IMPPACT will develop a physiological model of the liver and simulate the RFA intervention's result, accounting for patient specific physiological factors. Mathematical modelling together with experimental validation lead to a patient specific intervention planning system.

Patients can only be examined radiologically and prediction therefore has to rely on macroscopic parameters and tissue properties that can be measured minimally invasive. However, during the heating process microscopic changes at a cellular level affect the end result and should be incorporated the macroscopic equations to allow patient specific prediction. Therefore, both scales are dynamically linked together implying the multi-scale modelling approach to gain iteratively the optimum description. IMPPACT will use several approaches on the macroscopic scale:

On the microscopic scale the key feature of the modelling approach is the characterisation of detailed models at one length scale such that they provide 'lumped' parameters for use at larger length scales: By integrating modelling, imaging and visualization in a full simulator, IMPPACT will create the Intervention Planning System (IPS) as a clinically relevant application. Creating a tool for validation on the highest level by visualizing treatment results together with simulation results, the models will provide input to the clinical practice whereas feedback from the clinical practice will in turn support the modelling. This way confidence in the model predictions will be established.

Expected Results & Impacts

IMPPACT will be modelling a physiological organ including the metabolism and patient specific tissue properties. This alone is a huge step forward as compared to the state-of-the-art intervention planning systems that do not address this issue.

The IPS will allow prediction of treatment results on a patient specific base. It will therefore bring down the risk of local recurrences and eliminate the nowadays so common repeated treatments of the same tumour, making RFA an as effective treatment as resection.

At the same time the IPS will make RFA treatment much safer. By reliably predicting tissue heating it will warn of possible damage to surrounding organs in advance and allows choosing a safe needle position and path.

The greatest impact will be achieved by installing the created application in many hospitals in Europe. To be able to directly use the IPS in clinical practice medical personnel in those hospitals needs to be trained in using it. The augmented reality training simulator provides an excellent opportunity as it trains surgeons directly with the IPS

All developed software will be open source and run with common hospital equipment. Its deployment to virtually every hospital in Europe is solely a question of using a deployment infrastructure.


1st year:

During the first year of the project, the consortium has set up a common framework for an experimental cycle to be conducted on pigs. The framework structure defines a closed-loop experiment with several complementary components. Each component generates a set of data in a compatible format to be consumed by one or several other components. Overview of the structure and data flow between the components in one experimental cycle is illustrated in Figure 1.

Figure 1. The experimental loop conducted on pigs comprises 4 layered components. The components in layers I through III are consecutive: each component generates data to be input to the subsequent component. Components in layers III and IV are engaged into an iterative adjustment of model parameters: Model prediction is compared against CECT built 3D reference model using the visualization component. Model parameters are updated to minimize inconsistencies. Note that micro-CT image modality has been added to the loop to enable registration of histology lesion with the 3D reference model.

Modelling approaches and algorithms were developed for each component of the experimental cycle. These included several modelling activities such as 1) macroscopic mathematical description of bio-heat processes on macroscopic and single-cell levels, 2) numerical modelling of heat transfer, and 3) visualization model. Algorithmic solutions for automatic and semi-automatic image analysis have also been part of this objective.
The modelling activities were based upon experimental data obtained in animal experiments with pigs. As stated by the objectives, the modelling activities and corresponding algorithms had to address the macroscopic and microscopic nature of the problem at hands. The experimental objective for the reporting period was to conduct extensive animal experiments in order to collect sufficient amount of macroscopic and microscopic experimental data. In particular:

To reach the above objectives a detailed list of tasks has been worked out by the project partners. This comprises:

2nd year:

Key objectives for the second reporting period Month 12-Month 24 was to 1) conduct the complete the RFA experimental loop (Figure 1) using pigs; 2) register data resulted from each step of the loop into a unifying 3D reference model of liver structures; and 3) perform cross-validation of the developed RFA modelling within the reference model. While conducting the experimental loop the consortium has further developed necessary methods, algorithms and procedures used in different experimental components leading to a much higher level of their maturity. Novel procedures and algorithms were established to close the experimental loop in a way, which allows optimal cross validation of the physiological / numerical model with established radiological understanding about RFA related processes.

Experimental loop using pigs

The aim of the experimental loop on pigs was to establish understanding of the RFA processes in terms of:

In support of the above, IMPPACT consortium scheduled a set of RFA experiments with healthy pigs sacrificed 1) right after the RFA intervention; as well as 2) one day; 3) one week; 4) one month after the RFA intervention (Figure 2).

