A pixel-based autofocusing technique for digital histologic and cytologic slides Vincenzo Della Mea a, * , Federico Viel b , Carlo A. Beltrami b a Department of Mathematics and Computer Science, University of Udine, v.delle Scienze 206, 33100 Udine, Italy b Department of Medical Morphological Research, University of Udine, Udine, Italy Received 24 November 2004; revised 23 February 2005; accepted 23 February 2005 Abstract The present paper describes a method for autofocusing specifically studied for the acquisition of digital slides, i.e. full histologic and cytologic slides, utilising low-cost equipment. At first, experimentations with some of the most used focus measures and algorithms have been made, in order to choose the most suitable for histologic and cytologic images. Then, a study of the specific features of digital slides has been preliminarily carried out in order to understand the constraints of the domain. These included the capability of autofocusing in an unattended way on thousands of microscope fields, while fast performance is not a strict requirement. Based on the findings, an algorithm based on a dynamic focus position space, with a variance-based focus measure has been adapted to the specific situation. A qualitative and quantitative evaluation of the proposed algorithm allowed us to show that the proposed algorithm is suitable for the acquisition of digital slides, and furthermore it can be implemented starting from a basic microscope with an inexpensive motorised stage. The algorithm is currently implemented into a complete digital slide acquisition system, which is in turn being used for a Quality Assurance Programme in cervicovaginal cytologic screening. q 2005 Elsevier Ltd. All rights reserved. Keywords: Microscope imaging; Digital slides; Telepathology; Autofocus function; Focus search algorithm 1. Introduction Telepathology has been traditionally constituted by the set of techniques for transmitting images acquired from a glass slide seen through a microscope. Main aims for doing that are distant diagnosis, expert consultation, distant education, remote image processing and analysis [1]. The two usual telemedicine categories of real-time and store-and-forward systems apply to telepathology too, with the same cost and practical consequences, but with a peculiar aspect: store-and-forward telepathology usually involves the pre-selection of supposedly representative images, which are then sent at distance for second opinion diagnosis, continuing education, etc. Although the diag- nostic performance, when the image sampler is an expert pathologist, is reported as sufficiently good [2,3], pathologists do not like to receive pre-selected information [4]: they fear the possibility of missing the diagnosis by not having the right images. In the very recent times, the availability of faster computers, with huge amounts of storage available, and of novel acquisition devices such as digital cameras, drove towards attempts to acquire the complete glass slide and use it instead of selected images for representing the whole case, thus overcoming the main limitation of store-and-forward telepathology and also opening the way to novel applications. The basic idea is straightforward: by means of a microscope equipped with a motorised stage, and using an high-magnification objective, a matrix of images is acquired in an automatic way. The mosaic obtained by collating the images together is a digital representation of the glass slide. The number of pixels theoretically needed for capturing all information on a glass slides has been calculated as about 10,000 pixel/mm (linear) [5], although in common diag- nostic practice this value might be easily divided by 3–4. The quality of the image is however also dependent on the focus. Although a glass slide appears like a really flat Computerized Medical Imaging and Graphics 29 (2005) 333–341 www.elsevier.com/locate/compmedimag 0895-6111/$ - see front matter q 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.compmedimag.2005.02.004 * Corresponding author. Fax: C39 0432 558461. E-mail address: dellamea@dimi.uniud.it (V. Della Mea).