Engineering an image processing based computer aided diagnostic (CAD) systems is very popular research topic nowadays, and a really hard problem. The reason for the popularity of CAD systems, that making diagnose from radiographs always monotone and hard work. In many cases (for example the subject of my thesis: detecting nodules in chest X-rays) the radiologists don't agree about the question that is it possible to make good quality diagnoses, with enough sensitivity by using a kind of imaging process (in this case the chest X-rays), or not. This shows that the nodule detection from chest X-ray is not trivial, nor for the radiologists.
In my thesis I show a new, own engineered fully automated CAD system, which task is detecting lung nodules in chest X-rays. The main expectation was to implement an alternative approach of the Department's implemented CAD system in order to cooperate with that system. I implemented the CAD system in MATLAB environment. The exact goal of my CAD is detecting nodules according to malignant soft tissue tumors, segmenting these nodules in posterior – anterior chest radiographs. I tested the full CAD system both with the database of the Japanese Society of Radiological Technology (this is a standard benchmark database of the problem), and with the private databese of the Pulmonological Clinic of Semmelweis University (called seventh database).
I begin my thesis with representing the theoretical background in medical context, then I detail the classical CAD systems architectures, then I represent the components, what I designed in detail. I validate every component with both the JSRT and the seventh database. In the end of my thesis I analyse the capability of the full CAD system with images not used in the CAD construction from the Pulmonological Clinic of Semmelweis University’s database, and with the JSRT database due to the fact, that I used only the seventh database in the CAD system creating.