by the model is within physiological ranges. Both in vivo and in the model, the spike amplitude and the spike conduction velocity are reduced in the relative refractory period. It is concluded that single-node models (although widely used) cannot replace this multiple nonlinear node model, as the stimulus repetition rates that can be followed by the simulated nerve fibre are limited by impulse conduction properties, rather than by the frequency following behaviour of a single node.
56 citations
TL;DR: Preliminary results are presented, showing a significant beneficial effect of the biofeedback method in increasing the step length of paretic limbs and in correcting step-length asymmetry.
Abstract: A biofeedback gait training system for step length is proposed, adapted to the correction of spatial walking asymmetries by means of a simple, quick and reliable method for daily clinical use. The system is composed of a walkway and a gait analysis device (locometer) measuring the main temporal and distance factors of gait. The step length is imposed on the subject by lighted targets appearing on the walkway, alternately on the right and left side; the subject is asked to place a swinging foot on the lighted target. Feedback to the subject is supplied by direct visual information (the subject looking at the movement and the position of the foot with respect to the lighted target) and an acoustic signal delivered in real time when the length step error is greater than an allowed value. The method is validated on a population of hemiparetic patients who have suffered from a stroke and who have been reeducated with traditional rehabilitation methods. The patients were divided into two groups; one group following a gait training with biofeedback (BFB group)_ and one group following a gait training without biofeedback (reference group). Preliminary results are presented, showing a significant beneficial effect of the biofeedback method in increasing the step length of paretic limbs and in correcting step-length asymmetry.
52 citations
TL;DR: The authors present a method to estimate primary-current distributions in a homogeneous, realistically shaped boundary-element torso model that can be used both for heart and brain studies.
Abstract: The paper deals with the bioelectric and biomagnetic inverse problems. The authors present a method to estimate primary-current distributions in a homogeneous, realistically shaped boundary-element torso model. The reconstruction surface is triangulated to keep the procedure computationally feasible. The minimum-norm estimate is computed on the basis of separate electric and magnetic signals, as well as from combined data. The method can be used both for heart and brain studies. Simulation results for current-dipole sources in a homogeneous realistic torso are discussed.
TL;DR: The construction and fundamental studies are described for a subkilohertz X-ray generator for producing low-dose rate flash X-rays and high-speed radiography was performed.
Abstract: The construction and fundamental studies are described for a sub-kilohertz X-ray generator for producing low-dose rate flash X-rays. The X-ray tube was a glass-enclosed cold-cathode triode, composed of a tungsten plate target, a rod-shaped graphite cathode, a mesh-type trigger electrode made of tungsten wires, and a glass tube body. The coaxial condenser was charged up to 60 kV by a power supply, and the electric charges in the condenser were discharged to the X-ray tube repetitively when a negative high-voltage pulse was applied to the trigger electrode. The maximum tube voltage before the discharging was equivalent to the initial charged voltage of the condenser, and the maximum tube current was about 0.3 kA with a charged voltage of 60 kV. The X-ray durations were about 1 microsecond, and the X-ray intensity was about 0.47 microC kg-1 at 0.5 m per pulse with a charged voltage of 60 kV. The maximum repetition rate of the X-rays was about 0.4 kHz, and high-speed radiography was performed.
TL;DR: The paper describes the finite-element analysis software package developed, including an interactive graphical mesh generator and fast algorithms for solving linear systems of equations using sparse-matrix and vector techniques.
Abstract: In electrical impedance tomography (EIT), current patterns are injected into a subject and boundary voltages are measured to reconstruct a cross-sectional image of resistivity distribution. Static EIT image reconstruction requires a computer model of a subject, an efficient data-collection method and robust and fast reconstruction algorithms. The finite-element method is used as the computer model. The paper describes the finite-element analysis software package developed, including an interactive graphical mesh generator and fast algorithms for solving linear systems of equations using sparse-matrix and vector techniques. Various models of irregularly shaped subjects are developed using mesh-design tools, including automatic mesh generation and optimisation using the Delaunay algorithm. Even though the software package is customised for use in electrical impedance tomography, it can be used for other biomedical research areas, such as impedance cardiography, cardiac defibrillation and impedance pneumography.
