Artificial intelligence together with mechanical imaging in mammography
(2019) MSFT01 20191Medical Physics Programme
- Abstract
- Artificial intelligence (AI) and mechanical imaging (MI) have been used in separate studies
in breast imaging. They have individually shown great possibilities within the field of
mammography, but the use of the two techniques together have never been evaluated. The
artificial intelligence application used in this work was Transpara, a deep learning
convolutional neural network. It distinguishes patterns in the mammographic images and
provides scores of individual findings and the whole mammographic examination, which
indicates a level of suspicion for breast cancer. Mechanical imaging is a surface stress
measurement, that provides information of the mechanical structure of the underlying tissue.
Since malignant tumours often... (More) - Artificial intelligence (AI) and mechanical imaging (MI) have been used in separate studies
in breast imaging. They have individually shown great possibilities within the field of
mammography, but the use of the two techniques together have never been evaluated. The
artificial intelligence application used in this work was Transpara, a deep learning
convolutional neural network. It distinguishes patterns in the mammographic images and
provides scores of individual findings and the whole mammographic examination, which
indicates a level of suspicion for breast cancer. Mechanical imaging is a surface stress
measurement, that provides information of the mechanical structure of the underlying tissue.
Since malignant tumours often express a higher relative pressure compared to the surrounding
tissue in the breast, mechanical imaging is comparable with palpation but could provide even
more information of the mechanical structures.
The purpose of this work was to study if the combination of the two methods could be used
to directly detect breast cancer. Screening images of 118 women were analysed in Transpara,
and the pressure distribution measurement of the same women was obtained from a previous
study on MI. For 46 cases, there was compression pressure present over the AI-findings, and
these were chosen to be included in the analysis. Locations of findings with the highest level
of suspicion and the corresponding locations in the pressure measurement were used to
calculate the mean relative pressure over a finding. The cases were divided into three groups
by diagnosis; biopsy-proven cancer, biopsy-proven benign and non-biopsied, very likely
benign. The increased pressure was then compared among these three groups and the two
groups of cancer and healthy, to evaluate if the increased pressure over Transpara scores of
women diagnosed with cancer was different from those diagnosed as healthy. The correlation
between increased pressure and Transpara score was evaluated for each group, to evaluate if
the two methods found the same indications for breast cancer.
The results of this study indicated that there probably are differences in increased pressure
between cases with breast cancer and healthy, but it remains to further evaluated for a larger
material. A significant and relatively strong correlation between the relative pressure increase
over an AI-finding and the Transpara scores was established in the group with cancer, but
the other groups showed no correlation.
This study indicates that MI combined with AI can potentially be used to improve the
performance of mammography screening. It suggests that AI and MI find independent
markers that coincide in breast cancer. Therefore, the two methods have the potential of
lowering the recall rate in mammography, but this needs to be further evaluated. (Less) - Popular Abstract (Swedish)
- Bröstcancer är den vanligaste cancerdiagnosen för kvinnor och årligen drabbas över 9000 i
Sverige. För att hitta cancern i ett tidigt stadium genomförs regelbunden screening
(massundersökning) med mammografi (bröströntgen) för kvinnor i åldern 40–74. Artificiell
intelligens (AI) och mekanisk avbildning, Mechanical Imaging (MI), har använts i separata
studier inom mammografi för att underlätta granskningen av mammografibilderna.
Teknikerna har individuellt visat stor potential, men användningen av de två teknikerna
tillsammans har aldrig utvärderats.
AI-programmet som använts i den här studien är ett djupinlärningsprogram som hittar fynd
utifrån misstänksamma mönster i mammografibilderna. Programmet poängsätter också
fynden,... (More) - Bröstcancer är den vanligaste cancerdiagnosen för kvinnor och årligen drabbas över 9000 i
Sverige. För att hitta cancern i ett tidigt stadium genomförs regelbunden screening
(massundersökning) med mammografi (bröströntgen) för kvinnor i åldern 40–74. Artificiell
intelligens (AI) och mekanisk avbildning, Mechanical Imaging (MI), har använts i separata
studier inom mammografi för att underlätta granskningen av mammografibilderna.
Teknikerna har individuellt visat stor potential, men användningen av de två teknikerna
tillsammans har aldrig utvärderats.
