Aid for prognosis – because our algorithm is based on a progressive inclusion of inputs, higher-order outputs gain in complexity and allow finer discrimination of states. This informs about 1) the overall state of the patient, 2) small and large risk factors, and 3) more importantly, the interaction between these risk factors. Because the algorithm is based on parallel processing and integration, any kind of human data can be inputted to the system: physical, mental, personal history, daily-routines, etc. The model is scalable and adaptive.
Aid for diagnosis and monitoring of the treatment, either in loco or remotely. This can easily be done by integrating the algorithm into the system’s architecture of telemedicine and remote biofeedback training
Bio-signals and their processes carry many information channels, which existing algorithms cannot integrate as a unity. For example, EEG and ECG, because they are electric charges, they carry electric information, magnetic information, temporal characteristics, frequency characteristics, etc. All of these signal properties are decoded in the body in very specific ways, but current human analytic procedures are incomplete and segmented
As briefly mentioned the SSS may be used to improve medical diagnosis. Current diagnosis technology tools are restricted to one task or diagnosis. Information from a predetermined databased is entered, and the system simply searches the database for likely matches. A general problem of current diagnosis practice concerns the segregation of conditions per biological subsystem and specialty of the doctors. NAME will be able to combine information across biological subsystems, adding information from other medical domains.
Suppose a patient has the following clinical symptoms: a) they have high pressure in their carotid artery; b) there is liver failure; c) they are very weak, in a way that appears like heart failure. Even if, for example, this individual might be seen by a top cardiologist, and also, separately by other top specialists, these specialists may not by themselves figure out what is wrong. One type of specialist might not have enough information about how other systems might be having an impact. Similarly, a computerized diagnostic system that is fed information on only one system may likely fail. But if all of the information from each of these specialties was ‘fed into’ Stack 1, Stack 2 then learns to weight and combine them. The correct diagnosis of pericarditis would more likely be the output.
Stacked-neural Networks will make it possible to learn from the experience of surgeons abut rare events and how to diagnose and treat them. For example, consider if the intestine is not free to be cut and moved as in a gastric, bypass due to a tumor or some other malformation. This result could instantly be added to a database for a Stacked-neural Network. It is the cases that occur less than 5% of the time that cause many of the problems in medicine. Combining information from different databases is more likely to diagnose these more rare and problematic cases.
Other kinds of diagnoses, for example, the diagnosis of different kinds of cancer would also be greatly improved with the use of a two-stack neural network. Information both about different kinds of cancer cells as well as other physiological markers relevant to cancer and its treatment could be read into neural networks in Stack 1. Stack 2 will then learn, by weighting and combining patterns, to better diagnose the specific type of cancer.
Besides diagnosis, but still in the medical domain, a two-stack neural network also has applicability in medical recovery technological tools. An example would be in neurofeedback techniques. Biofeedback and neurofeedback techniques are gaining high prevalence as a solution for many medical conditions, for example, psychosomatic and mental conditions. They are being shown to reliably substitute for the use of some drugs in treatment. However, current training schedules of neurofeedback are first programmed to train a specific region of the brain; then, the training must be temporarily suspended so that the technician can analyze how the activity in all regions changed as a result of a change in the trained one. After this limited analysis, the technician devises a second training schedule. This procedure is quite obsolete. With a second stack, this could be done in real-time, as the changes in all regions could be computed, compared and reflected upon in the second stack, automatically updating the training schedule. The technician would only need to monitor the results of training, eventually manipulating the database.