Inter-institutional variability of knowledge-based plan prediction of left whole breast irradiation.

PURPOSE: Within a multi-institutional project, we aimed to assess the transferability of knowledge-based (KB) plan prediction models in the case of whole breast irradiation (WBI) for left-side breast irradiation with tangential fields (TF). METHODS: Eight institutions set KB models, following previously shared common criteria. Plan prediction performance was tested on 16 new patients (2 pts per centre) extracting dose-volume-histogram (DVH) prediction bands of heart, ipsilateral lung, contralateral lung and breast. The inter-institutional variability was quantified by the standard deviations (SDint) of predicted DVHs and mean-dose (Dmean). The transferability of models, for the heart and the ipsilateral lung, was evaluated by the range of geometric Principal Component (PC1) applicability of a model to test patients of the other 7 institutions. RESULTS: SDint of the DVH was 1.8 % and 1.6 % for the ipsilateral lung and the heart, respectively (20 %-80 % dose range); concerning Dmean, SDint was 0.9 Gy and 0.6 Gy for the ipsilateral lung and the heart, respectively (<0.2 Gy for contralateral organs). Mean predicted doses ranged between 4.3 and 5.9 Gy for the ipsilateral lung and 1.1-2.3 Gy for the heart. PC1 analysis suggested no relevant differences among models, except for one centre showing a systematic larger sparing of the heart, concomitant to a worse PTV coverage, due to high priority in sparing the left anterior descending coronary artery. CONCLUSIONS: Results showed high transferability among models and low inter-institutional variability of 2% for plan prediction. These findings encourage the building of benchmark models in the case of TF-WBI.

Empirical methods for the validation of time-to-event mathematical models taking into account uncertainty and variability: application to EGFR + lung adenocarcinoma.

BACKGROUND: Over the past several decades, metrics have been defined to assess the quality of various types of models and to compare their performance depending on their capacity to explain the variance found in real-life data. However, available validation methods are mostly designed for statistical regressions rather than for mechanistic models. To our knowledge, in the latter case, there are no consensus standards, for instance for the validation of predictions against real-world data given the variability and uncertainty of the data. In this work, we focus on the prediction of time-to-event curves using as an application example a mechanistic model of non-small cell lung cancer. We designed four empirical methods to assess both model performance and reliability of predictions: two methods based on bootstrapped versions of parametric statistical tests: log-rank and combined weighted log-ranks (MaxCombo); and two methods based on bootstrapped prediction intervals, referred to here as raw coverage and the juncture metric. We also introduced the notion of observation time uncertainty to take into consideration the real life delay between the moment when an event happens, and the moment when it is observed and reported. RESULTS: We highlight the advantages and disadvantages of these methods according to their application context. We have shown that the context of use of the model has an impact on the model validation process. Thanks to the use of several validation metrics we have highlighted the limit of the model to predict the evolution of the disease in the whole population of mutations at the same time, and that it was more efficient with specific predictions in the target mutation populations. The choice and use of a single metric could have led to an erroneous validation of the model and its context of use. CONCLUSIONS: With this work, we stress the importance of making judicious choices for a metric, and how using a combination of metrics could be more relevant, with the objective of validating a given model and its predictions within a specific context of use. We also show how the reliability of the results depends both on the metric and on the statistical comparisons, and that the conditions of application and the type of available information need to be taken into account to choose the best validation strategy.

Multi-view graph representation with similarity diffusion for general zero-shot learning.

Zero-shot learning (ZSL) aims to predict unseen classes without using samples of these classes in model training. The ZSL has been widely used in many knowledge-based models and applications to predict various parameters, including categories, subjects, and anomalies, in different domains. Nonetheless, most existing ZSL methods require the pre-defined semantics or attributes of particular data environments. Therefore, these methods are difficult to be applied to general data environments, such as ImageNet and other real-world datasets and applications. Recent research has tried to use open knowledge to enhance the ZSL methods to adapt it to an open data environment. However, the performance of these methods is relatively low, namely the accuracy is normally below 10%, which is due to the inadequate semantics that can be used from open knowledge. Moreover, the latest methods suffer from a significant "semantic gap" problem between the generated features of unseen classes and the real features of seen classes. To this end, this paper proposes a multi-view graph representation with a similarity diffusion model, applying the ZSL tasks to general data environments. This model applies a multi-view graph to enhance the semantics fully and proposes an innovative diffusion method to augment the graph representation. In addition, a feature diffusion method is proposed to augment the multi-view graph representation and bridge the semantic gap to realize zero-shot predicting. The results of numerous experiments in general data environments and on benchmark datasets show that the proposed method can achieve new state-of-the-art results in the field of general zero-shot learning. Furthermore, seven ablation studies analyze the effects of the settings and different modules of the proposed method on its performance in detail and prove the effectiveness of each module.

Knowledge-based multi-institution plan prediction of whole breast irradiation with tangential fields.

PURPOSE: To quantify inter-institute variability of Knowledge-Based (KB) models for right breast cancer patients treated with tangential fields whole breast irradiation (WBI). MATERIALS AND METHODS: Ten institutions set KB models by using RapidPlan (Varian Inc.), following previously shared methodologies. Models were tested on 20 new patients from the same institutes, exporting DVH predictions of heart, ipsilateral lung, contralateral lung, and contralateral breast. Inter-institute variability was quantified by the inter-institute SDint of predicted DVHs/Dmean. Association between lung sparing vs PTV coverage strategy was also investigated. The transferability of models was evaluated by the overlap of each model's geometric Principal Component (PC1) when applied to the test patients of the other 9 institutes. RESULTS: The overall inter-institute variability of DVH/Dmean ipsilateral lung dose prediction, was less than 2% (20%-80% dose range) and 0.55 Gy respectively (1SD) for a 40 Gy in 15 fraction schedule; it was < 0.2 Gy for other OARs. Institute 6 showed the lowest mean dose prediction value and no overlap between PTV and ipsilateral lung. Once excluded, the predicted ipsilateral lung Dmean was correlated with median PTV D99% (R2 = 0.78). PC1 values were always within the range of applicability (90th percentile) for 7 models: for 2 models they were outside in 1/18 cases. For the model of institute 6, it failed in 7/18 cases. The impact of inter-institute variability of dose calculation was tested and found to be almost negligible. CONCLUSIONS: Results show limited inter-institute variability of plan prediction models translating in high inter-institute interchangeability, except for one of ten institutes. These results encourage future investigations in generating benchmarks for plan prediction incorporating inter-institute variability.

