Automated Recognition of Lung Tumors in CT Scans:
Quantitative comparison of tumors before and after treatment.
Computed Tomography (CT) image taken before
treatment shows a large tumor in the bottom left lung
(right in this image). This type of lung tumor, a
Bronchioloalveolar Carcinoma or BAC, can be highly
variable in appearance, from small nodules to large areas
such as imaged here.
treatment shows a large tumor in the bottom left lung
(right in this image). This type of lung tumor, a
Bronchioloalveolar Carcinoma or BAC, can be highly
variable in appearance, from small nodules to large areas
such as imaged here.
The strategy to recognize and measure the
tumor is to progressively isolate the lung cavity
using a series of thresholds to remove the
background from the body, and then to isolate the
pleural cavity as much as possible. The threshold
shown here in red contains all the normal lung
parenchyma and much of the tumor.
tumor is to progressively isolate the lung cavity
using a series of thresholds to remove the
background from the body, and then to isolate the
pleural cavity as much as possible. The threshold
shown here in red contains all the normal lung
parenchyma and much of the tumor.
Segmenting the thresholded portion of the image
above and adjusting the threshold to include only
normal lung tissue permits the segmentation of the
normal lung tissue in the image below.
above and adjusting the threshold to include only
normal lung tissue permits the segmentation of the
normal lung tissue in the image below.
egmented normal lung tissue can be measured
for area in this image, and in successive slices of
the CT scan, the volume can be determined from
the distance between slices.
for area in this image, and in successive slices of
the CT scan, the volume can be determined from
the distance between slices.
Algorithms must be developed to distinguish tumor
areas from normal tissue in the isolated pleural
cavity image. Other areas of increased density such
as blood vessels and normal tissues at the lung
boundaries will be included in any simple threshold
segmentation.
areas from normal tissue in the isolated pleural
cavity image. Other areas of increased density such
as blood vessels and normal tissues at the lung
boundaries will be included in any simple threshold
segmentation.
Once the tumor has been recognized,
it can be segmented for further processing.
it can be segmented for further processing.
The tumor image is first bitmapped to
allow for binary processing.
allow for binary processing.
an Erosion and Dilation cycle remove thin
processes that are likely artifacts.
processes that are likely artifacts.
The separated "particles" of the tumor can then be joined into a
single "particle" using an algorithm that joins the closest parts.
single "particle" using an algorithm that joins the closest parts.
A Convex Hull algorithm creates a mask that can be used
to find the complete tumor on the original image.
to find the complete tumor on the original image.
The original image can now be masked with the
outline of the tumor found in the above processing
routine, creating a Region of Interest (ROI). The
ROI is copied from the original image and pasted
to the segmented pleural cavity image as shown below.
outline of the tumor found in the above processing
routine, creating a Region of Interest (ROI). The
ROI is copied from the original image and pasted
to the segmented pleural cavity image as shown below.
Thresholding all tissues denser than
normal lung parenchyma now completely
segments the tumor.
normal lung parenchyma now completely
segments the tumor.
Pasting the copy from the original image
completes the tumor.
completes the tumor.
Thresholding all tissues denser than
normal lung parenchyma now completely
segments the tumor.
normal lung parenchyma now completely
segments the tumor.
The segmented tumor can be measured
and veiwed in 3D.
and veiwed in 3D.
The segmented tumor can be measured and veiwed in 3D.
