Individual micro calcifications are difficult to be detected as they are variable in shape and size and may be embedded in areas of dense parenchymal tissues. One of the most important problems of medical diagnosis, in general, is the subjectivity of the pattern recognition by diagnosis experts. This is due to the fact that the results are depended on the interpretation of the input from the patients but not on systematic procedure. In this paper, an adaptive neuro-fuzzy model optimized by PSO algorithms has been proposed. The symptoms and signs are gathered and the fuzzy membership values are defined. Feed forward multilayer networks are used to accept the fuzzy input values and is trained using back-propagation algorithm. The system is tested for detecting the micro-calcifications in breast sonograms. Later the results are compared for its performance.

No abstract

Frequent patterns play vital role in generating association rules. The frequent patterns are generated from a huge transaction database as a first step and strong association rules are generated as the next step. The input database contains transactions which consist of transaction identifier and a set of items. A number of algorithms have been proposed to determine frequent patterns. Apriori algorithm is the first and foremost algorithm proposed in this field. It mines the frequent patterns by scanning the database as {Tid, itemset}. Vertical data format technique uses {item, TidSet} way of scanning the database to mine frequent patterns efficiently. In the second approach, the transaction database is transformed into vertical format for mining frequent patterns and intersection method is used to find support count. In both the above algorithms, a huge number of candidate sets are generated which are then pruned using Apriori property. This pruning process generates a huge collection of subsets for each candidate set. These subsets are pruned for existence in prior level frequent sets. This makes an overhead in terms of memory and time. In this paper, a different technique namely Direct-vertical, is proposed that improves the performance in terms of memory and time consumption. This algorithm is based on both Apriori and vertical data format and is proved better than other algorithms in terms of number of subsets and candidate sets.