MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel
  1. Home
  2. Meeting Program
  3. All Sessions
  4. All Young Investigator Symposium
  5. Machine Learning Method to Automate Incident Triage in Radiotherapy Incident Reporting and Analysis System (RIRAS)

Machine Learning Method to Automate Incident Triage in Radiotherapy Incident Reporting and Analysis System (RIRAS)

K Syed1*, J Nalluri2 , W Sleeman IV2, 3 , P Ghosh1 , M Hagan2, 3 , J Palta2, 3 , R Kapoor2, 3 , (1) Virginia Commonwealth University, Department of Computer Science, Richmond, VA (2) Virginia Commonwealth University, Department of Radiation Oncology, Richmond, VA (3) National Radiation Oncology Program, Department of Veteran Affairs, Richmond, VA

Presentations

(Monday, 7/15/2019) 1:45 PM - 3:45 PM

Room: Stars at Night Ballroom 2-3

Purpose: RIRAS receives incident reports from facilities across the VA enterprise. This effort is to automate the triage process based on the severity of the reported incidents using sophisticated machine learning methods.

Methods: The reported incidents in RIRAS are reviewed by a radiation oncology subject matter expert (SME); who initially triages incidents based on the severity by evaluating salient elements in the reporter’s form including the incident narrative. To build a machine learning model, we used 345 reported incidents spanning the whole spectrum of a RT workflow and for which an SME had manually completed the analysis. These incidents were triaged into four levels of severity, A through D, where A is most severe and D is least. The distribution of events based on severity was A (62), B (52), C (162), and D (67). The dataset was split into 70:30 ratio as training and testing datasets. Natural language processing (NLP) technique; term frequency-inverse document frequency (tf-idf) was applied to the bag-of-words model to generate feature vectors. These feature vectors were passed to the machine learning classification algorithm using a support vector machine (SVM) with a linear kernel to identify the type of severity. We built two models, model-1 by merging the severity of A&B and C&D. Model-2: just considering type A and C.

Results: The model was designed to determine the severity of the incident. The accuracy of the model-1 0.83 and model 2 was 0.87. The precision, recall, and f-measure for model-1 was 0.84, 0.83 and 0.83 and model-2 was 0.86, 0.87 and 0.87.

Conclusion: Automated triage and severity determination helps SMEs focus on high severity incidents first. The SVM model can identify high severity incidents based on the initial report and narrative. Additional data should help improve computer model accuracy and provide human-level fidelity and performance.

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email

Additional Links