Image Process Research

The brain of the human or other mammals consists of numerous nerve cells called neurons. Each neuron, in turn, connects with other neurons through projection fibers called axons. Neuronal information, in the form of electrical pulses, are sent by the neuron through its axon(s) to the target cells to control the activity of those targets. Neurons in the brain are subdivided into systems, responsible for different functions. Examples of these functions are vision, hearing, motion, thinking, emotion, and so on. Each system are subdivided into units, which assume another level of division of labor. For example, in vision, a unit would be responsible for the detection of an object, another unit for its location in space, another for its texture, and others for its relevance to the perceiving subject, etc.

With the huge number of nerve cells in the brain (a ``handwaving'' rough estimate of ten billion), any reliable quantitative estimate of the number of neurons or axons even in a small functional unit of the brain is a very formidable, and often impractical, job if done manually. Yet, a quantitative knowledge of the number of neurons, or the number of axons is very useful. This is because the number of neurons or axons changes during the developmental stage of the subject. How it varies may also depend upon the environment and experience of the subject. How many survives to adulthood may affect the subject's adult performance. As the subject ages, or when certain diseases or damages occur, the number of neurons or axons diminishes. How this reduction in neurons or axons affect the performance of the subject can be better understood if a quantitative counting of the number can be achieved.

Even if the number of neurons or axons do not change, there are many other morphological factors that can affect performance. For example, if the sheath wrapping around the axons (called myelin sheath) deteriorates, the subject will suffer, as in the disease called multiple sclerosis. Also, the diameter of the axon is known to affect conduction rate of the electrical pulses transmitted by the axon.

Traditionally, the counting process was done by human experts. However, since each nerve bundle contains a huge number of fibers, this approach is extremely time consuming and ineffective. As a result, there is a pressing need to have such counting process be automated. The problem we study in this research involves counting and sizing of axons in a set of electron microscope images. We are interested in estimating the total number of fibers, together with their area, diameter and myelin thickness distributions, in each picture.

Image "cell.ras" shows a typical nerve image used in this research. Since the the raw image is corrupted by unwanted microstructures, it is desirable to have them removed as early as possible, otherwise it would hinder subsequent processes. Image "cellc.f.ras" shows the processed version of the raw image.

Then, we apply a boundary detection technique, commonly known as snake, to extract each fiber. The final result is shown in image "cella.out.ras". The number of fibers can be estimated by counting the number of detections whereas their size distributions can be computed from the corresponding detected boundaries.

roland@cs.ust.hk