Journal Article•
TL;DR: The FEM stress analysis proved that the stress distributions in the second cervical vertebra (axis) are influenced by the direction of the applied load on the anterior surface of the odontoid process.
Abstract: A human vertebra was modelled using the finite element method (FEM), to investigate likely sites of failure under prescribed loading. The FEM stress analysis proved that the stress distributions in the second cervical vertebra (axis) are influenced by the direction of the applied load on the anterior surface of the odontoid process. The regions of high compressive stress could be found on the inner lateral edges of the superior facets, and the maximum tensile stress at the anterior junction of the dens and the vertebral body; both correspond to recorded fracture locations in clinical observations and judicial hangings.
TL;DR: Water-mediated bioeffects further present additional proof that water has the capability to remember the imposed electromagnetic field characteristics for an extended period of time.
Abstract: Athermal physiological effects of continuous wave and modulated microwaves were studied on a cynobacterium Nostoc muscorum. The study shows that different microwave frequencies in continuous wave and modulated modes produced significantly different physiological effects on the algae. Water-mediated bioeffects further present additional proof that water has the capability to remember the imposed electromagnetic field characteristics for an extended period of time.
TL;DR: It is suggested that the pathway of the majority of sound transmission from the trachea to the chest wall changes from a more radial to airway-borne route over the measured frequency range.
Abstract: Sonic noise transmission from the mouth to six sites on the posterior chest wall is measured in 11 healthy adult male subjects at resting lung volume. The measurement sites are over the upper, middle and lower lung fields and are symmetric about the spine. The ratios of transmitted sound power to analogous sites over the right (R) and left (L) lung fields are estimated over three frequency bands: 100–600 Hz (low), 600–1100 Hz (mid) and 1100–1600 Hz (high). A R-L dominance in transmission is measured at low frequencies, with a statistically significant difference observed at the upper site. No significant asymmetry is observed in any measurement site at mid or high frequencies. A theoretical model of sound transmission that includes the asymmetrical anatomy of the mediastinal structures is in agreement with the observed asymmetry at low frequencies. These findings suggest that the pathway of the majority of sound transmission from the trachea to the chest wall changes from a more radial to airway-borne route over the measured frequency range.
TL;DR: Classification of non-averaged task-related EEG responses with different types of classifier, including self-organising feature map and learning vector quantiser, K-mean, back-propagation and a combination of the last two, is reported.
Abstract: Classification of non-averaged task-related EEG responses with different types of classifier, including self-organising feature map and learning vector quantiser, K-mean, back-propagation and a combination of the last two, is reported. EEG data are collected from approximately one second periods prior to movement of the right or left index finger. A cue stimulus indicating which hand to use is employed. Feature vectors are formed by concatenating spatial information from different EEG electrodes and temporal information from different time incidents during the planning of hand movement. Power values of the most reactive frequencies within the extended alpha-band (5–16 Hz) are used as features. The features are derived from an autoregressive model fitted to the EEG signals. The performance of the classifiers and their ability to learn and generalise is tested with 200 arbitrarily selected event-related EEG data from a normal subject. Classification accuracies as high as 85–90% are achieved with the methods described here. A comparison of the classifiers is made.
TL;DR: The electrode/electrolyte interface impedance can be represented by the parallel combination of a non-faradaic pseudocapacitance and a faradaic, charge transfer resistance, which becomes non- linear at large applied voltage amplitudes and dominates the high-frequency non-linear behaviour of the overall interface impedance.
Abstract: The electrode/electrolyte interface impedance can be represented by the parallel combination of a non-faradaic pseudocapacitance and a faradaic, charge transfer resistance. The non-linearity of the overall electrode/electrolyte interface impedance is largely due to that of the faradaic resistance which is derived from the Butler-Volmer equation. As the charge transfer resistance dominates the interface impedance at low frequencies, it is in this region that non-linearities are first observed. The voltage limit of linearity has been investigated and found to increase gradually for higher frequencies. Although relatively linear compared with the charge transfer resistance, the non-faradaic impedance becomes non-linear at large applied voltage amplitudes and dominates the high-frequency non-linear behaviour of the overall interface impedance. Mid-frequencies are affected by a combination of the faradaic and non-faradaic non-linearities.
