Q/A

Q: How could a computer recognise cancer?

Answer: Humans during their childhood learn to identify colors projected on their retina and to classify them (apple/not apple, car/no car, etc.). We also learn to classify objects if they are slightly different from what we have known so far (a SUV is also a car, etc.). In the second half of the last century scientist have tried to implement the histologic architecture of the brain in computers. With some workarounds they succeeded in that and called it artificial intelligence (AI). AI uses neuronal networks, which is an algorithm that is composed of layers. Between each pair of layers there is a complex algorithm that transforms the information as it passes thru. The bottom most layer consists of two boxes labelled "No cancer" and "cancer". To use a neuronal network one has to "train" it before: You take a set of images of which you know if on them is shown cancer or not. You insert them in the topmost layer and wait what will be the result in the two boxes. The first time you do that there will be many wrong classifications. But knowing what should be the right classification you can use another algorithm called "backpropagation" to adjust the mathematical operations of the neuronal network. After that you put again the same pictures into the topmost layer of the neuronal network and look what is the result. Repeating this procedure if you are lucky the classification of the neuronal network will become more and more accurate. If not you have to adjust the architecture (e.g. how many layers you use). There are several parameters that can be set in order to modify the architecture. These parameters are called "hyperparameters".

Q: Do you need the graphic card to show pictures of your findings?

Answer: To just show them not. These are png or jpg images and you don't need a graphics card for that. To mark tumour areas on a new picture it is adviseable to use a GPU (graphics Card). With a good multicore CPU the computer can do that without a graphics card, but it takes longer (several minutes).

Q: What kind of impact will this research have in the public eye?

Answer: Pathologists have a very reserved attitude towards this new approach. For the time beeing there is no need to fear that all pathologists will be abolished within the next few years. The trained eye of a pathologist is - for the moment - far better that the best neuronal networks. But science is developping fast, and therefore perhapes some time the computer will be as good as the pathologist. For the moment the computer can help the pathologist to identify interesting areas but the neuronal network cannot replace the pathologist. AI will not only be used in histopathology: also in radiology neuronal networks can be applied with success.

Q: Once developed, what kind of change will happen in the surgical process?

Answer: The surgical procedure won't change at all. Our work focuses only on the diagnostic part. Patients will hardly notice anything because normally it is a tumour board that decides how to threat the patient after the diagnosis has been established.

Q: What type of equipment will surgeries need in the future if they implement your process?

Answer: We use exactly the same material as has been used in traditional microscopy. There is no need to change any surgical procedures.

Q: Must a patient feat that he/she is treated by a computer instead of a doctor.

Answer: No. Many fear modern technology and I can understand that very well. But computers will help the physicians not replace them. For example nobody wounders today that in a medical laboratory computers do most of the analysis in blood samples. Once this work was done by hand by lab technicians. And secretaries weren't abolished because typewrites were replaced by computers.