Robotics Reference
In-Depth Information
danger or it could be a false alarm. Even highly trained radiologists some-
times miss a tell-tale sign that would necessitate further investigation of
the breast. A computerized vision system can now assist in the mammo-
gram analysis process at Innovis Health in Fargo, North Dakota, where
radiologists have accepted the system as a screening tool, and the tech-
nology is becoming standard practice for many medical centres.
The computer produces an image, onscreen or on a printout, with
markings to call attention to suspicious areas of the breast. Usually a
spot will turn out to be benign. But studies show that computer-aided
detection has found a significant number of cancers missed by radiolo-
gists. The goal is to help doctors do a better job of catching cancer early,
which is often crucial to a patient's chances of survival. The five-year
survival rate for breast cancers spotted at an early stage is 96 percent,
compared with 20 percent for cancers detected so late that they have that
have spread.
The radiologists who use the system do not expect computers to re-
place them. Ultimately it is still up to an experienced radiologist to de-
cide on the best course of action once the computer system has done its
job. The computer prompts doctors to take more second looks, which
means calling the patient back for another mammogram, this time tar-
geted at the suspicious area. The computer is especially good at identi-
fying tiny deposits of calcium, which sometimes are cancerous but are
usually harmless. Such deposits can be difficult to see for a human radi-
ologist, but the computer imaging system excels at identifying them.
Speech Recognition
The “ears” of a computer are microphones, devices that contain some
sort of diaphragm that vibrates in concert with any audible sound. The
vibrations of the diaphragm are converted into electrical signals, which
in turn can be displayed as a waveform on a screen or measured electroni-
cally. A speech waveform has an irregular shape and no two are the same.
Figure 9 in Chapter 1 (page 33) shows a typical speech waveform.
In essence the speech recognition problem is one of recognizing wave-
forms such as this one. If everyone's voice produced the same waveform
for the same word then there would be no problem at all, but this is far
from what actually happens. Not only do different people say the same
word in different ways, with different accents, different stresses, different
Search WWH ::




Custom Search