Digital Health Innovation and Informatics

PV QA 2 - Poster Viewing Q&A 2

MO_18_2656 - Machine learning can improve the accuracy and efficiency of output factor estimation for electron cutouts

Monday, October 22
10:45 AM - 12:15 PM
Location: Innovation Hub, Exhibit Hall 3

Machine learning can improve the accuracy and efficiency of output factor estimation for electron cutouts
L. Kofman1,2, and J. Chang2,3; 1Computer Science and Biology, Tufts University, Medford, MA, 2Radiation Medicine, Northwell Health, Lake Success, NY, 3Department of Physics and Astronomy, Hofstra University, Hempstead, NY

Purpose/Objective(s): Estimating the output factor of an electron cutout from the library of previously measured data is a tedious task for physicists. This manual task generally involves area approximation of the electron cutout, reference to measured output factors of similar cutouts, and data interpolation. The hypothesis of this study is that artificial intelligence can learn from the process performed by physicists and estimate output factor for electron cutouts. The purpose of this study is to develop an artificially intelligent tool that can automatically calculate the output factor of electron cutouts with minimal human intervention.

Materials/Methods: This artificial intelligent tool first reads the DICOM RT Plan files and extracts the electron beam information, including cone size, field size, energy, source-to-surface distance, beam block (i.e., contour of the electron cutout), and other data necessary to compute the output factor for an electron beam. The program executes area approximation by interacting with the OpenCV computer vision module; it blurs the reference image of the cutout contour with Gaussian Blur, coupled with canny detection to achieve optimal contour detection. The program then bounds a rectangle of minimum area around the contour. The dimensions of the rectangle (in pixels) are extracted and converted into field size units (in centimeters), and square field sizes are estimated. The output factor is computed by the square root method and interpolated using the measured output factors of standard electron cutouts (e.g., 10×10 cm2, 6×6 cm2, 3×3 cm2, etc.) for each electron energy value, stored in an on-line database, and accessed by Python’s indexing methods.

Results: Two methods of interpolation, by field size and by area, were first tested using the output factors of standard cutouts. Interpolation by field size yielded a smaller percent error for 60% of the cases. The average percent error was 0.8% for interpolation by field size, and 1.0% for interpolation by area. This tool was then tested with 14 electron cutouts, with output factors that were already measured for clinical use. All 14 DICOM RT files were successfully imported, and the computer vision module recognized the shape of each electron cutout. When interpolation by field size was used, the estimated output factors were within an average error of 1.7% in comparison to the measured data. Seven of the 14 output factor values featured percent errors of less than 0.7%.

Conclusion: An artificial intelligent tool using computer vision to mimic the operation of human physicists has been developed for interpolating the electron output factors from previously measured data of standard electron cutouts. Preliminary tests indicate that interpolation by field size is more precise than interpolation by area. This tool is accurate and reliable in that its sufficient intelligence can potentially eliminate the tedious manual calculations of output factor for electron cutouts and significantly reduce calculation time.

Author Disclosure: L. Kofman: None. J. Chang: None.

Send Email for Lior Kofman


Assets

MO_18_2656 - Machine learning can improve the accuracy and efficiency of output factor estimation for electron cutouts



Attendees who have favorited this

Please enter your access key

The asset you are trying to access is locked. Please enter your access key to unlock.

Send Email for Machine learning can improve the accuracy and efficiency of output factor estimation for electron cutouts