Study on the feasibility of quality improvement for automatic plans based on rapid plan model.

Objective: To study whether an interactive improved internal feedback system with the model can be established, we compared the plans generated by two automatic planning models. Methods: Seventy cases of pelvic patients were selected. Intensity-modulated radiation therapy (IMRT) plans (P0) generated by the clinical model (M0) were imported into the Rapid plan model to establish a dose-volume histogram. The predicted model through automatic planning model in clinical, and the new rapid plan model (M1) was generated by training and structure matching settings. The 70 new IMRT plans (P1) were generated by M1, and the new rapid plan model (M2) was trained by P1. In this same method, 70 IMRT plans (P2) were generated by M2. Dosimetric differences between P1 and P2 were then compared and analyzed. Results: For the model parameters, R2 and X2 in P2 were higher than those in P1, and the CD values of the bladder, right femoral head, and rectum in P1 were higher than those of corresponding organs in P2. The studentized residual (SR) value of the bladder and SR and difference of estimate values of the left femoral head and right femoral head in P1 were lower than P2. In planning, (D2, D98, and HI) P1 were better than P2 (P < 0.01); the bladder V10 and left femoral head V40 in P2 were lower than in P1 by 0.08% and 0.15%, respectively (P < 0.05); others in P2 were higher than those in P1 (P < 0.05) except the bladder V20, Dmean, rectum V10, V20, V30, right femoral head V10, and V40; and the MUs of P2 was lower than that of P1 for 132.2 (P < 0.05). Conclusion: The stability of M2 is stronger than that of M1. Therefore, the interactive improved internal feedback system within the model of "plan-model-plan-model" is feasible and meaningful.

Protein structure prediction: making AWSEM AWSEM-ER by adding evolutionary restraints.

Protein sequences have evolved to fold into functional structures, resulting in families of diverse protein sequences that all share the same overall fold. One can harness protein family sequence data to infer likely contacts between pairs of residues. In the current study, we combine this kind of inference from coevolutionary information with a coarse-grained protein force field ordinarily used with single sequence input, the Associative memory, Water mediated, Structure and Energy Model (AWSEM), to achieve improved structure prediction. The resulting Associative memory, Water mediated, Structure and Energy Model with Evolutionary Restraints (AWSEM-ER) yields a significant improvement in the quality of protein structure prediction over the single sequence prediction from AWSEM when a sufficiently large number of homologous sequences are available. Free energy landscape analysis shows that the addition of the evolutionary term shifts the free energy minimum to more native-like structures, which explains the improvement in the quality of structures when performing predictions using simulated annealing. Simulations using AWSEM without coevolutionary information have proved useful in elucidating not only protein folding behavior, but also mechanisms of protein function. The success of AWSEM-ER in de novo structure prediction suggests that the enhanced model opens the door to functional studies of proteins even when no experimentally solved structures are available.

Bioenergy and biodiversity: Intensified biomass extraction from hedges impairs habitat conditions for birds.

Biomass is increasingly used as an alternative source for energy in Europe. Woody material cut from hedges is considered to provide a suitable complement to maize and oilseed rape, which are currently the dominant biomass sources. Since shrubs and trees are also important habitats for birds, however, coppicing of hedges at the landscape scale may adversely affect the diversity of the avifauna. To evaluate this risk, we estimated the response of hedge birds to three management scenarios differing in cutting intensity and hedge selection. The analysis was done using hedge data of the Lautertal municipality (n = 339 hedges; Vogelsberg area, Hesse, Germany). It focused on 25 bird species, which are all listed in the hedge programme of the German Ornithological Stations. Information on the preferences of these birds for certain hedge features such as height or width was gathered by an extensive literature review. A cluster analysis on the consolidated literature data allowed us to identify three groups of birds according to their preference for certain hedge attributes. Two groups, which included Yellowhammer (Emberiza citrinella L.) (i) and Blackbird (Turdus merula L.) (ii), favoured trees located in hedges, but differed in their preference for hedge shape, with (i) being attracted by long and broad hedges and (ii) by high hedges. The third group, which included the Whitethroat (Sylvia communis L.), preferred small hedges with gaps and medium vegetation density. Spatially explicit suitability models based on these data allowed us to predict the status quo of hedge suitability for these species groups. Field surveys proved the accuracy of the predictions to be sufficient, since the hedge suitability predicted was significantly and positively correlated to the occurrence of 9 out of the 12 testable focal species. Our models predicted biomass extraction to almost always reduce hedge suitability for the three bird groups. Concerning the Yellowhammer and the Blackbird group, a high level of biomass extraction reduced hedge suitability by approximately 20%. We thus conclude that intensively extracting biomass can significantly reduce hedge suitability for birds, despite considerable differences in habitat requirements. Considering the variable response of the bird groups to our scenarios as well as the variation in habitat occupancy by birds, however, cutting woody material from hedges nevertheless provides an option to reduce adverse effects of bioenergy production on biodiversity at the landscape scale, as long as hedge management is based on the best knowledge available.