Physics and Technology Forefronts

Flat Panel Medical Imaging

By R. A. Street

X-ray imaging for medical diagnosis began soon after Roentgen discovered x-rays in 1895, but it has taken 100 years to replace x-ray film with a digital imaging technology. Digital imaging has the benefits of immediacy in acquiring the image, electronic storage, retrieval and transmission, and enables image enhancement and computer-assisted diagnosis.

X-rays cannot be easily focused, so the big challenge in creating a digital imager is to make the detector as large as the object to be imaged. For human medical diagnosis, the size must reach 17"x17", well beyond the capability of a conventional silicon chip. Previous approaches include image intensifier vacuum tubes and laser scanned storage phosphors. Hydrogenated amorphous silicon (a-Si:H) transistors and sensors, deposited on large glass sheets and patterned into devices by photolithography, provide a compact, fully solid state digital x-ray imager. These are manufactured with the same large area processing technology that is used for liquid crystal computer displays.

The flat panel x-ray detector is a pixel array, comprising up to 10 million pixels, with pixel size from 50 to 500 microns depending on the imaging application. The essential elements are a sensor that absorbs x- rays and creates a corresponding electric charge, a capacitor to store the charge, and the active matrix addressing that organizes the readout of the signal to external electronics, which amplify, digitize and display the image. Each pixel contains one addressing transistor, and a single row of pixels is activated simultaneously by a common gate contact. The rows that make up the array are addressed in sequence to read out the whole image.

The imager technology is more than just transistors, and involves a combination of material science, semiconductor physics, imaging science, device processing and electronics design. A-Si:H transistors and sensors make the detectors possible. The a-Si:H transistor is fast enough to operate the detector at video rates. Its large (>109) on-to-off ratio is needed to hold charge on the pixel between readout events. Since the signal is stored on the sensor capacitance, low leakage is a must. Amorphous silicon has the important property of high resistance to radiation damage, in part because it has a disordered atomic structure.

Sensitive detection requires that the x-rays passing through an imaged object must be stopped in the thin film sensor. Detectors are best made from high atomic number materials. Even then, a thickness of 200-500 micron is required for the energies that are best suited to medical imaging. Since the a-Si:H sensor has neither of these attributes, a different approach is used in which the a-Si:H sensor detects light emitted by an adjacent phosphor layer, as shown in Figure 1. An incident x-ray excites a high-energy electron in the phosphor by the photoelectron effect. The electron loses energy by ionization, creating many low energy electron-hole pairs, which then recombine to emit visible light. This light emerges from the phosphor and is detected by the a-Si:H, which has excellent quantum efficiency at the 550nm emission wavelength of the two most common phosphors, CsI:Tl and GdO2S2:Tb.

Present flat-panel x-ray detectors operate in two basic modes, one of which is radiographic imaging, in which a single image is obtained, for example to identify a broken bone. Fluoroscopic imagers operate at video rates to monitor the continuous motion of internal organs or the progress of non-invasive surgical procedures. The same detectors are also used to inspect inanimate objects for security or quality control.

Most detectors now on the market use the phosphor/sensor combination, and several companies have made recent product introductions. General Electric offers several radiographic systems with image sensors made by Perkin Elmer, and Varian Medical Systems sells fluoroscopy detectors using arrays made by dpiX. Similar products have been introduced in Europe by Trixell and in Japan by Canon. Hologic offers radiographic systems using selenium sensors with a-Si:H transistor addressing.

The detector has gain in the sense that many charges are detected for each absorbed x-ray. The ionization within a semiconductor typically creates an electron-hole pair for each 5-10 eV of electron energy loss, so that a 100kV photon might develop 10,000 to 20,000 e-h pairs. However the rather complex detection process of the current technology reduces the measured sensitivity by a factor of 5-10. Furthermore, scattering in the phosphor tends to spread out the recombination light and therefore reduces the spatial resolution of the detector.

Figure 1. Diagram illustrating x-ray detection with a phosphor/sensor combination (left) and with an x-ray photoconductor (right).

An alternative approach replaces the phosphor/sensor combination with a thick x-ray photoconductor, as illustrated on the right in Figure 1. The ionization charge is collected directly by the action of an applied electric field. There is minimal lateral diffusion of the charge to reduce the spatial resolution. The photoconductor must have a large mobility-lifetime product for the collection of charge in an applied field, and low leakage, low charge trapping and low temperature deposition are also essential.

Can materials be found with suitable properties? Amorphous selenium has proven effective, although the sensitivity is no greater than the phosphor detectors. More recently, we find that the semiconductors PbI2 and, particularly, HgI2 can achieve close to the theoretical sensitivity of one electron- hole pair detected for each 5 eV of x-ray energy absorbed, which gives nearly a 10-fold improvement.

Detector sensitivity is important because x-rays damage tissue, and it is essential that the required diagnostic information be obtained with the lowest x-ray dose. However, high sensitivity is only one component of the detector performance. The modulation transfer function (MTF) describes the spatial resolution of the array, and the detective quantum efficiency (DQE) describes how well the information content of the x-ray image is captured by the detector. The photoconductor generally gives an MTF that approaches the ideal for a pixel array. It has also proved possible to obtain a high DQE (>70%) with large x-ray dose, but the challenge is to maintain this at low dose, when electronic noise starts to dominate. This is particularly important in fluoroscopy applications where the image is obtained with less than 100 photons per pixel.

In the present detectors, the charge on a pixel is transferred through the a-Si:H transistor to a common address line where it is detected by an external amplifier. Large detectors have electronic noise values of 1000-2000e, which is roughly the charge detected from a single 100 kV x-ray in present systems. The noise can be reduced by placing an amplifier in each pixel, taking advantage of the small input capacitance. We have demonstrated arrays of this type, using transistors made from laser recrystallized polycrystalline silicon rather than a-Si:H, because its larger mobility is better suited to amplifiers. A source follower provides a gain of about 10, which is enough to overcome the noise in the external amplifiers. Although this is a simple 3-transistor circuit, it opens the way to more complicated pixel electronics for added functionality. With the extra electronic devices on the pixel, there is no space for the sensor. The solution is a 3-dimensional structure in which the sensor is placed above the addressing electronics, separated by a thick passivation layer. The surface is coated completely with the a-Si:H sensor, which also follows the contours of the underlying electronics.

Flat panel x-ray detectors have only recently been introduced to the medical imaging market, but could prove to be the technology of choice for the next 100 years. Continued improvements such as those described here will allow lower patient dose, higher resolution imaging, and will extend the range of applications.

Bob Street is senior research fellow and manager of the Large Area Systems group at the Palo Alto Research Center.