Early detection of lung cancer by excavating the bit planes of a LDCT scan
Principal Investigator
Name
Swapan Chakrabarti
Degrees
Ph.D.
Institution
Immersion 3D Plus, LLC.
Position Title
CTO
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-815
Initial CDAS Request Approval
Jul 15, 2021
Title
Early detection of lung cancer by excavating the bit planes of a LDCT scan
Summary
Detecting lung cancer early from Low-dose Computed Tomography (LDCT) scan is often a challenging task. The feeble modulation of intensities on an LDCT scan from the cancerous segments may not be noticeable even by the trained radiologists. A straightforward intensification of the images will inevitably amplify noise and create more obstacles for cancer detection. We propose to analyze an LDCT scan or image by accumulating and visualizing information from one bit plane at a time. Such a process reveals information from the low-intensity segments of the image that may not be visible on a complete image—the image formed by adding contributions from all the bit planes at once. We are especially interested in analyzing LDCT scans that are characterized as questionable or inconclusive for detecting lung cancer.
A monochrome LDCT scan with PxQ pixels usually possesses 8 bits per pixel as Pixel Excitation Values (PEVs). Such an image can be broken down in to 8 bit planes or binary images of PxQ pixels and label those as bit planes 0 to 7 where the bit plane number “n” has the weight 2n in forming the complete image. On adding the weighted contributions from all these 8 bit planes we get a number between 0 and 255 for every pixel, and that contributes to the formation of 256 levels of shades with luminance varying between 0.1 to 300 nits (say). So, when a trained radiologist visualizes such a complete image, the limited intensity variation from the early-stage cancerous segments may not become noticeable in the presence of a wider intensity distribution on the image. Thus, an alternative approach is proposed for this task.
We propose to analyze seven partially summed images formed by adding the weighted contributions from one bit plane at a time of the original image. Thus, the seven images are formed as the weighted summation of the bit planes from 0 to 1, from 0 to 2, and so on till we compute that summation of the bit planes from 0 to 7 which is also the complete image. When these shaded partially summed images are visualized simultaneously, the view will reveal how the imaging information has been evolved from the lower intensities to reach its complete form. Some of the images of the partial sums computed using the lower levels of the bit planes might reveal the signature of lung cancer before it disappears on the images formed by including the higher levels of partial sums. Our tests with synthetic cancer data show highly encouraging results, and we are eagerly waiting to test our approach with clinical data. The eventual goal is to show all those images automatically for a test image and help a radiologist detect lung cancer early. Our approach does not require to collect any extra images from the patient’s LDCT scan while providing the potential to detect lung cancer early, especially from those LDCT scans that are found questionable.
A monochrome LDCT scan with PxQ pixels usually possesses 8 bits per pixel as Pixel Excitation Values (PEVs). Such an image can be broken down in to 8 bit planes or binary images of PxQ pixels and label those as bit planes 0 to 7 where the bit plane number “n” has the weight 2n in forming the complete image. On adding the weighted contributions from all these 8 bit planes we get a number between 0 and 255 for every pixel, and that contributes to the formation of 256 levels of shades with luminance varying between 0.1 to 300 nits (say). So, when a trained radiologist visualizes such a complete image, the limited intensity variation from the early-stage cancerous segments may not become noticeable in the presence of a wider intensity distribution on the image. Thus, an alternative approach is proposed for this task.
We propose to analyze seven partially summed images formed by adding the weighted contributions from one bit plane at a time of the original image. Thus, the seven images are formed as the weighted summation of the bit planes from 0 to 1, from 0 to 2, and so on till we compute that summation of the bit planes from 0 to 7 which is also the complete image. When these shaded partially summed images are visualized simultaneously, the view will reveal how the imaging information has been evolved from the lower intensities to reach its complete form. Some of the images of the partial sums computed using the lower levels of the bit planes might reveal the signature of lung cancer before it disappears on the images formed by including the higher levels of partial sums. Our tests with synthetic cancer data show highly encouraging results, and we are eagerly waiting to test our approach with clinical data. The eventual goal is to show all those images automatically for a test image and help a radiologist detect lung cancer early. Our approach does not require to collect any extra images from the patient’s LDCT scan while providing the potential to detect lung cancer early, especially from those LDCT scans that are found questionable.
Aims
1. Test the effectiveness of an approach in detecting lung cancer early from clinical data. The approach has worked very well with synthetic data .
2. Specifically interested in studying the efficacy of the software in detecting lung cancer early from LDCT scan that are considered before as questionable or inconclusive.
3. Compare the performance of the software with imaging data that have been evaluated as (i) questionable and (ii) obvious cases.
Collaborators
Anjan Ghosh; CEO, Immersion 3D Plus, LLC.