Automated Z-drive for Nikon FN-1 microscope

A problem of conventional microscopy is that you never just see the plane you focus on, but rather a composite of the plane that is in focus plus all the planes above and below that are not in focus. This blurs the image and makes it difficult to see details. There are techniques that solve this problem such as confocal and multi-photon microscopy. However, these techniques have their own shortcomings (e.g. long time to acquire images, bleaching, cost of equipment, etc.). One way to improve the results of conventional fluorescence microscopy is to not just take a single image, but rather a stack of images. An image is acquired, the focus slightly changed, a second image is acquired, the focus slightly changed.... You get the idea. This stack of images can then be deconvoluted, i.e. software is used to guess the relevant information, and to substract the out-of-focus blur. That is possible because one knows how a particular point is blurred above and below the focus plane - it's so called point spread function. Deconvolution reverses the process. It starts with the blurred image of a point and calculates back how the point must looked like to create this image.

Figure 1: Z-drive controller and stepper motor which connects to the rear focus drive of the FN-1 microscope via a rubber O-ring.

For this technique to work, you must know the image parameters (magnification, pixel size, etc.) and have a stack of precisely spaced images. E.g. for a image taken with a 10x lens the focus planes should be something like 1.17 µm apart, for 40x lens 0.293 µm apart. And you might need several dozen images to make up a stack that goes sufficiently above and beyond the structure of interests - and to give the deconvolution algorithm something to work with. A motorized focus (Fig. 1) is clearly an advantage for such an undertaking.

Figure 2: The stand-alone controller can also be connected to the imaging computer via serial port. Figure 3: The stepper motor that dirves the microscope's focus. It's axis is mounted in a ball bearing.

The z-drive developed is based on a PIC 16F690 microcontroller that drives a stepper motor via a MC3479 stepper motor driver. A HIN232ECP RS232 driver, a two line LCD display, and 4 buttons provide an interface to control the functions (Fig. 2). In the first version, the stepper motor was connected to the back focus knob on the right side of the FN-1 with a rubber O-ring. The size of the wheel on the stepper motor was slightly smaller (1.056") than the knob on the microscope (1.1") so that 100 steps would result in one full turn of the focus knob (100 µm, Fig. 3). This is necessary as each step of the motor moves the wheel by 3.75° (instead of the 3.6° needed). The slippage of the rubber belt caused some imprecision (and we hate those), so it was replaced by a timing belt (Fig. 5, SmallParts INC. part #TBPN6-25/D, #TBPN6-24/D, and # TB6-160).

Figure 4: Inside of the controller box. A PIC 16F690 microcontroller is located on the upper circuit board.

The software for PIC micro controller is written in PicBasic Pro. It allows to set the step size in 1 µm steps, and to step one step up or down. This would be the most basic way to capture an image stack: focus on the bottom most image of the stack, take a picture, press the 'up'-button, take a picture, ... More convenient is the computer controlled mode. In this mode the controller listens for commands at the RS232 port. Nikon's NIS Elements AR is set to initialize the RS232 port at startup. You then capture an automatic "Timelapse Series" which is set to execute a script after each image taken. This script contains a single command: it sends the 'focus upwards' command to the controller via the RS232. Keith Gembala from Morrell Instruments kindly suggested this way - and it works great. It provides a simple, fully automatic method to capture an entire image stack at the push of a single button. A 100 image stack takes less than 5 minutes to acquire.

Figure 5: Z-drive mounted on a Nikon FN-1 research microscope. The final version of the drive uses a timing belt instead of the O-ring to increase precision.
A heat sink was also added to the motor as it gets quite a bit warm during prolonged operation.

The stack is then imported into commercial image deconvolution software (AutoDeblur from MediaCybernetics) and processed. The result is shown in Fig. 6. On the left side is the image of part of a neuron that was filled with a fluorescent dye (40x lens). The detailed structure of secondary and tertiary branches is pretty much obscured. A few minutes later a complete image stack was acquired and after processing it we obtained a much more detailed picture of the neuron's process.

Figure 6: The result: Left is the tip of a neuron's process as seen in the brightfield microscope and - in direct comparison - after deconvolution. The image on the right was obtained by filling two neurons with fluorescent dyes of different emission wavelengths. The two neurons were then imaged separately through different filters, the two stacks deconvoluted, and the resulting projections aligned in Photoshop. The relative position of the two neurons within the ganglion is clearly visible. Considering the minimal effort the image took (~ 2 minutes acquisition time per stack plus the time needed for the deconvolution and alignment) a great result!