Among the speakers at Wired Health, the event organized last March 15 in Milan by Wired with the scientific and editorial collaboration of Humanitas and dedicated to health technologies, was Professor Arturo Chiti, Head of Nuclear Medicine at Humanitas and Professor at Humanitas University with a speech entitled “The image of artificial intelligence.
Image diagnostics and shape recognition
“Diagnostic imaging uses various technologies (such as X-ray, ultrasound, CT, MRI, PET, scintigraphy) with the aim of seeing within our body. However, by image diagnostics we also mean the evaluation of tissues, and so we are also trying to integrate the images observed by fellow pathologists.
The concept on which we are working and which we would like to carry forward into the future is to use the acronym AI not only in terms of artificial intelligence, but also in terms of Augmented Imager, in order to be able to offer an aid to what we do every day with diagnostic imaging or the recognition of forms.
It has been shown that the recognition of forms in our brains is a mechanism very similar, if not equal, to what we call things. When a radiologist looks at an image, he recognizes something unusual within a structure and gives it a name: it can be for example a pneumonia, an aortic aneurysm, a fracture, a tumor. This process requires a great deal of experience, which must be passed on to students and junior doctors.
We then need to try to go beyond naming things and integrate the information we have with others information (the patient’s record, previous clinical data, data obtained through the visit, information provided by colleagues). The diagnosis, therefore, will be the result not only of what we see, but also of all the other information collected,” explained Professor Chiti.
Help with algorithms
“This task can also be carried out by algorithms, which need to be trained so that they can facilitate clinicians and radiologists in the diagnosis and planning of treatments. However, training has its critical points: technology alone is not enough, the ability to introduce human knowledge into it is necessary; otherwise, this technique will not be successful. It is therefore essential that, at this stage of exponential technological development, we are able to offer quality content, producing and certifying information that allows algorithms to learn correctly.
Another fundamental aspect is the ethical one: we have a huge amount of data available, but these are patients and therefore we must pay attention to preserving the quality of the data and the ethical aspects linked to the privacy of patients.
Once trained, these algorithms can, for example, give a name to the lesions and see how they evolve over time thanks to some features that allow you to recognize the lesion always as itself.
They can also integrate different images (CT or pathological anatomy), clinical data, the history of the patient and his life habits and in fact can help us to work faster (and perhaps over time we will find out in a different way) toward an analysis of all the information we can have from the patient, to the patient. These algorithms reason by similarity and are able to identify similar elements and therefore a patient in a group of similar patients will have a more correct diagnosis, resulting in better therapy, better therapeutic approach, less hospitalization, fewer side effects and lower costs for society,” said the professor.
The tool of Humanitas and Orobix for lung cancer
“At the moment we are focusing on lung cancer and have made a classification based on CT and PET images. We then developed a tool called Radiomix (born from the collaboration between Humanitas and Orobix) that is able to recognize primitive lesions of the lung to classify the T parameter in this disease. Once the experimental phase has been completed, the instrument will therefore help radiologists and clinicians with the initial evaluation of patients suffering from lung cancer.
That is what we are doing and will continue to do in the coming years. Artificial intelligence algorithms are already a reality and we are very happy to have been able to realize this experience.
In conclusion, it should be stressed that we do not know exactly what is the mechanism by which an algorithm – even if well trained on certified data – makes a diagnosis and this is another of the ethical aspects on which we will have to work because the radiologist will have to sign a diagnosis that in fact was made by a machine. Moreover, this could be difficult to explain to a patient: because although the machine has within it the experience of many doctors, in the patient’s eyes it remains a machine. However, the future is certainly bright and I am sure that we will be able to provide an increasingly better service to our patients”, concluded Professor Chiti.
Watch Professor Chiti’s full speech, click here