Histology image analysis – background and challenges 

Cytology imagery provides interesting characteristics that ease the visual analysis like isolated/clustered cells and the absence of complicated structures such as glands. Moreover, this type of image often results from the least invasive biopsies, contributing to their common use in disease screening and biopsy purposes (Gurcan et al., 2009). Compared to cytology imagery, histology slides provide a more comprehensive view of diseases and their effects on tissue (Hipp et al., 2011) since their preparation preserves the underlying tissue structure (He et al., 2012). Histology analysis is performed by inspecting a thin slice (i.e. section) of tissue under an optical or electron microscope (Gartner & Hiatt, 2006; Kiernan, 1990; Mescher, 2013; Murphy & Davidson, 2001; Sternberg, 1997). The study of histology images is considered as the gold standard for clinical diagnosis of cancer and identification of prognostic and therapeutic targets. Histopathology, the microscopic study of biopsies to locate and classify diseases, has roots in both clinical medicine and basic science (Sternberg, 1997). In this section, we first summarize the production of histology images, from tissue preparation to imaging technologies. Then, we briefly review histology image analysis, its relation to other types of medical imaging, and its main challenges.

Image production

Fixation is the first stage of preparation for subsequent procedures, which should be conducted in real time to preserve the samples as well as possible. Different fixatives (e.g. precipitant and crosslinking) or methods (e.g. heat fixation and immersion) may be used. For example, the precipitant fixatives (e.g. methanol, ethanol, acetone, and chloroform) dehydrate the tissue samples, removing lipids and reducing the solubility of proteins. After fixation, the tissue must be adequately supported, e.g. frozen or embedded in a solid mold, to allow sufficiently thin sections to be cut for microscopic examination. Common treatments employ a series of reagents to process the fixed tissue and embed it in a stable medium such as paraffin wax, plastic, or resin. Such treatments include the main steps of dehydration, clearing, infiltration, and embedding (Chandler & Roberson, 2009; Nelson, Lehninger & Cox, 2008; Wootton, Springall, Polak et al., 1995).

The embedded tissue sample is finally cut into thin sections (e.g. 5μm for light microscopy and 80 − 100nm for electron microscopy). The transparent sections are usually produced with a microtome, an apparatus feeding the hardened blocks through a blade with high precision. After cutting, the sections are floated in warm water to smooth out any wrinkles. Then, they are mounted (by heating or adhesives) on a glass slide. Once they are attached on the slide, the process is reversed prior to staining. The wax is removed with a solvent (usually xylenes) and the tissue is re hydrated through a series of solutions in which the alcohol – water ratio is changed. The gradual rehydration preserves tissue architecture. Now, the sections are ready for staining, which helps to enhance the contrast and highlight specific intra- or extra cellular structures. A variety of dyes and associated staining protocols are used. The routine stain for light microscopy is hematoxylin and eosin (H&E); other stains are referred to as special stains for specific diagnostic needs. Each dye binds to particular cellular structures, and the color response to a given stain can vary across tissue structures. For example, hematoxylin stains the nuclear components of cells dark blue and eosin stains the cytoplasmic organelles varying shades of pink, red, or orange. (Kiernan, 1990; Ross, Kaye & Pawlina, 2003) provide a detailed description of common laboratory stains. After staining, the stained section on the slide is covered to protect the tissue and provide better visual quality for microscope examination.

After the tissue has been prepared, light microscope (Murphy & Davidson, 2001; Török & Kao, 2007) is used to take digital histology images of the stained sections. Additional details on different types of microscopes and image production are provided in (He et al., 2010,1).

Image analysis

In histology image analysis for cancer diagnosis, histopathologists visually inspect the regularities of cell shapes and tissue distributions. Such histopathological study has been extensively employed for cancer detection and grading applications, including prostate (Doyle, Madabhushi, Feldman et al., 2006a; Doyle, Hwang, Shah et al., 2007), breast (Basavanhally, Agner, Alexe et al., 2008; Doyle, Agner, Madabhushi et al., 2008), cervix (Guillaud, Cox, Malpica et al., 2004; Guillaud, Adler Storthz, Malpica et al., 2005), and lung (Jütting, Gais, Rodenacker et al., 1999; Kayser, Riede, Werner et al., 2002) cancer grading, neuroblastoma categorization (Gurcan et al., 2006; Kong, Shimada, Boyer et al., 2007b), and follicular lymphoma grading (Cooper, Sertel, Kong et al., 2009; Kong, Sertel, Lozanski et al., 2007a)

Histopathology has attracted researchers from different disciplines including clinical medicine, biology, chemistry and machine learning. Computer-based image analysis has become an increasingly important field due to the high rate of production and the increasing reliance on these images by the biomedical community. Medical image processing and analysis in radiology (e.g. X-ray, ultrasound, CT, MRI) and cytology have been active research fields for several decades with numerous systems (Bankman, 2008; Greenberg, 1984; He, 2009; Yoo, 2004) and products (Lamprecht, Sabatini & Carpenter, 2007; Schroeder, Ng & Cates, 2003) developed. However, the application of these systems in histology analysis is not straightforward due to the significant difference in the imaging techniques and image characteristics.

The complexity of histology images is defined by several factors including overlapping tissue types and cell boundaries and nuclei corrupted by noise; some structures, such as cell boundaries, may appear connected or blurred. These factors make it difficult to extract cell regions (e.g. nuclei and cytoplasm) by traditional image segmentation approaches. On the other hand, cytology images are taken at higher magnification level which results in clearly identified cell compartments. Computer-based histology analysis systems generally exploit a much larger quantity of image features to derive clinically meaningful information than similar systems for radiology and cytology (He et al., 2012). Nevertheless, the image analysis systems for these three domains generally consist of a common sequence of steps of image restoration, segmentation, feature extraction, and pattern classification.