Computer-Assisted Differential Diagnosis of Pyoderma Gangrenosum and Venous Ulcers with Deep Neural Networks.

(1) Background: Pyoderma gangrenosum (PG) is often situated on the lower legs, and the differentiation from conventional leg ulcers (LU) is a challenging task due to the lack of clear clinical diagnostic criteria. Because of the different therapy concepts, misdiagnosis or delayed diagnosis bears a great risk for patients. (2) Objective: to develop a deep convolutional neural network (CNN) capable of analysing wound photographs to facilitate the PG diagnosis for health professionals. (3) Methods: A CNN was trained with 422 expert-selected pictures of PG and LU. In a man vs. machine contest, 33 pictures of PG and 36 pictures of LU were presented for diagnosis to 18 dermatologists at two maximum care hospitals and to the CNN. The results were statistically evaluated in terms of sensitivity, specificity and accuracy for the CNN and for dermatologists with different experience levels. (4) Results: The CNN achieved a sensitivity of 97% (95% confidence interval (CI) 84.2−99.9%) and outperformed dermatologists, with a sensitivity of 72.7% (CI 54.4−86.7%) significantly (p < 0.03). However, dermatologists achieved a slightly higher specificity (88.9% vs. 83.3%). (5) Conclusions: For the first time, a deep neural network was demonstrated to be capable of diagnosing PG, solely on the basis of photographs, and with a greater sensitivity compared to that of dermatologists.

Evaluation of a 2D detector array for patient-specific VMAT QA with different setups.

For pre-treatment plan verification of advanced treatment techniques such as intensity-modulated arc therapy, a fast and reliable dosimetric device is required. In this study, we investigated the suitability of MatriXX in different setups for verification of volumetric modulated arc therapy (VMAT) plans. If MatriXX is used in a stationary phantom (MULTICube), the measured dose is dependent on the beam angle. For the first setup (MatriXX/MULTICube), we developed correction factors (CFs) for each detector element (1020 CFs). We investigated the accuracy of these CFs by verifying 12 VMAT plans. In the second setup, we also assessed the suitability of MatriXX in a dedicated holder. Using this setup (MatriXX/Holder), 30 additional VMAT plans were verified. Deviations of up to ∼17% and ∼11% were noted for one of the ion chambers at 90° and 180° gantry positions. The influence of the beam angle dependence (MULTICube) can explicitly be seen when a gamma criterion of 2%/2 mm was chosen. An overall improvement of 4.3% of passing pixels (pp) was noted after applying beam angular-dependent CFs. When the gamma criterion was 3%/3 mm, the %pp was ≥ 95% without and ∼100% with correction. With the second setup, MatriXX/holder, we showed excellent agreement between measurements and calculations. The %pp averaged over all plans (30 VMAT treatment plans) was nearly ∼100%. The combination of MatriXX with MULTICube or with holder proved to be a fast and reliable method for pretreatment verification of arc therapy with sufficient accuracy.

Experimental validation of a commercial 3D dose verification system for intensity-modulated arc therapies.

We validate the dosimetric performance of COMPASS®, a novel 3D quality assurance system for verification of volumetric-modulated arc therapy (VMAT) treatment plans that can correlate the delivered dose to the patient's anatomy, taking into account the tissue inhomogeneity. The accuracy of treatment delivery was assessed by the COMPASS® for 12 VMAT plans, and the resulting assessments were evaluated using an ionization chamber and film measurements. Dose-volume relationships were evaluated by the COMPASS® for three additional treatment plans and these were used to verify the accuracy of treatment planning dose calculations. The results matched well between COMPASS® and measurements for the ionization chamber (≤3%) and film (73-99% for gamma((3%/3 mm)) < 1 and 98-100% for gamma((5%/5 mm)) < 1) for the phantom plans. Differences in dose-volume statistics for the average dose to the PTV were within 2.5% for three treatment plans. For the structures located in the low-dose region, a maximum difference of <9% was observed. In its current implementation, the system could measure the delivered dose with sufficient accuracy and could project the 3D dose distribution directly on the patient's anatomy. Slight deviations were found for large open fields. These could be minimized by improving the COMPASS® in-built beam model.

Model-independent, multimodality deformable image registration by local matching of anatomical features and minimization of elastic energy.

With respect to the demands of adaptive and 4D-radiotherapy applications, an algorithm is proposed for a fully automatic, multimodality deformable registration that follows the concept of translational relocation of regularly distributed image subvolumes governed by local anatomical features. Thereby, the problem of global deformable registration is broken down to multiple independent local registration steps which allows for straightforward parallelization of the algorithm. In a subsequent step, possible local misregistrations are corrected for by minimization of the elastic energy of the displacement field under consideration of image information. The final displacement field results from interpolation of the subvolume shift vectors. The algorithm can employ as a similarity measure both the correlation coefficient and mutual information. The latter allows the application to intermodality deformable registration problems. The typical calculation time on a modern multiprocessor PC is well below 1 min, which facilitates almost-interactive, "online" usage. CT-to-MRI and CT-to-cone-beam-CT registrations of head-and-neck data sets are presented, as well as inhale-to-exhale registrations of lung CT data sets. For quantitative evaluation of registration accuracy, a virtual thorax phantom was developed; additionally, a landmark-based evaluation on four lung respiratory-correlated CT data sets was performed. This consistently resulted in average registration residuals on the order of the voxel size or less (3D-residuals approximately 1-2 mm). Summarizing, the presented algorithm allows an accurate multimodality deformable registration with calculation times well below 1 min, and thus bears promise as a versatile basic tool in adaptive and 4D-radiotherapy applications.

Analysis of the rigid and deformable component of setup inaccuracies on portal images in head and neck radiotherapy.

