4 A typology for CDSS
4.5 An example of a CDSS in the therapeutic cycle: CAESAR
To determine the type of CDSS that is the focus of this research, the CAESAR project is explained in detail. This section builds on the general information about care processes that was provided in section 3.1.
4.5.1 THE GOAL
At this point, the goal of CAESAR is to:
- Aid in selecting and assessing the best way of treatment for an individual patient (personalized medicine) and improve patient care. This is done through generating as much valuable and high-quality data and data analysis as possible from medical imaging that can be used by a clinician to decide upon. Outputs can come in the form of images or text.
- Reduce unnecessary treatment (either by minimizing the treatment or not performing it at all) in order to improve efficiency and cost reductions in healthcare.
Figure 12: A typology for CDSS in the therapeutic cycle.
4.5.2 THE PROBLEM
CAESAR focuses solely on patient with liver metastases (an extensive analysis on this disease can be found in Appendix B). Metastases in general are an indication that a curative treatment is impossible. There is however hope for these specific patients. If the patient only has metastases in the liver (and not in any other organs), then there is a chance to cure these patients. Research has shown that if these metastases can be fully removed from the liver, the patient can be cured [69]. This can be done in two ways: local treatment (like resection) or systemic therapy followed by resection.
There are two main problems here:
1. Evaluating the response of a patient to systemic therapy. As will be seen later this section, the criteria for measuring this response are very basic and do not provide a fully accurate picture of current situation and the response to the chemo.
2. Determining whether a local treatment strategy for a metastasis is possible or not. This is tricky due to the anatomy of the liver. The liver contains 9 segments. To keep a fully functioning liver, only 2 of these segments need to be present. This is of course remarkable, but it is important to remember that these 2 segments cannot be chosen arbitrarily. The two constraints are that there must be a functioning in- and outflow of blood and that there must be a functioning bile duct. Determining whether a local treatment strategy is possible is not only subject to anatomical constraints, but also to constraints represented by the confidence of the surgeon in his/her own skills.
4.5.3 THE SOLUTION
To address the two problems, SAS wants to create a system that can perform several actions. Development is divided into three phases: generate as much valuable information as possible from CT- scans, auto segment CT-scans, determine resectability using a model of the liver and advise on systemic therapy.
Phase 1. The development starts with the generation of as much valuable metastasis information as possible from CT-scans. To illustrate why this is extremely valuable, Figure 14 shows a CT-scan of a patient with liver metastases. The left image shows a situation before systemic therapy, the right one after systemic therapy. Now, clinicians assess these images according to the guidelines on roughly two criteria: are the metastases smaller and are there less metastases? For this example, it is clear: there are less metastases and they are smaller as well. The treatment works and should continue.
Figure 14: CT-scan treatment effect assessment of liver metastases A
In other images, this assessment is harder to perform. Figure 15 shows that the metastases are smaller but there are more metastases. However, with a bit of logic, we can see that the extra metastasis is actually a former large metastasis split in two due its shrinkage.
Figure 15: CT-scan treatment effect assessment of liver metastases B
Figure 16 shows that the metastases are not smaller and there are not less of them, so one might say that the treatment stabilizes the disease. However, what we do see is that the edges of the metastases are far clearer and that the greyscale within the metastases is spread evenly and this does represent some sort of progress (whether this is positive is up to the clinician). However, current guidelines do not account for these features.
These images illustrate that there are a lot of unused characteristics that cannot be used at this moment. ANN allows us to see and measure the characteristics. Adding the greyscale, volume, density, maximum diameter etc. to the list of available data can lead to a better
analysis of the response to systemic therapy.
Phase 2. Once this is done, an effort is made to auto-segment the metastases, organs and other tissues from a CT-scan. The purpose for this is three-fold: 1) it removes the inter-observer variability between radiologists, 2) it gives better insight into
the structure of an individual’s organs and metastases and 3)
it makes the process more efficient and less time-consuming. Early efforts have provided an indication of what this might look like. In Figure 17, the liver (in green) and metastases (in red) can be visualized in the x,y,z-plane and this allows for an easier assessment of 3D-characteristics.
Phase 3. When the system is able to provide all relevant characteristics of the liver and metastases and can model this in a 3D environment on a patient level (given the differences in anatomy), the next step will be to create a model that can determine the technical feasibility of local treatment strategies of the metastases in the liver. An AI could provide insights into this feasibility if it is presented with enough relevant data. Moreover, the AI would be better at providing an analysis of the needs and possibilities of the individual patient rather than a population of patients (like surgeons often do). This is due to the individual unique data that the AI can analyze.