TL;DR: The study confirms the potential of artificial networks for the classification of bioprosthetic valve closure sounds and describes the design, training and testing of a three-layer feedforward back-propagation neural network.
Abstract: The paper describes the design, training and testing of a three-layer feedforward back-propagation neural network for the classification of bioprosthetic valve closure sounds. Forty-seven patients with a porcine bioprosthetic valve inserted in the aortic position were involved in the study. Twenty-four of them had a normal bioprosthetic valve, and the other 23 had a degenerated valve. Five features extracted from the Fourier spectra and 12 linear predictive coding (LPC) coefficients of the sounds were used separately as the input of two neural-network classifiers. The performance of the classifiers was tested using the leave-one-out method. Results show that correct classifications were 85 per cent using the spectral features, and 89 per cent using the LPC coefficients. The study confirms the potential of artificial networks for the classification of bioprosthetic valve closure sounds. Clinical use of this method, however, still requires further investigation.
TL;DR: Comparison with experimental data shows that the normal aterial wall can be appropriately modelled using a hyperelastic material definition, and results show good agreement with previous histologic studies.
Abstract: Finite-element modelling is used to simulate the response of atherosclerotic arteries to a balloon angioplasty procedure. Material properties for the normal wall are derived from experimental data, and the properties of the plaque are varied over a wide range. Comparison with experimental data shows that the normal arterial wall can be appropriately modelled using a hyperelastic material definition. Large strain, non-linear analysis was used to simulate the dilatation of three typical plaque configurations by an angioplasty balloon. Stress contour plots are presented for each configuration. Results show good agreement with previous histologic studies.
TL;DR: The procedure, which conducts clustering by UOFC, is the adaptive and machine-learning nature of its operation, which correctly differentiates between waveforms evoked by the same stimuli which differ in their context to the performance of a memory-scanning task.
Abstract: A procedure for clustering evoked potentials (EPs) according to their waveforms is presented. Clustering is performed without a priori selection of basis waveforms, the number of basis waveforms or the number of clusters. The method uses the principal-component-analysis coefficients of EP records as features for unsupervised optimal fuzzy clustering (UOFC) of the records. The validity of the procedure is demonstrated in two instances: visual evoked potentials (VEPs) and cognitive event-related potentials (ERPs) from humans in a memory-scanning task. In the clustering of VEPs, the procedure differentiates between waveforms judged to be clinically normal and abnormal. In the clustering of ERPs, the procedure correctly differentiates between waveforms evoked by the same stimuli which differ in their context to the performance of a memory-scanning task (memorised items against probes). Within this classification, the procedure detects two subgroups to probeevoked waveforms, which are not obvious from visual inspection of the waveforms. The advantage of the procedure, which conducts clustering by UOFC, is the adaptive and machine-learning nature of its operation.
TL;DR: Time-frequency transforms are employed to characterise first heart sound frequency dynamics from 27 sites across the human thorax implying that the instantaneous power and the power spectrum contain most of the diagnostic information in the first heartSound frequency law.
Abstract: Power spectral analysis has attracted attention because of its potential for non-invasive cardiac diagnosis. However, time-frequency analysis of first heart sound frequency dynamics from canine epicardium has demonstrated that cardiac vibrations are fundamentally multi-component and non-stationary, questioning the validity of power spectral techniques. In this study, we employed time-frequency transforms to characterise first heart sound frequency dynamics from 27 sites across the human thorax. In contrast to the dynamics observed epicardially, the first heart sound frequency law was dominated by quasi-stationary and impulse-like components implying that the instantaneous power and the power spectrum contain most of the diagnostic information in the first heart sound.
TL;DR: A mathematical model of cardiovascular system mechanics, on which both quantitative diagnosis and evaluation of drug effects can be made, which provides a useful basis for analysing pathophysiological states in heart failure and evaluating drug effects on the disease.