Figure 2. RFA related development and associated data acquisition in time.

Each RFA experiment comprises sequential steps for data acquisition, data processing and data cross-validation with a rough outline as follows:
  1. RFA intervention accompanied by required data acquisition (i.e. RITA position data; Rita temperature record, CT images, blood pressure);
  2. Follow up acquisition of CT images and other data as required by the established protocol until the date of sacrificing;
  3. Sacrificing, conservation of the ablation zone, acquisition of its micro-CT images;
  4. Slicing, preparation of histology samples and image acquisition of histological slices;
  5. Segmentation of CT images and building of a 3D intervention model. This model is referred to as reference 3D model of liver & vessel structures. All segmentation and reconstruction results originating from later image data are registered into the reference 3D model.
  6. FEM simulation using the reference vessel tree and RITA positioning and power record during the RFA.
  7. Segmentation of RFA lesion in follow up CT images, registration of the lesion volume into the reference model.
  8. Segmentation of RFA lesion after sacrificing from micro CT and histology. Mutual registration of the two lesion volumes segmented in these different image modes.
  9. Visualization of the reference 3D model including all lesion volumes (reconstructed in steps: 5, 7, 8) and the FEM lesion volume simulated in 6. Validate the RFA modelling.

3rd year:

Key objectives for the third reporting period were: 1) Conducting further experiments on pigs within the RFA experimental loop (Fig. 1) and using the extended protocol (i.e. Micro-CT imaging and artificial landmarks); 2) Process and register data resulted from each step of the loop into a unifying 3D reference model of liver structures; 3) Perform cross-validation of the registration results from Step 2 and benchmark against the prediction by RFA model. 4) Perform cell line experiments to address the critical comment #3 (“Problems with cell biology studies “) from the consolidated report of the second review meeting. 5) Build 3D models of the available patient data 6) Process patient data using the RFA model and refine the model to obtain best prediction results. Benchmark the model prediction against the computed 3D models. 7) Develop clinical user interface fro the IPS and training simulator. 8) Implement final release of the IPS and training simulator. Test these tools in the clinical conditions.

Experimental loop using pigs

Table lists pigs that undergone the complete experimental loop required for the model validation:

Exp. # Time after RFA Pig # Status
1 1 month Pig 26, healthy, 2 lesions 1 lesion fully processed. No Micro-CT. Registration is based on the reconstructed vessel tree.
2 1 day Pig 37, healthy, 2 lesions Same as above. Difficult histology segmentation due to short survival period.
3 1 week Pig 42, cirrhotic, 2 lesions 1 lesion fully processed. Registration based on the MicroCT and the reconstructed vessel tree.
4 1 week Pig 60, cirrhotic, 2 lesions 1 lesion fully processed. Registration based on MicroCT and the artificial landmarks.
5 1 week Pig 64, cirrhotic, 2 lesions Same as above.
Four full data sets (i.e. histology/MicroCT) for pigs 37, 42, 60, and 64 have been processed and the lesion volume reconstructed.
We have applied probabilistic classification followed by the level-set segmentation for robust identification of essential structures (i.e. lesion region, vessels, the artificial landmarks and others) from CAB- and MB-stained histology slides. Three additional registration steps have been included into a modified processing protocol that uses MicroCT image modality: (1) histology – embedded MicroCT registration; (2) embedded MicroCT – native MicroCT; 3) native MicroCT – CECT. The latter step 3 utilises the artificial landmarks. All necessary algorithms were implemented and were integrated within an interactive segmentation tool. Nevertheless the created segmentation tools retain a significant amount of manual adjustments as the large variability in histology appearance makes it necessary to adapt the algorithms to each data set. The final native Micro-CT and CT registration have been reliably performed based on the artificial landmarks.
The reconstructed lesion volumes were registered with the corresponding 3D reference Models. Statistical comparison between the histological, CT and the predicted by the physiological model lesion volumes has been carried and out. This showed accurate correspondence between the overlapping volumes well within the clinically required accuracy of 5mm.

Figure 3. Example histology (green)
and CT-reconstructed (red) lesions.
The results show accurate overlap of
the CT and histology lesions with the
CT lesion almost entirely contained
within the histology lesion, which was
a clinically desired result

Copyright 2006 - Design by LS