The customer is improving their Electron Paramagnetic Resonance Imaging (EPRI) system that is used for the diagnosis of tumors. Using this non-invasive method, the concentration of oxygen in living tissues can be determined without the need for biopsies. Tumors that have little or no oxygen content (i.e. hypoxia) also have increased resistance to radio-frequency radiation, making it less effective as a choice for cancer treatment. Identifying those patients with hypoxic tumors allow alternative treatments to be used instead.
The requirement is to capture the returned RF signals with the highest possible vertical-resolution (i.e. 12-bits or more). The transmitted 200 to 1000 Watt RF pulse lasts 30 to 75 nanoseconds with a central frequency of 300 MHz. Since the same coil is used both for transmission and for reception, there is a short 200 to 300 ns dead-time before the receiver is turned on. They must capture the returned decay signals for about 4 to 8 microseconds. In order to eliminate systematic noise build-up, the signal is collected from four different quadrant angles. Each detected RF signal is down-converted to 20 MHz. Since the received signal-level is so very small, the signal must be averaged extensively.
Their first set-up used an 8-bit oscilloscope communicating via GPIB, which sampled at 500 MS/s. A complete measurement frame consisted of 37 gradient-scans. Each scan entailed capture of 4,000 samples on two channels, each of which was averaged 50,000 times to improve signal quality. This signal repeated at a Pulse Repeat Frequency (PRF) of 50 kHz. A 2-dimentional image frame took about 2 minutes to complete. The 145 gradient-scans required for a 3-dimensional image took approximately 8 minutes.
Their current set-up uses two 8-bit ISA-bus A/D cards. The same 37 scans can occur at a faster PRF of 200 kHz and now only takes about 1.5 minutes for a 2-D image (or 145 scans within about 4 minutes for a 3-D image).
In the two cases above, summing was done by onboard accumulators of 24-bits (i.e. 64K averages of 8-bit data) and 22-bits (i.e. 16K averages of 8-bit data), respectively. Note that, in the second case, three iterations would be necessary to accumulate an overall 48K (i.e. 49152) averages.
Their intended set-up, using a new "single-point" imaging technique, would increase the number of scans from 37 for 2-D all the way up to 400 to 900 scans at a resolution of 12-bits or more.
Obviously, the one thing that the customer needs is speed. GaGe's CompuScope 14100 provides both the necessary high-resolution and the fast PCI-bus transfer-speeds. The customer will use two CS14100-64M cards configured in Master/Slave mode for true simultaneous sampling on 4 channels at up to 50 MS/s.
The CS14100's 63 dB of Signal-to-Noise Ratio would allow the reduction of the number of averages from 50,000 for an 8-bit digitizer down to only about 1,000. Since summing/averaging is done using the PC's processor, the hardware limitations of an on-board processor is removed, allowing for a more flexible averaging algorithm.
Acquiring the minimum-sized record of 256 samples at 50 MS/s with the CS14100 would take about 5 m s. Using the PCI bus-master transfer-rates of over 100 MB/s, transferring the resulting 256*2 = 512 Bytes (since there are two bytes per 14-bit sample-point) from the card into the PC-RAM would take about 5ms per channel or 20ms for 4 channels. Rearming the cards to prepare for the next trigger takes about 30ms per card or 60ms for 2 cards. In PC RAM, summing the data from each 250 sample record into a running signal averaging buffer will take about 7ms. The total trigger handling time for one record is, therefore; 5 + 20 + 60 + 7 » 100ms. Consequently, the capture and processing of 50,000 triggers would take about 5 seconds. This is a best-case estimate, however, it significantly exceeds the more than one minute measurement times of the previous two solutions.
Instead of off-loading each record individually as described above, the Multiple Record mode of operation could be used to stack the records into on-board memory for a later bulk offload. Acquiring each of the minimum-sized records of 256 samples at 50 MS/s would again take about 5ms. Within 10's of sampling clock cycles (say, within 0.5ms at 50 MS/s), the cards would automatically rearm themselves to be prepare for the next trigger. The capture of 50,000 records would require an on-board memory of:
256 Samples/record * 50,000 records/scan * 2 channels » 25 MegaSamples
Using the PCI bus-master transfer-rates of over 100 MB/s, transferring the resulting bulk of data from the cards into PC-RAM would take:
Total Data = 25 MegaSamples/card * 2 cards * 2 bytes/Sample = 100 MB
PCI Transfer Time = Total Data / (100 MB/s) = 100 MB / (100 MB/s) = 1 second
The averaging time is again 50,000 * 25 ns = 1.25 seconds. The capture and processing of 50,000 triggers would therefore take less than 2.5 seconds. Because of this improved scan time, and because Multiple Record mode is resistant to Windows PCI bus latencies, GaGe recommended Multiple Record Mode and an on-board memory of 64 MegaSamples per card. Between scans, all the data points for one MulRec scan are offloaded across the PCI-bus into PC-RAM.
The customer was easily able to incorporate the Multiple Record sample-program from the CompuScope "Software Development Kit (SDK) for C/C++ for Windows" into his existing Windows application. This sample-program is intentionally simplified for easy transfer into any given software application.
The combination of GaGe's superior hardware and versatile software products enabled the customer to quickly move from a preliminary prototype to a fully-commercialized diagnostic system. Using the GaGe solution is at least double the current performance (and for over 10 times the data).
We encourage you to contact us and discuss your medical application in more detail with our engineering team. GaGe can provide tailored custom data acquisition hardware and software solutions to meet specific application requirements.