Image Analysis and Stereology for Discovery and Safety Assessment

AI-Assisted Morphometric Digital Image Analysis Services

With the advent of deep learning for computer vision, we can now generate numerical endpoints using digital image analysis (DIA) with greater precision than ever before.

The adaptive nature of artificial intelligence (AI) technologies such as convolutional neural networks (CNNs) has created new challenges for regulatory agencies. But, with cooperative input from Charles River scientists and pathologists across the globe, responsiveness of the developers of AI-assisted image analysis platforms, and published guidelines from the FDA, we have developed a method of phenomenological “black box” verification testing, which allows us to generate GLP-compliant algorithms for digital image analysis of pathology endpoints.

Superior to manual collection of quantitative data, which is often performed on limited sets of study material, CNN-assisted DNA allows us to interrogate entire datasets in much less time.

Our experienced scientists can discuss endpoint-specific options and guide study and protocol design for the most appropriate fixation, collection, and sampling techniques to ensure the highest quality and most accurate, unbiased data.

Pathology Image Analysis Services and Expertise

  • CD68 stain for macrophages within a mouse tumor.Static and dynamic measurements for bone
  • Small intestine villus length, crypt depth, and villus/crypt ratios
  • Image analysis of immunohistochemically or histochemically stained slides, including number of labeled cells, area of positive staining, and tissue composition analysis
  • Determination of cell proliferation index (Ki67, BrdU, PCNA) or apoptotic index (caspase-3, TUNEL)
  • Diameter of stented blood vessels
  • Extent of infarcts in the heart
  • Capillary angiogenesis in cutaneous wound defects, including microvascular density quantification
  • Goblet cell number in the respiratory tract
  • Adult and juvenile rat cerebellar and cerebral measurements
  • Axon and nerve fiber changes in peripheral nerves
  • Skeletal muscle characterization
  • Pulmonary fibrosis assessment
  • Thyroid gland epithelial height and colloid area for endocrine disruptor assessment

 

Stereology Services and Support

Standard morphometry (pathology image analysis) relies on homologous sections captured from the structure of interest. Comparing quantitative endpoints are limited to those sections and should not be extended to the whole structure. Where homologous sections are captured is a great source of bias.

Additionally, the plane of view between two-dimensional profiles can be very misleading, resulting in overrepresentation of large and oblong objects. Stereology randomly takes samples from the entire structure and relies on unbiased geometric probes to derive quantitative endpoints from which statistically valid conclusions can be derived for the whole structure.

The principles of stereology logically extend the design of histomorphometry studies that use planar sampling methods in order to obtain measurements of the properties of whole structures and objects, rather than simply measuring the properties of the profile of these targets in mounted sections. Typical stereological endpoints are volume, number, surface area, and/or length.

For stereological analysis, we offer comprehensive services for discovery and GLP-compliant studies from conception and design through data analysis.


Stereology Services Technical Sheet

Stereology Services Technical Sheet

Charles River is the leading contract provider for state-of-the-art stereology services, analyzing whole slide images using validated software systems.
Download the Technical Sheet


Our experienced scientists can discuss endpoint-specific options and guide study and protocol design for the most appropriate fixation, collection, and sampling techniques to ensure the highest quality and most accurate, unbiased data.

  • Capabilities
    • Unbiased sampling consultation for all tissues and species
    • Systematic uniform random sampling (SURS) of all tissue types
    • Paraffin, plastic, and frozen embedding media
    • Histochemical, immunohistochemical, and immunofluorescence staining for endpoints of interest
    • Brightfield and fluorescent whole slide scanning
    • Software analysis of whole slide digital images
    • Microscope analysis of thick sections
    • Ultrastructural analysis with transmission electron microscopy
    • Statistical analysis, graphing, and data reporting
  • Applications
    • Neuronal analysis for neurodegenerative disease research
    • Sensory and sympathetic neuronal toxicity evaluation
    • Intra-epidermal nerve fiber length or density estimates
    • Alveolar assessments in animal models of lung disease
    • Bone surface area and osteoblast quantification in osteoporosis models
    • Pancreatic cell evaluations for diabetes research
    • Glomerular function assessment for diabetic nephropathy
    • Quantification and size of any cell type in mode of action (MOA) toxicity assessments

Our image analysis and stereology experts are ready to help you with a wide range of quantitative measurements for your program.

Get Started

 

Frequently Asked Questions (FAQs) about Image Analysis and Stereology

  • How many sections should be captured for image analysis and stereology assessments?

    For pilot work, we generally collect sections from 8-12 random intervals for most small tissues/regions (i.e., tissue regions that can be embedded in a single cassette or sectioned from a single block of tissue) for image analysis & stereology. For larger tissues requiring a block sampling fraction, we will apply the same rule of thumb to each block of tissue. Based upon the precision metrics generated by the image analysis & stereology pilot work, we may increase or decrease the frequency of our sectioning interval.

  • Do we need to capture thick sections for object counting?

    No; We prefer to use thin sections (3-5 microns) for object counting, using the physical dissector method for image analysis & stereology. The stereology platform we use allows automatic, seamless alignment of dissector pairs for unbiased counting. We reserve the optical dissector method for image analysis & stereology projects with high object density, which must be analyzed under an oil immersion lens at higher magnifications (60x, 80x, or 100x) than standard whole slide scanners allow (20x or 40x).

  • What is the proportionator in stereology?

    The proportionator is a method of randomly sampling from a distribution of the weights of all fields of view from the area of a region of interest based upon the presence of an object of interest. This is a feature exclusive to the Visiopharm stereology toolkit, which reduces the amount of work while also increasing precision.

  • What is a good coefficient of error for quantitative image analysis services?

    The coefficient of error (CE) by itself is not as useful for estimating precision when combined with the coefficient of variation (CV). If sampling methods are truly random, and the measurements are unbiased, then the CV will be a genuine reflection of biological variability, while the CE represents your sampling error.

    If the CE is sufficiently lower than the CV, you may be confident that any differences detected can be attributed to your independent variables rather than spurious fluctuations due to your sampling method. We measure the relative contribution of CE and CV using the precision range of an optimally balanced estimator (PROBE), which is calculated simply as CV2/CE2. At a minimum, we like to see a PROBE value of two.

  • Does artificial intelligence improve the accuracy of Digital quantitative image analysis?

    AI improves the precision of image analysis services by offering greater consistency than manual scoring, reducing variability around the mean, which can improve the ability to detect more subtle differences between groups. However, an AI-assisted algorithm is only as accurate in finding the true mean as the scientist that designed it. Additionally, algorithms are still susceptible to sampling bias. Currently, the best way to improve the accuracy of quantitative image analysis is to employ the unbiased techniques of stereology.