News |  Sitemap |  Contact
PDF Export  | 

Intelligent imaging using Discrete Mereotopology

Abstract

We introduce methods suited for developing intelligent imaging procedures with ImageJ, based on the Region Connection Calculus. In this context, intelligent imaging applications are those designed to perform a certain level of mechanical reasoning about the image contents [1].

The usefulness of intelligent procedures in image analysis is two-fold: i) they relax the need for human interaction in, for example, high throughput applications where observer-based confirmation of results is not possible owing to the size of the collected data or to timing constraints and ii) they enable an algorithmic assessment of the results of imaging procedures. For example, segmentation methods can be mechanically tested against an expected model of image contents. Such an approach can be used to qualify the correctness of the segmentation results. Furthermore, knowledge of the relations between regions can be enriched by means of conceptual neighbourhood diagrams [1]. These are pre-computed graphs encoding the possible changes in a relation when one of the regions undergoes a specified minimal change (e.g. after a morphological dilation, the relation can change from external contact to partial overlap, or after one morphological erosion, it can change from external contact to being disconnected). Knowledge of the possible changes in the relations can be used to indicate which additional morphological operations in the algorithms can be applied so the expected image content models are better fulfilled.

The methodology presented grafts a spatial logic called Discrete Mereotopology [2] onto the Mathematical Morphology domain to provide a set of spatial relations which can be used to describe the topology and structural organisation of digital images of cells and tissues. These spatial relations are a set of contact and part-whole relationships in discrete 2D space that can hold between pairs of binary regions in a single image as well as between regions across different images (such as in multi-channel images, where labelled structures of the same image are encoded in separate channels). Specifically, the discrete versions of the relation sets RCC5 and RCC8, well known in Qualitative Spatial Reasoning [3], were implemented as a plugin for ImageJ. The plugin provides a means to compute spatial relationships between regions and also enables the implementation of histologically relevant models such as cells and tissues where the relations of their parts can be assessed algorithmically. While we use a 2D model of discrete space, the models for Discrete Mereotopology are not restricted to this, and regions could be modelled as volumes or include an additional temporal dimension.

References

1. Randell DA, Landini G, Galton A. Discrete mereotopology for spatial reasoning in automated histological image analysis. IEEE Transactions on Pattern Analysis and Machine Intelligence (in press).

2. Galton AP, “The Mereotopology of Discrete Space”, in Spatial Information Theory: Cognitive and Computational Foundations of Geographic Science, C. Freksa and D. M. Mark eds., Proceedings of COSIT'99, 1999 (Lecture Notes in Computer Science 1661), pp. 251-266. Berlin: Springer, 1999.

3. Randell DA, Cui Z, Cohn AG, “A Spatial Logic Based on Regions and Connection”, in Proceedings of KR-92, pp. 165-176, 1992.

Keywords

Mereotopology, Mathematical morphology, Artificial intelligence, Automated reasoning, Qualitative spatial reasoning,

Short CV

Gabriel Landini is a Professor of Analytical Pathology and Head of the Oral Pathology Unit at the School of Dentistry, The University of Birmingham, UK. He received a degree in Dentistry from the Universidad de la Republica, (Montevideo, Uruguay) and a PhD in Oral Pathology from Kagoshima University (Kagoshima, Japan). His research is focussed on the development of reliable, quantitative, mathematically-based measures of complexity on the spatial domain applied to diagnostic problems in Pathology. Prof. Landini's research background covers both classical Pathology and Medical Digital Imaging, specially applied to the problems of tumour shape, cell complexity, tissue segmentation.

Administrative data

Presenting author: Gabriel Landini
Organisation: University of Birmingham, UK

co-authors: David A Randell, University of Birmingham, UK

Antony Galton, University of Exeter, UK

© Luxembourg Institute of Science and Technology | Legal Notice