Fusion Techniques in Biomedical Information Retrieval

Abstract

For difficult cases clinicians usually use their experience and also the information found in textbooks to determine a diagnosis. Computer tools can help them supply the relevant information now that much medical knowledge is available in digital form. A biomedical search system such as developed in the Khresmoi project (that this chapter partially reuses) has the goal to fulfil information needs of physicians. This chapter concentrates on information needs for medical cases that contain a large variety of data, from free text, structured data to images. Fusion techniques will be compared to combine the various information sources to supply cases similar to an example case given. This can supply physicians with answers to problems similar to the one they are analyzing and can help in diagnosis and treatment planning.

Biomedical compound figure detection using deep learning and fusion techniques

“…The individual model class scores are combined with score-based fusion operators defined by Fox and Shaw [21]: combMAX, combMIN, combSUM, combANZ, combMNZ. An additional operator combPROD is defined as the cumulative product of all the individual scores [30]. The input scores to the fusion operators are the estimated probabilities of an input image belonging to each class.…”

“…The combSUM and equivalent operators were adopted as the common point of comparison as they behave like the simple average or arithmetic mean of scores applied in other models [14,30]. The combPROD operator was adopted as it behaves differently from combSUM and equivalent operators in that it penalises low scores among combined individual classifier results more heavily.…”

Late fusion of deep learning and handcrafted visual features for biomedical image modality classification

“…However, usually only the average score is taken, which behaves similarly to the ‘combSUM’ fusion operator. We also explore other fusion operators, such as ‘combMAX’, ‘combMIN’ and ‘combPROD’ [56] or ‘combMED’ that takes the median score. The input scores are taken as the estimated posterior probabilities from the SVM and Softmax classifiers.…”

Medical case retrieval

“…An overview of late fusion techniques based on ranks or scores of individual result lists as well as proposals for linear score combination and score normalization are given by Wu [32]. A recent comparison of late fusion techniques on the ImageCLEF 2013 MCR data set showed that linear score combination of text and image retrieval is superior over individual retrieval [13]. Rahman et al [22] successfully applied fusion techniques to multimodal biomedical image retrieval.…”