![](https://csdnimg.cn/release/download_crawler_static/19898327/bg1.jpg)
Adaptive Classification of Urinary Sediment Images Using
Feedback Training
Satoshi Mitsuyama,
1
Jun Motoike,
2
and Hitoshi Matsuo
1
1
Central Research Laboratory, Hitachi, Ltd., Kokubunji, 185-8601 Japan
2
Instrument Group, Hitachi, Ltd., Tokyo, 100-8220 Japan
SUMMARY
In urinary sediment examination, untypical particles
may appear, even though very infrequently. Neural network
with feedback training proposed in this study offers im-
proved classification of such untypical particles, which was
difficult using conventional classification algorithms. The
proposed method suggests that network is optimized using
classification results obtained for typical objects. The
method may be easily applied to complicated patterns with
multiple parameters. Operation and efficiency of the pro-
posed method were confirmed by computer simulation.
Additional validation was obtained by applying the pro-
posed method to classification of urinary sediment images.
© 2001 Scripta Technica, Syst Comp Jpn, 32(2): 1118,
2001
Key words: Urinary sediment; neural network;
feedback training; pattern recognition.
1. Introduction
Urinary sediment examination aims at classification
and quantification of particles included in urine, with re-
sults being used for diagnosis of renal and urinary disorders.
Normally, urine of a healthy person includes very few red
and white blood cells as well as squamous cells. On the
other hand, with renal and urinary disorders, many blood
cells and other cells as well as crystals and other particles
appear in urine. Previously, urine was centrifuged, and thus
obtained sediment specimens were examined under micro-
scope by an expert. Since microscope examination requires
considerable effort, automatic examination was very desir-
able but this field remained most undeveloped in terms of
automation.
Aiming at automation of urinary sediment check, we
conducted R&D in image processing technology for close-
up still images of urinary particles (urinary sediment im-
ages) [1, 2], and in pattern recognition using neural
networks [3, 4]. The techniques developed so far support
automatic classification of common particles offering typi-
cal sizes and shapes, which made possible automation of
urinary screening tests.
However, certain problems still remain to be solved
before fully automatic urinary check can be realized. One
such problem is classification of rare particles with untyp-
ical shapes.
For example, images of typical red and white blood
cells are shown in Figs. 1(a) and 1(b). As seen, red blood
cells are normally much smaller than white ones, hence the
two types of blood cells can be easily distinguished by size.
However, very large swollen red blood cells as in Fig. 1(c)
or very small atrophic white blood cells as shown in Fig.
1(d) may appear, though very seldom. Classification of
such untypical particles is a difficult task even for an expert.
When an expert classifies untypical particles, he first
deals with typical particles in a given specimen, and then
proceeds to untypical particles taking into account general
trends in shape and size of typical ones. For example, when
© 2001 Scripta Technica
Systems and Computers in Japan, Vol. 32, No. 2, 2001
Translated from Denshi Joho Tsushin Gakkai Ronbunshi, Vol. J83-D-II, No. 1, January 2000, pp. 237244
11