The Department of Electronics, Informatics and Systems (DEIS) at the University of Calabria in Italy has designed an integrated software system, called Computer Aided Medical Diagnosis or CAMD, as a partnering contribution within the Euromed project. Under the coordination of Professor Lucio Grandinetti, the team has developed an automatic classifier which discriminates between benign and malignant cells from a breast cancer. The CAMD tool has been embedded into a telemedicine system, based on two different functionalities, one using a Client/Server approach and the other relying on Web facilities. This implementation enables the physician to perform remote diagnostical consultations. In the future, the research team hopes to extend the application of the automatic diagnostic system to other types of cancers or diseases, such as brain and thyroid cancers.
Early and accurate diagnosis of breast cancer largely improves the patient's chances for long term survival. Usually, the cancer is detected as a lump in the breast by self-examination or mammography. The physician determines the nature of this lump using Fine Needle Aspirate (FNA), a simple, non-invasive, outpatient procedure by which a sample of fluid is taken from the breast mass and mounted on a microscope slide for visual interpretation of the morphological characteristics of the cellular nuclei. However, the assessment of this clinical data turns out to be highly subjective, depending upon the skills and experience of the examining doctor. Therefore, an urgent need exists to increase the sensitivity of FNA diagnosis by computerising the process with use of image processing techniques and machine learning methods.
In the Parallel Computing Laboratory (Parcolab) at the University of Calabria, the CAMD system has been composed of two stages in order to refine the degree of objectivity within the diagnosis. First, a graphical computer programme is activated for the analysis of cytological features, based on digital scanning of the FNA samples by means of a video camera and a frame-grabber board. The digital image then is stored as a file in a suitable format to be displayed on a PC monitor for further determination of each nucleus and its exact boundary. The programme, running on Windows 95, computes ten features for each nucleus, resulting in a total of 30 real-valued nuclear features for each FNA sample.
Second, a training set, based on the creation of a 30-dimensional features vector, is generated by performing the analysis for each individual on a large set of patients, for which the actual diagnostic outcome is known. An automatic classifier applies the process of inductive learning, using a Linear Programming model to discriminate among points related to benign cells and points related to malignant ones. To implement the CAMD tool for remote diagnosis, two ways of proceeding are possible. The first one is based on the client-server paradigm functioning via Internet. The server might be situated in an advanced medical centre where a whole collection of tested clinical data is being stored to build an efficient training set as well as a solid automatic classifier upon.
On the client-side, which can be an isolated region without medical expertise on cancer diagnosis, the physician is able to extract and measure the cytological characteristics from the FNA sample in order to send them to the server. In turn, the server receives the queries for execution of the classification stage and transmits the results of the diagnosis to the local physician for visualisation. The second method consists of the actual CAMD- integration into the Web environment by means of Common Gateway Interface (CGI) or Java Web-based applications.
Using CGI, the local doctor has to access a protected medical image archive hosted in a central hospital to select the needed image with his web browser. Subsequently, a sophisticated image processing algorithm constructs the suspected cells boundaries by extracting the morphological features. They serve as inputs to a classifier operating via linear programming. Using Java, the physician equally accesses the hospital web server to select the relevant image and start the image and diagnosis processing with the help of an applet. All these computing tasks are performed by the browser, keeping the server free to handle other incoming requests. Currently, the Java-approach is still suffering from insufficient speed. The research team is therefore working hard on optimising the telemedicine facilities of the CAMD tool. You can follow all recent developments on the Parcolab site of the University of Calabria.