The issue of setup errors consisting of translation, rotation and deformation components in head and neck radiotherapy is addressed with a piecewise registration of small independent regions on a portal image to their reference position. These rectangular regions are termed featurelets as they contain relevant anatomical features. The resulting displacement vectors of each featurelet reflect both the center-of-mass (COM), i.e. the rigid, and the non-rigid component of the setup error. The displacement vectors of a series of daily portal images were subjected to a principal component analysis. In addition to the mean, systematic displacement of each featurelet, this analysis yields correlated patterns of anatomical deformations. Hence, the physiological movements of an individual patient can be obtained without a biomechanical model. It is shown that in the presence of setup errors that are due to rotations or deformations a correction by the COM displacement may deteriorate the error of parts of the anatomy further. The featurelet analysis can be used to refine setup correction protocols, tune spatially variable setup margins in treatment planning and optimize patient immobilization devices.

Robust treatment planning for intensity modulated radiotherapy of prostate cancer based on coverage probabilities.

BACKGROUND AND PURPOSE: To evaluate an optimization approach where coverage probabilities are incorporated into the optimization of intensity modulated radiotherapy (IMRT) to overcome the problem of margin definition in the case of overlapping planning target volume and organs at risk. PATIENTS AND METHODS: IMRT plans were generated for three optimization approaches: based on a planning CT plus margin (A), on prostate and rectum contours from five pre-treatment CT plus margin (B), and on coverage probabilities (C). For approach (C), the probability of organ occupation was computed for each voxel from five pre-treatment CTs and the population distribution of systematic setup error and it was used as local weight in the costfunctions. Monte Carlo simulations of treatment courses were used to compute the probability distribution of prostate and rectal wall equivalent uniform dose (EUD). RESULTS: Treatment simulations showed best and most robust results for prostate and rectal wall EUD within the population for (C). For (A) the rectal wall EUD was on average about 1.5 Gy greater than in (C), while the prostate EUD was lower than those from (C) for most of the patients for (B) (especially for those with great organ motion). CONCLUSIONS: The incorporation of coverage probabilities as local weights allows for dose escalation as well as improved rectal sparing and results in a safer and more robust IMRT treatment.

Intensity modulated radiotherapy for high risk prostate cancer based on sentinel node SPECT imaging for target volume definition.

BACKGROUND: The RTOG 94-13 trial has provided evidence that patients with high risk prostate cancer benefit from an additional radiotherapy to the pelvic nodes combined with concomitant hormonal ablation. Since lymphatic drainage of the prostate is highly variable, the optimal target volume definition for the pelvic lymph nodes is problematic. To overcome this limitation, we tested the feasibility of an intensity modulated radiation therapy (IMRT) protocol, taking under consideration the individual pelvic sentinel node drainage pattern by SPECT functional imaging. METHODS: Patients with high risk prostate cancer were included. Sentinel nodes (SN) were localised 1.5-3 hours after injection of 250 MBq 99mTc-Nanocoll using a double-headed gamma camera with an integrated X-Ray device. All sentinel node localisations were included into the pelvic clinical target volume (CTV). Dose prescriptions were 50.4 Gy (5 x 1.8 Gy / week) to the pelvis and 70.0 Gy (5 x 2.0 Gy / week) to the prostate including the base of seminal vesicles or whole seminal vesicles. Patients were treated with IMRT. Furthermore a theoretical comparison between IMRT and a three-dimensional conformal technique was performed. RESULTS: Since 08/2003 6 patients were treated with this protocol. All patients had detectable sentinel lymph nodes (total 29). 4 of 6 patients showed sentinel node localisations (total 10), that would not have been treated adequately with CT-based planning ('geographical miss') only. The most common localisation for a probable geographical miss was the perirectal area. The comparison between dose-volume-histograms of IMRT- and conventional CT-planning demonstrated clear superiority of IMRT when all sentinel lymph nodes were included. IMRT allowed a significantly better sparing of normal tissue and reduced volumes of small bowel, large bowel and rectum irradiated with critical doses. No gastrointestinal or genitourinary acute toxicity Grade 3 or 4 (RTOG) occurred. CONCLUSION: IMRT based on sentinel lymph node identification is feasible and reduces the probability of a geographical miss. Furthermore, IMRT allows a pronounced sparing of normal tissue irradiation. Thus, the chosen approach will help to increase the curative potential of radiotherapy in high risk prostate cancer patients.

Dosimetric consequences of the application of off-line setup error correction protocols and a hull-volume definition strategy for intensity modulated radiotherapy of prostate cancer.

PURPOSE: To evaluate the consequences of a planning volume definition based on multiple CTs and the application of off-line setup error correction for the treatment of prostate cancer with intensity-modulated radiotherapy (IMRT). Further, to compare various setup correction protocols (SCP) by their influence on the average dose distributions. MATERIALS AND METHODS: A planning target volume (PTV) consisting of the bounding volume of prostate contours of five CTs (CTV_hull) plus an additional margin of 5mm and a virtual Rectum_hull volume (the solid bounding volume of the five corresponding rectum contours) are used for treatment planning. Simulations of treatment courses with the non-parametric bootstrap method allow to estimate the distribution of the expected equivalent uniform dose (EUD). The impact of off-line setup error correction protocols is evaluated based on estimated EUD distributions. RESULTS: Off-line SCP allow to achieve the intended prostate and rectum EUD and a reliable coverage of the CTV despite the reduced margins. The EUD of the virtual hull volumes is a good estimate for the EUD of prostate and rectal wall. CONCLUSION: Treatment planning based on Rectum_hull and CTV_hull plus setup margin as PTV in combination with SCP results in a robust and safe IMRT planning concept.