Program Information
Datamining and Distributed Learning in Radiation Oncology to Help Clinical Decision Support: The Australian Computer Aided Theragnostics Network for Oncology
D Thwaites1*, L Holloway2 , M Field3 , S Barakat4 , S Vinod5 , G Delaney6 , M Carolan7 , M Bailey8 , A Miller9 , J Sykes10 , R Alvandi11 , E Hau12 , V Ahern13 , J Lehmann14 , J Ludbrook15 , A Ghose16 , D Stirling17 , T Lustberg18 , J van Soest19 , S Walsh20 , A Dekker21 , (1) University of Sydney, Camperdown, Sydney, ,(2) Liverpool & Macathur Cancer Therapy Centre, Sydney, New south Wales, (3) Ingham Institute, Liverpool, NSW, (4) Ingham Institute, Liverpool, NSW, (5) Liverpool Hospital, Liverpool, NSW, (6) Liverpool Hospital, Liverpool, NSW, (7) Illawarra Cancer Care Centre, Wollongong, NSW, (8) Illawarra Cancer Care Centre, Wollongong, NSW, (9) Illawarra Cancer Care Centre, Wollongong, NSW, (10) Western Sydney Local Health District, Sydney, NSW, (11) Crown Princess Mary Cancer Centre, Westmead, NSW, (12) Crown Princess Mary Cancer Centre, Westmead, NSW, (13) Crown Princess Mary Cancer Centre, Westmead, NSW, (14) Newcastle Mater Hospital, Hunter Region Mail C, ,(15) Newcastle Cancer Centre, Newcastle, NSW, (16) University of Wollongong, Wollongong, NSW, (17) University of Wollongong, Wollongong, NSW, (18) MAASTRO, Maastricht, NT, (19) MAASTRO, Maastricht, NT, (20) MAASTRO, Maastricht, NT, (21) MAASTRO, Maastricht, NT
Presentations
TU-L-GePD-JT-4 (Tuesday, August 1, 2017) 1:15 PM - 1:45 PM Room: Joint Imaging-Therapy ePoster Theater
Purpose: Radiotherapy treatment guidelines are based on randomised clinical trial (RCT) evidence. However, many patients are not eligible for RCTs. The concept of a distributed learning network, where radiotherapy data remains at local centres but can be learnt from jointly can provide additional clinical evidence. Such a network has been established in NSW, Australia and tested on non small cell lung cancer (NSCLC) patient data across four centres.
Methods: A NSCLC radiotherapy dataset of 298 patients was extracted from one centre's database and was reviewed to determine patient numbers meeting RTOG 9410 and 0617 criteria. Lung radiotherapy data was extracted from the four centres' databases for local modelling (per centre) and for comparing and sharing model parameters. Initial work has been carried out on imputation of missing parameter values and on the impact of incorporating radiomics features into the models.
Results: Only 26-30% of patients within the NSCLC cohort met trial criteria. Lung cancer radiotherapy data has been successfully extracted at the four centres and transfer of model learning parameters between centres has been successfully demonstrated. With small size data sets, statistical imputation was shown to cope with 5%-10% of missing values with a reasonable error margin. A DSS trained with clinical variables and radiomics features achieved an AUC of 0.69 compared to 0.67 with clinical variables only (using only complete datasets, this further improved to 0.75).
Conclusion: Many patients do not meet RCT criteria on which radiotherapy treatment guidelines are based. A growing (increasing number of centres) Australian distributed radiotherapy data network has been established with international links to enable generation of additional clinical evidence to support treatment decisions for such patients. Imputation of missing values and inclusion of radiomics features show promise. The work has focussed attention on improvement of data collection and data accuracy.
Contact Email: