Strategies for efficient virtual microscopy in pathological samples using JPEG2000

Abstract

This paper describes the design, implementation and validation of a new strategy for efficiently browsing large microscopical images (mega-images). A mega-image is constructed by registering a sequential set of microscopic fields of view, compressed and stored in hard disk using the JPEG2000 standard (J2K). Navigation is accelerated by fully exploiting J2K properties through the introduction of a cache strategy and an optimal delivering of quality information. Cache is introduced at the level of the spatial and resolution dimensions while optimal delivering is implemented on the organisation of minimal information units. Navigation with the conventional use of J2K results in extraction times of about 500 ms. We show that these strategies can improve navigation velocities up to a 30%, while we can efficiently represent high-quality and high-resolution colour images of microscopic specimens.

Introduction

Navigation through large images has come to be a useful tool in many different domains and for different applications, among which we can include various types of images such as satellite, astronomic, medical images or image mosaicking Chaudhuri, 2001, Li and Sun, 2003. Acquisition, archiving and navigation on large pathological images is a very recent domain for the image processing community. An entire digitalisation of this type of samples is called a virtual slide, which makes possible, among others, image retrieval for latter studies, medical training, distribution by electronic media and image exchange between pathologists Johansen et al., 2005, Chang et al., 2003. Virtual microscopy is associated with a virtual slide, which can be considered as a digital representation of an entire slide at a high magnification. Slides are merely too large to be acquired as a single image, so instead it is needed to capture many fields of view before combining them into a single slide image, called here a mega-image. In general, although this pathology field, called virtual microscopy Humphries et al., 1997, Chang et al., 2003, Bustamante et al., 1998, has reached a high degree of functionality, it is still limited regarding the mega-image construction, storing and navigation procedures. Hence, a virtual microscope requires different strategies at each of these processes.

In the acquiring process, it is quite frequent that the capturing frame is overlapped with its neighbours in order to avoid possible information losses. Several sources of errors come out in the process: variable illumination conditions between different fields of view, geometric deformations due to the radial camera distortion and aligning errors because of the microscope stage backlashes (Bradley et al., 2005). Provided that the latter factor is the major source of inaccuracy, the problem of registration is straightforwardly introduced i.e the mega-image reconstruction is similar to a puzzle problem in which these small pieces must be precisely gathered together and registration methods are then used to find an optimal displacement match. This registration problem for virtual microscopy has been approached by greedy algorithm solutions (Bradley et al., 2005) or by dynamic programming (Wildermoth et al., 2005). Similarity has been measured using normalised cross correlation (Rosenfeld and Kak, 1982) or Fourier Phase Shifting (Bracewell, 1965).

Storage and compression for virtual microscopy have been approached with different strategies: JPEG compression, TIFF, GIF Mamata et al., 2003, Mullick et al., 2003 or LZW (Johansen et al., 2005) algorithms. Fontelo and co-workers (Johansen et al., 2005) propose an archiving system for virtual microscopy using JPEG standard with losses. Yet this system yields a good compression rate, it fails at following the pathologist quality needs. Besides, the system makes no spatial or resolution progressive access, which results in a very inefficient navigation since even the simpler magnification versions need an extraction of the entire set of data. Zhang Zhang and Wang, 2002, Lo et al., 1999 takes advantage of the multiresolution nature of the Haar wavelets for obtaining resolution scalability and carries out coding by blocks for spatial progressivity. However, this approach makes no quality progressivity which results in a major drawback for image transmission through narrow band channels. Lately, J2K turns out to be an actual alternative for navigation due to its multiple advantages and compromise between fast access and minimum disk space requirements. Bradley and co-workers (Wildermoth et al., 2005) propose a virtual microscopy system with an extended depth of field or multilayer biopsy, which in the registering phase uses a normalised cross-correlation similarity measurement. Storing is performed through J2K, but because of some implementation limits the least information units for decompression are the tiles, which correspond to the primary J2K image partition that could not be optimal for navigation since they need to be large enough for avoiding distortion effects at low bit rates (Rabbani and Rajan, 2002). Moreover, Bradley does not exploit the minimal coding unit allowed by the standard: the precinct (ISO/IEC JTC1/SC29 WG1, 2003). Finally, the proposed system has no strategies for improving the navigation velocity.

In this paper an innovative J2K-based virtual microscope was designed, developed and evaluated. Virtual microscopy was here three-fold approached by developing registration, store and navigation modules. Firstly, mega-images are automatically constructed with a standard registration method, using similarity measures such as the normalised cross correlation, phase registration or mutual information. These mega-images are then compressed and stored, using the J2K standard with an appropriate setting of spatial, resolution and quality parameterisations. For navigation, the structure of the J2K codestream is conveniently accessed for obtaining specific regions of the image. Besides, a cache strategy is designed and implemented for improving navigation velocity. The entire system is a user friendly interface which allows the pathologist to navigate through mega-images as he would by using a microscope. Main contribution of this paper lies on a exhaustive analysis of the most relevant issues of a virtual microscope system, so that each of the modules can be modified after the presented recommendations, based on a detailed study from different perspectives by exploiting up-to-date tools.

The paper is organised as follows. In Section 2 we describe the problem which is approached from the standpoint of these three issues:

• Registering;

• Compression and storage;

• Displaying and navigation.

Section 3 describes a general methodology for constructing a mega-image, as well as a complete study of all the compression, storage, displaying and navigation issues, focusing on the capabilities of the J2K standard. Section 4 presents the results and finally Section 5 is dedicated to discussion and conclusions