Use of computer based decision tools to aid clinical decision making, has been a primary goal of research in biomedical informatics. Research in the last five decades has led to the development of Medical Decision Support (MDS) applications using a variety of modeling techniques, for a diverse range of medical decision problems. This paper surveys literature on modeling techniques for diagnostic decision support, with a focus on decision accuracy. Trends and shortcomings of research in this area are discussed and future directions are provided. The authors suggest that-(i) Improvement in the accuracy of MDS application may be possible by modeling of vague and temporal data, research on inference algorithms, integration of patient information from diverse sources and improvement in gene profiling algorithms; (ii) MDS research would be facilitated by public release of de-identified medical datasets, and development of opensource data-mining tool kits; (iii) Comparative evaluations of different modeling techniques are required to understand characteristics of the techniques, which can guide developers in choice of technique for a particular medical decision problem; and (iv) Evaluations of MDS applications in clinical setting are necessary to foster physicians' utilization of these decision aids.

This work relates to the development of computational algorithms to provide decision support to physicians. The authors propose a Fuzzy Naive Bayesian (FNB) model for medical diagnosis, which extends the Fuzzy Bayesian approach proposed by Okuda. A physician's interview based method is described to define a orthogonal fuzzy symptom information system, required to apply the model. For the purpose of elaboration and elicitation of characteristics, the algorithm is applied to a simple simulated dataset, and compared with conventional Naive Bayes (NB) approach. As a preliminary evaluation of FNB in real world scenario, the comparison is repeated on a real fuzzy dataset of 81 patients diagnosed with infectious diseases. The case study on simulated dataset elucidates that FNB can be optimal over NB for diagnosing patients with imprecise-fuzzy information, on account of the following characteristics - 1) it can model the information that, values of some attributes are semantically closer than values of other attributes, and 2) it offers a mechanism to temper exaggerations in patient information. Although the algorithm requires precise training data, its utility for fuzzy training data is argued for. This is supported by the case study on infectious disease dataset, which indicates optimality of FNB over NB for the infectious disease domain. Further case studies on large datasets are required to establish utility of FNB.

The potential of computer based tools to assist physicians in medical decision making, was envisaged five decades ago. Apart from factors like usability, integration with work-flow and natural language processing, lack of decision accuracy of the tools has hindered their utility. Hence, research to develop accurate algorithms for medical decision support tools, is required. Pioneering research in last two decades, has demonstrated the utility of fuzzy set theory for medical domain. Recently, Wagholikar and Deshpande proposed a fuzzy relation based method (FR) for medical diagnosis. In their case studies for heart and infectious diseases, the FR method was found to be better than naive bayes (NB). However, the datasets in their studies were small and included only categorical symptoms. Hence, more evaluative studies are required for drawing general conclusions. In the present paper, we compare the classification performance of FR with NB, for a variety of medical datasets. Our results indicate that the FR method is useful for classification problems in the medical domain, and that FR is marginally better than NB. However, the performance of FR is significantly better for datasets having high proportion of unknown attribute values. Such datasets occur in problems involving linguistic information, where FR can be particularly useful. Our empirical study will benefit medical researchers in the choice of algorithms for decision support tools.

A new method for analysis of microarray gene expression experiments referred to as Sum-based Meta-analytical Enrichment(SME) is proposed in this manuscript. SME is a combined enrichment and meta-analytical approach to inferon the association of gene sets with particular phenotypes. SME allows enrichment to be performed across datasets,which to our knowledge was not earlier possible. As a proof of concept study, this technique is applied to datasets fromOncomine, a publicly available cancer microarray database. The genes that are significantly up-/down-regulated(p-value ≤ 10-4) in various cancer types in Oncomine were listed. These genes were assigned to biological processesusing GO annotations. The SME algorithm was applied to identify a list of GO processes most deregulated in 4 majorcancer types. For validation we examined whether the processes predicted by SME were already documented in literature.SME method identified several known processes for the 4 cancer types and identified several novel processeswhich are biologically plausible. Nearly all the pathways identified by SME as common to the 4 cancers were found tocontribute to processes which are widely regarded as cancer hallmarks. SME provides an intuitive yet objective ‘process-centric’ interpretation of the ‘gene-centric’ output of individual microarray comparison studies. The methods describedhere should be applicable in the next-generation sequencing based gene expression analysis as well