DECISION SUPPORT SYSTEM BASED ON DCT TEXTURE FEATURES FOR DIAGNOSIS OF ESOPHAGITIS

Esophagitis is essentially inflammation of the esophageal squamous mucosa. One of the major reasons for cause of Esophagitis is the acid reflux from the stomach. This condition is observed in the process of upper gastro-intestinal tract endoscopy and the diagnosis is arrived at by examining the images of the esophagus. The diagnosis is based on the observation of the lesions and coloration of the digestive mucosa. Our paper reports an implementation of Decision Support System (DSS) for diagnosis of Esophagitis based on the analysis of color and texture features of the images captured during the process of endoscopy. The Hue Saturation and Intensity color model is adapted. The statistical features of the Hue and Saturation form the color features and the texture features are determined by Discrete Cosine Transform coefficients of the image. The decision making structure is a feed forward neural network. The DSS has been tested and results are reported.

A new concept toward computer-aided medical diagnosis--a prototype implementation addressing pulmonary diseases

The original concept that led to the structuring of a computer-based medical decision support system (DSS) that is able to support a physician's diagnosis is introduced in this paper. The concept's implementation modeled a generic DSS, the core of which are an integrated knowledge/information base (KIB) along with the inference properties of a data evaluator. The KIB encapsulates the necessary medical knowledge and experience in the form of rules and constraints, preemptive tasks, and actual patients' clinical data. The data evaluator handles approved medical subjective and objective criteria for assessing the KIB's data. The data evaluator incorporates a medical standard data gathering and decision process, structured upon the principles of the clinical differential diagnosis methodology and has been integrated in the system by means of both algorithmic and artificial intelligence techniques. The novel model and the resulted computer-based package have been extensively tested under the Pulmonary Department, University Regional Hospital Patras, Patras, Greece.

A new perspective in learning pattern generation for teaching neural networks

This article deals with the Learning Patterns (LPs)' generation, a major aspect of Feed-Forward Artificial Neural Networks (FANNs)' learning process. Currently, more work is done to understand the mechanisms and improve the speed, learning accuracy, and implementation features of FANNs' teaching algorithms, though little is done towards the development of enhanced techniques that would extract experts' knowledge (from examples, rules, etc.) and obtain standardised LPs that would improve this learning process. A new approach in generating LPs is thereby introduced, that is used to train a new Medical Decision Support System (MDSS) based on FANNs, and its performance is analysed and compared with previous methods. It can handle incomplete data archives, individually boost any particular dataum special characteristics, and its application induces the FANNs to show better convergent facets. The efficiency of the resulting MDSS was thoroughly tested by pulmonologists and haematologists using medical data archives of a regional hospital.