Cancer in moles is not always easy to see with the
naked eye. By analysing images, a new computer
program can detect cancerous moles automatically.
We get hot in the sun and shed our clothes. No one
wants to wreck a summer’s idyll by reminding people
about the risk of skin cancer from sunbathing, but
the risk is there nonetheless. Sunshine and melanoma
are related.
The incidence of cancer in young adults, especially
in women, appears to be on the rise. Melanoma is one
of the most dangerous forms of skin cancer, but if
it is detected and treated early, the chance of
recovery is good. Researchers at NTNU have developed
a diagnostic program to help doctors decide if a
mole is cancerous or if it’s harmless.
“We aren’t trying to take over making the diagnosis
— that’s still the doctor’s job. But this method can
be a valuable tool to make better diagnoses,” says
Jon Yngve Hardeberg, a professor at the Norwegian
Colour and Visual Computing Laboratory.
Melanoma can be difficult to detect, because in the
earliest stages of development cancerous moles are
very similar to benign moles.
Having good procedures to detect melanoma at an
early stage and being able to differentiate it from
benign changes to the skin is critical.
Since the doctor initially does this through a
visual examination, a lot depends on a doctor’s
experience in assessing moles.
“Our aim was to develop a fast and effective
solution to classifying melanoma. This method
simulates the sense of sight and looks for certain
patterns in the mole. The method is very similar to
human visual perception,” says postdoctoral fellow
Tomas Majtner at NTNU’s Colourlab.
More specifically, a set of algorithms does the job
in five stages.
A mole is photographed and the image is then
processed by a computer program, which detects and
singles out the interesting area of skin.
Then the program analyses the texture of the mole on
the basis of numerous properties, in order to
automatically classify it as cancerous or harmless.
The special feature of this method is that it
penetrates into and uses the texture of the mole as
the subject of analysis, rather than looking at its
colour.
The method was developed based on a set of test
images from the University of Tromsø. With a
classification tool like this, doctors can avoid
removing benign moles unnecessarily.
Besides saving valuable time, this method can be an
important aid in cases where the doctor is in doubt,
and especially in training doctors to detect
melanoma or for general practitioners who don’t work
with melanoma everyday. The challenge ahead will be
to develop user-friendly and reasonably priced
systems that can be used by more doctors.
“Collaborating with other disciplines is important
in this field. We have expertise in colour imaging
and image quality, which can provide exciting and
useful results when combined with others’ expertise
in biomedical optics and dermatology. We’ll continue
to cooperate with medical research groups,” says
Hardenberg.
Majtner says that the method has an accuracy of 97
per cent in tests so far. Researchers plan to test
the method on another dataset of images that are
more difficult to interpret, and see if they can
develop the method for additional types of skin
cancer.
“We’re not quite there yet, but one possible future
application of the method would be an app that
allows you to check your own moles,” says Majtner.
The research is part of the research project IQ-MED.
For more information
This article has been accepted for presentation at
the International Conference on Image Analysis and
Recognition conference which will take place from 13
to 15 July in Portugal
Link...
Lecture Notes in Computer Science
Efficient Melanoma Detection Using Texture-Based
RSurf Features
Link...
Norwegian Colour and Visual Computing Laboratory
Link...
NTNU - Norwegian University of Science and
Technology
Link...
MDN |