MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel
  1. Home
  2. Meeting Program
  3. All Sessions
  4. All Therapy SNAP Oral
  5. A Machine Learning-Based Approach to Specify the Cause of Error in IMRT Patient Specific QA

A Machine Learning-Based Approach to Specify the Cause of Error in IMRT Patient Specific QA

S Utsunomiya1*, M Sakai1 , H Koarai1 , T Takizawa2 , N Kushima2 , S Tanabe2 , H Aoyama3 , (1) Graduate School of Health Sciences, Niigata University, Niigata, Japan, (2) Niigata University Medical and Dental Hospital, Niigata, Japan, (3) Graduate School of Medical and Dental Sciences, Niigata University, Niigata, Japan

Presentations

(Sunday, 7/29/2018) 5:05 PM - 6:00 PM

Room: Karl Dean Ballroom A1

Purpose: To develop a machine learning-based approach to specify the dominant cause of error in patient specific quality assurance (QA) for intensity modulated radiation therapy (IMRT).

Methods: We created IMRT treatment plans (“error plans�) with an intentional error for 10 prostate cancer patients previously treated with IMRT using Novalis TX (Brainlab) at our hospital between June and September 2014. The type of errors were single MLC positional error in 5 mm (type-I), setup error of the detector used in IMRT QA (type-II), and an error in MLC transmission factor of the treatment planning system Eclipse (Varian) in +10% (type-III). We performed a measurement of dose distribution of the “error plans� using a 2D diode array MapCHECK (SUN NUCLEAR) and compared it to the case of no error. We created 2D histograms in which the vertical axis shows the measured dose and the horizontal one the calculated one. We analyzed them to extract the patterns of three types of error using machine learning methods (k-NN, SVM and ensemble method) with MATLAB (MathWorks) and evaluated the created models using ROC analysis.

Results: There were some outliers in the histogram for the type-I error. The points in the histograms were spread on both sides from the diagonal line symmetrically for the type-II error. There were no patterns detected for the type-III error. The AUC (area under the curve) of ROC analysis ranged from 0.63 to 0.92 for the type-I error and 1.00 for type-II error.

Conclusion: We have shown that the proposed machine learning-based approach to specify the dominant cause of error in patient specific QA for IMRT was feasible for a MLC positional error and setup error of the detector.

Keywords

Intensity Modulation, Quality Assurance

Taxonomy

TH- External beam- photons: Quality Assurance - IMRT

Contact Email

Additional Links