Although machine learning (ML) has shown promise across disciplines, out-of-sample generalizability is concerning. This is currently addressed by sharing multi-site data, but such centralization is challenging/infeasible to scale due to various limitations. Federated ML (FL) provides an alternative paradigm for accurate and generalizable ML, by only sharing numerical model updates. Here we present the largest FL study to-date, involving data from 71 sites across 6 continents, to generate an automatic tumor boundary detector for the rare disease of glioblastoma, reporting the largest such dataset in the literature (n = 6, 314). We demonstrate a 33% delineation improvement for the surgically targetable tumor, and 23% for the complete tumor extent, over a publicly trained model. We anticipate our study to: 1) enable more healthcare studies informed by large diverse data, ensuring meaningful results for rare diseases and underrepresented populations, 2) facilitate further analyses for glioblastoma by releasing our consensus model, and 3) demonstrate the FL effectiveness at such scale and task-complexity as a paradigm shift for multi-site collaborations, alleviating the need for data-sharing.
With the increased use of cone beam CT (CBCT) for daily patient setup, the accumulated dose from CBCT may be significantly higher than that from simulation CT or portal imaging. The objective of this work is to measure the dose from daily pelvic scans with fixed technical settings and collimations. CBCT scans were acquired in half-fan mode using a half bowtie and x-rays were delivered in pulsed-fluoro mode. The skin doses for seven prostate patients were measured on an IRB-approved protocol. TLD capsules were placed on the patient's skin at the central axis of three beams: AP, left lateral (Lt Lat) and right lateral (Rt Lat). To avoid the ring artefacts centred in the prostate, the treatment couch was dropped 3 cm from the patient's tattoo (central axis). The measured AP skin doses ranged 3-6 cGy for 20-33 cm separation. The larger the patient size the less the AP skin dose. Lateral doses did not change much with patient size. The Lt Lat dose was approximately 4.0 cGy, which was approximately 40% higher than the Rt Lat dose of approximately 2.6 cGy. To verify this dose asymmetry, surface doses on an IMRT QA phantom (oval shaped, 30 cm x 20 cm) were measured at the same three sites using TLD capsules with 3 cm table-drop. The dose asymmetry was due to: (1) kV source rotation which always starts from the patient's Lt Lat and ends at Lt Lat. Gantry rotation gets much slower near the end of rotation but dose rate stays constant and (2) 370 degrees scan rotation (10 degrees scan overlap on the Lt Lat side). In vivo doses were measured inside a Rando pelvic heterogeneous phantom using TLDs. The left hip (femoral head and neck) received the highest doses of approximately 10-11 cGy while the right hip received approximately 6-7 cGy. The surface and in vivo doses were also measured for phantoms at the central-axis setup. The difference was less than approximately 12% to the table-drop setup.
1 January 2015 in International Journal of Radiation Oncology - Biology - Physics (IJROBP)