Abstract: The planning of drug therapy for heart failure should involve both the diagnostic analysis of the patient's defective state and a prediction of the drug effects on the identified state. We have devised a mathematical model of cardiovascular system mechanics, on which both quantitative diagnosis and evaluation of drug effects can be made. The model was composed of systemic and pulmonary circulatory networks including the dynamics of the left and right ventricles. The model of the ventricles can represent both systolic and diastolic problems in heart failure through the parameters of ventricular contractility and diastolic stiffness. Each vascular network was composed of arterial and venous resistances and total vascular capacitance. Patient's ventricular and vascular parameters were estimated simultaneously from the clinically measurable haemodynamic variables based on the model. Despite the simplicity of the model, the results showed good agreement with clinical and experimental data. The clinically significant haemodynamic classification of heart failure by Forrester et al. (Forrester et al., 1977) was simulated well by the model. This model provides a useful basis for analysing pathophysiological states in heart failure and evaluating drug effects on the disease.
Journal Article•
TL;DR: The present perivascular PDU system is able to register velocity profiles covering the entire vessel area in a plane perpendicular to the flow axis, as well as turbulent velocity fluctuations within the restrictions imposed by the Doppler ambiguity process.
Abstract: A pulsed Doppler ultrasound (PDU) multi-element system was developed for perivascular registration of velocity fields and turbulence in large vessels. In vivo evaluation and comparison with hot-film anemometry (HFA) was performed. C-shaped shells were designed with holes to fit five small 10 MHz ultrasonic probes directed at five measuring points along a diameter perpendicular to the vessel axis. By rotating the shell in 45 degrees steps, blood velocities were measured in 17 points covering the entire cross-sectional vessel area. Measurements were performed in the ascending aorta and at three axial locations in the descending thoracic aorta in pigs. Simultaneous PDU and HFA measurements were performed distal to induced vascular stenoses of different degrees. Three-dimensional visualisation of velocity profiles was made, and Reynolds normal stresses (RNS) were calculated for different levels of turbulence intensities based on the simultaneous PDU and HFA measurements. The velocity profiles in the ascending aorta were skewed at top systole with the highest velocities towards the posterior wall. In the descending thoracic aorta at the ligmentum of Botalli, the velocity profiles were skewed throughout the entire systole with the highest velocities at the right anterior vessel wall. Further downstream in the descending aorta the velocity profiles appeared blunter. The frequency response of the modified PDU system was determined by a 'random noise test' revealing an upper -3dB cut-off frequency of approximately 200 Hz. Regression analysis showed a linear relationship between RNS measured with PDU and RNS measured with HFA (r = 0.93). Two vessel diameters distal to a 75% stenosis RNS up to 28 N m-2 were measured.(ABSTRACT TRUNCATED AT 250 WORDS)
TL;DR: Filters of copper are generally recommended, as are filters of aluminium, while the use of bandpass filters (K-edge filters) should be restricted to examinations where the need for substantial variation in tube potential from patient to patient is small.
Abstract: The influence on image contrast, tube load and patient mean absorbed dose of different ways of shaping diagnostic X-ray spectra by placing filters in the beam is derived for two radiographic models (abdominal screen-film radiography and intra-oral, dental radiography) using a computational model. The filters are compared at either equal tube load (keeping tube potential constant) or equal contrast (adjusting the tube potential with the different filters), but always at equal energy imparted per unit area to the image receptor. Compared at equal tube load and relative to standard aluminium filtration, reductions in the mean absorbed dose in the patient of 15–25% can be achieved using filters of Cu, Ti, W and Au (increasing the tube load by 30–40% compared with standard aluminium filtration). However, contrast is also reduced by 7%. Compared at equal contrast, the dose reductions are smaller, about 10%. Filters of copper are generally recommended, as are filters of aluminium. The use of bandpass filters (K-edge filters) should be restricted to examinations where the need for substantial variation in tube potential from patient to patient is small. The benefit of using thicker filters than those commonly used today (increasing tube load by factors of 1.4–2.0 compared with no added filter) is small as the dose reduction is most rapid for small initial values of added filters, and the increase in tube load increases steadily with increasing filter thickness.