Are We There Yet? The Value of Deep Learning in a Multicenter Setting for Response Prediction of Locally Advanced Rectal Cancer to Neoadjuvant Chemoradiotherapy.

This retrospective study aims to evaluate the generalizability of a promising state-of-the-art multitask deep learning (DL) model for predicting the response of locally advanced rectal cancer (LARC) to neoadjuvant chemoradiotherapy (nCRT) using a multicenter dataset. To this end, we retrained and validated a Siamese network with two U-Nets joined at multiple layers using pre- and post-therapeutic T2-weighted (T2w), diffusion-weighted (DW) images and apparent diffusion coefficient (ADC) maps of 83 LARC patients acquired under study conditions at four different medical centers. To assess the predictive performance of the model, the trained network was then applied to an external clinical routine dataset of 46 LARC patients imaged without study conditions. The training and test datasets differed significantly in terms of their composition, e.g., T-/N-staging, the time interval between initial staging/nCRT/re-staging and surgery, as well as with respect to acquisition parameters, such as resolution, echo/repetition time, flip angle and field strength. We found that even after dedicated data pre-processing, the predictive performance dropped significantly in this multicenter setting compared to a previously published single- or two-center setting. Testing the network on the external clinical routine dataset yielded an area under the receiver operating characteristic curve of 0.54 (95% confidence interval [CI]: 0.41, 0.65), when using only pre- and post-therapeutic T2w images as input, and 0.60 (95% CI: 0.48, 0.71), when using the combination of pre- and post-therapeutic T2w, DW images, and ADC maps as input. Our study highlights the importance of data quality and harmonization in clinical trials using machine learning. Only in a joint, cross-center effort, involving a multidisciplinary team can we generate large enough curated and annotated datasets and develop the necessary pre-processing pipelines for data harmonization to successfully apply DL models clinically.

3D printing for rapid prototyping of low-Z/density ionization chamber arrays.

PURPOSE: To explore 3D printing for rapid development of prototype thin slab low-Z/density ionization chamber arrays viable for custom needs in radiotherapy dosimetry and quality assurance (QA). MATERIALS AND METHODS: We designed and fabricated parallel plate ionization chambers and ionization chamber arrays using an off-the-shelf 3D printing equipment. Conductive components of the detectors were made of conductive polylactic acid (cPLA) and insulating components were made of acrylonitrile butadiene styrene (ABS). We characterized the detector responses using a Varian TrueBeam linac at 95 cm SSD in slab solid water phantom at 5 cm depth. We measured the current-voltage (IV) curves, the response to different energy beam lines (2.5 MV, 6 MV, 6 MV FFF) for various dose rates and compared them to responses of a commercial Exradin A12 ionization chamber. We measured off-axis ratio (OAR) for several small field static multi-leaf collimators field sizes (0.5-3 cm) and compared them to OAR data obtained for commissioning of stereotactic radiotherapy. RESULTS: We identified the printing capability and the limitations of a low-cost off-the-shelf 3D printer for rapid prototyping of detector arrays. The design of the array with sub-millimeter size features conformed to the 3D printing capabilities. IV-curve for the array showed a strong polarity effect (8%) due to the design. Results for the parallel plate and the array compared well with A12 chamber: monitor unit (MU) dependence for the array was within a few % and the response to different energy beam lines was within 1%. Off-axis dose profiles measured with the array were comparable to dose profiles obtained in water tank and stereotactic diode after accounting for the size of the chambers. Dose error was within 2% at the center of the profile and slightly larger at the penumbra. CONCLUSIONS: Rapid prototyping of ion chambers by means of low-cost 3D printing is feasible with certain limitations in the design and spatial accuracy of the printed details.

Self-powered nano-porous aerogel x-ray sensor employing fast electron current.

PURPOSE: We developed a new class of aerogel-based thin-film self-powered radiation sensors employing high-energy electron current (HEC) in periodic multilayer (high-Z | polyimide aerogel (PA) | low-Z) electrode microstructures. MATERIALS: Low-Z (Al) and high-Z (Ta) electrodes were deposited on 50 µm-thick PA films to obtain sensors with Al-PA-Ta-PA-Al structures. Sensors were tested with x rays in the 40-120 kVp range and with 2.5 MV, 6 MV, and 6 MV-FFF linac beams (TrueBeam, Varian). Performance of PA-HEC sensors was compared to commercial A12 Farmer ionization chamber as well as to radiation transport simulations using CEPXS/ONEDANT with nanometer-to-micrometer spatial resolution. The computations included periodic and single-element structures N x (Al-PA-Ta-PA-Al) with variable layer thicknesses. RESULTS: Signal from PA-HEC sensors was proportional to the simulated net leakage electron current (averaged over the PA thickness). Experimental response was linear with dose and independent of dose rate. Detector responses to different x-ray sources show higher signals for kVp photon energies, as expected, though a strong signal was obtained for MV energies as well. The signal scaled with total effective area inside the multielemental structures; for example, the yield of a multielement sensor made with 20 Ta layers compared to a single-element structure with 1 Ta layer of the same total thickness of Ta was 10 times greater for 6 MV beam and 23 times greater for 120 kVp. Beam attenuation per element in the detector was 0.5%, 1%, 3%, and 46%, respectively for 6 MV, 6 MV FFF, 2.5 MV, and 120 kVp. CONCLUSION: We demonstrated the feasibility of aerogel-based multilayer HEC radiation detector and its application for flux/dose monitoring of kVp and radiotherapy MV beams with small beam attenuation.