AI-programmet som använts i den här studien är ett djupinlärningsprogram som hittar fynd
utifrån misstänksamma mönster i mammografibilderna. Programmet poängsätter också
fynden, vilka indikerar en nivå av misstänksamhet för bröstcancer. Mechanical Imaging är en
mätning av trycket vid bröstets yta med hjälp av mekaniska sensorer då det komprimeras i
samband med bildtagningen. Mätningen ger information om den underliggande vävnadens
mekaniska strukturer och om det finns något styvt område i bröstet. Elakartade tumörer har
påvisats vara styvare än godartade förändringar och tekniken skulle därför kunna användas
som ett komplement vid screeningtillfället.
I arbetet har mammografibilder med tillhörande tryckmätningar från 46 kvinnor använts för
att analysera AI-fynd tillsammans med tryckbilderna. Av dessa kvinnor hade 12 cancer, 15
hade benigna (godartade) tumörer och 19 hade diagnostiserats som friska. AI-programmet
analyserar mammografibilder närmast som en radiolog. Det ger ett värde för hela
mammografiundersökningen på en skala på 1–10 hur pass sannolikt det är att kvinnan har
cancer, där 1 motsvarar en låg sannolikhet och 10 motsvarar hög sannolikhet för att kvinnan
har cancer.
Resultatet av studien visar att det finns tendenser till att ett ökat tryck över ett AI-fynd
skiljer sig mellan sjuka och friska, och att metoderna tillsammans kan hitta oberoende
markörer som båda sammanfaller i bröstcancer. Användbarheten av MI och AI tillsammans
har därför stor potential att kunna minska andelen återkallade kvinnor, om metoden skulle
införas vid screeningtillfället. Det här är första utvärderingen av MI och AI tillsammans, och
vidare studier krävs för att utvärdera metodernas fulla potential. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9008576
- author
- Bejnö, Anna LU
- supervisor
- organization
- course
- MSFT01 20191
- year
- 2019
- type
- H2 - Master's Degree (Two Years)
- subject
- language
- English
- id
- 9008576
- date added to LUP
- 2020-05-06 20:49:40
- date last changed
- 2020-05-06 20:49:40
@misc{9008576,
abstract = {{Artificial intelligence (AI) and mechanical imaging (MI) have been used in separate studies
in breast imaging. They have individually shown great possibilities within the field of
mammography, but the use of the two techniques together have never been evaluated. The
artificial intelligence application used in this work was Transpara, a deep learning
convolutional neural network. It distinguishes patterns in the mammographic images and
provides scores of individual findings and the whole mammographic examination, which
indicates a level of suspicion for breast cancer. Mechanical imaging is a surface stress
measurement, that provides information of the mechanical structure of the underlying tissue.
Since malignant tumours often express a higher relative pressure compared to the surrounding
tissue in the breast, mechanical imaging is comparable with palpation but could provide even
more information of the mechanical structures.
The purpose of this work was to study if the combination of the two methods could be used
to directly detect breast cancer. Screening images of 118 women were analysed in Transpara,
and the pressure distribution measurement of the same women was obtained from a previous
study on MI. For 46 cases, there was compression pressure present over the AI-findings, and
these were chosen to be included in the analysis. Locations of findings with the highest level
of suspicion and the corresponding locations in the pressure measurement were used to
calculate the mean relative pressure over a finding. The cases were divided into three groups
by diagnosis; biopsy-proven cancer, biopsy-proven benign and non-biopsied, very likely
benign. The increased pressure was then compared among these three groups and the two
groups of cancer and healthy, to evaluate if the increased pressure over Transpara scores of
women diagnosed with cancer was different from those diagnosed as healthy. The correlation
between increased pressure and Transpara score was evaluated for each group, to evaluate if
the two methods found the same indications for breast cancer.
The results of this study indicated that there probably are differences in increased pressure
between cases with breast cancer and healthy, but it remains to further evaluated for a larger
material. A significant and relatively strong correlation between the relative pressure increase
over an AI-finding and the Transpara scores was established in the group with cancer, but
the other groups showed no correlation.
This study indicates that MI combined with AI can potentially be used to improve the
performance of mammography screening. It suggests that AI and MI find independent
markers that coincide in breast cancer. Therefore, the two methods have the potential of
lowering the recall rate in mammography, but this needs to be further evaluated.}},
author = {{Bejnö, Anna}},
language = {{eng}},
note = {{Student Paper}},
title = {{Artificial intelligence together with mechanical imaging in mammography}},
year = {{2019}},
}