Threshold-based Algorithm Research Articles

Sports Video Classification Framework Using Enhanced Threshold Based Keyframe Selection Algorithm and Customized CNN on UCF101 and Sports1-M Dataset.

The computer vision community has taken a keen interest in recent developments in activity recognition and classification in sports videos. Advancements in sports have a broadened the technical interest of the computer vision community to perform various types of research. Images and videos are the most frequently used components in computer vision. There are numerous models and methods that can be used to classify videos. At the same time, there no specific framework or model for classifying and identifying sports videos. Hence, we proposed a framework based on deep learning to classify sports videos with their appropriate class label. The framework is to perform sports video classification using two different benchmark datasets, UCF101 and the Sports1-M dataset. The objective of the framework is to help sports players and trainers to identify specific sports from the large data source, then analyze and perform well in the future. This framework takes sports video as an input and produces the class label as an output. In between, the framework has numerous intermediary processes. Preprocessing is the first step in the proposed framework, which includes frame extraction and noise reduction. Keyframe selection is carried out by candidate frame extraction and an enhanced threshold-based frame difference algorithm, which is the second step. The final step of the sports video classification framework is feature extraction and classification using CNN. The proposed framework result is compared with pretrained neural networks such as AlexNet and GoogleNet, and then the results are also compared. Three different evaluation metrics are used to measure the accuracy and performance of the framework.

The Online Knapsack Problem with Departures

The online knapsack problem is a classic online resource allocation problem in networking and operations research. Its basic version studies how to pack online arriving items of different sizes and values into a capacity-limited knapsack. In this paper, we study a general version that includes item departures, while also considering multiple knapsacks and multi-dimensional item sizes. We design a threshold-based online algorithm and prove that the algorithm can achieve order-optimal competitive ratios. Beyond worst-case performance guarantees, we also aim to achieve near-optimal average performance under typical instances. Towards this goal, we propose a data-driven online algorithm that learns within a policy-class that guarantees a worst-case performance bound. In trace-driven experiments, we show that our data-driven algorithm outperforms other benchmark algorithms in an application of online knapsack to job scheduling for cloud computing.

Throughput Optimization in Adaptive Transmit Antenna Selection Systems With Limited Feedback

We investigate the performance of a downlink wireless system with a multiantenna base station under the effect of limited channel feedback. In order to inactivate transmit antennas with poor channel quality and, thus, reduce the feedback overhead, a channel gain threshold-based multiantenna selection (CGT-MAS) scheme is proposed without optimization (NOCGT-MAS) and the corresponding single-antenna scheme (NOCGT-SAS) is considered as a baseline. Considering that the number of selected antennas affects both data transmission rate and the feedback overhead, an effective throughput defined as the difference between the data transmission rate and feedback rate is employed for evaluating the system performance. We conduct an effective throughput analysis for both the NOCGT-SAS and NOCGT-MAS schemes and show that an optimal channel gain threshold exists to maximize the effective throughput of the proposed NOCGT-MAS scheme. As a consequence, we propose the optimal instantaneous CGT-MAS algorithm (OICGT-MAS) to maximize the instantaneous effective throughput. Besides, we derive a closed-form expression of an average effective throughput for the CGT-MAS scheme and propose an optimal average channel gain threshold-based multiantenna selection algorithm (OACGT-MAS) for maximizing the derived average effective throughput. Numerical results demonstrate that the effective throughput of the proposed NOCGT-MAS is substantially better than that of conventional NOCGT-SAS. Moreover, although the OICGT-MAS scheme has a slight advantage over OACGT-MAS in terms of the effective throughput, it has a much higher computational complexity than the latter method.

A fast and reliable method for semi-automated planimetric quantification of dental plaque in clinical trials.

To develop a simple and reproducible method for semi-automated planimetric quantification of dental plaque. Plaque from 20 healthy volunteers was disclosed using erythrosine, and fluorescence images of the first incisors, first premolars, and first molars were recorded after 1, 7, and 14 days of de novo plaque formation. The planimetric plaque index (PPI) was determined using a semi-automated threshold-based image segmentation algorithm and compared with manually determined PPI and the Turesky modification of the Quigley-Hein plaque index (TM-QHPI). The decrease of tooth autofluorescence in plaque-covered areas was quantified as an index of plaque thickness (TI). Data were analysed by analysis of variance (ANOVA) and Pearson correlations. The high contrast between teeth, disclosed plaque, and soft tissues in fluorescence images allowed for a fast threshold-based image segmentation. Semi-automated PPI is strongly correlated with manual planimetry (r=0.92; p < .001) and TM-QHPI recordings (r=0.88; p < .001), and may exhibit a higher discriminatory power than TM-QHPI due to its continuous scale. TI values corresponded to optically perceived plaque thickness, and no differences were observed over time (p > .05, ANOVA). The proposed semi-automated planimetric analysis based on fluorescence images is a simple and efficient method for dental plaque quantification in multiple images with reduced human input.

Review and Evaluation of Eye Movement Event Detection Algorithms.

Eye tracking is a technology aimed at understanding the direction of the human gaze. Event detection is a process of detecting and classifying eye movements that are divided into several types. Nowadays, event detection is almost exclusively done by applying a detection algorithm to the raw recorded eye-tracking data. However, due to the lack of a standard procedure for how to perform evaluations, evaluating and comparing various detection algorithms in eye-tracking signals is very challenging. In this paper, we used data from a high-speed eye-tracker SMI HiSpeed 1250 system and compared event detection performance. The evaluation focused on fixations, saccades and post-saccadic oscillation classification. It used sample-by-sample comparisons to compare the algorithms and inter-agreement between algorithms and human coders. The impact of varying threshold values on threshold-based algorithms was examined and the optimum threshold values were determined. This evaluation differed from previous evaluations by using the same dataset to evaluate the event detection algorithms and human coders. We evaluated and compared the different algorithms from threshold-based, machine learning-based and deep learning event detection algorithms. The evaluation results show that all methods perform well for fixation and saccade detection; however, there are substantial differences in classification results. Generally, CNN (Convolutional Neural Network) and RF (Random Forest) algorithms outperform threshold-based methods.

An adaptive threshold-based quantum image segmentation algorithm and its simulation

An adaptive threshold-based quantum image segmentation algorithm and its simulation

Automated bow shock and magnetopause boundary detection with Cassini using threshold and deep learning methods

Two algorithms set for automatic detection of bow shock (BS) and magnetopause (MP) boundaries at Saturn using in situ magnetic field and plasma data acquired by the Cassini spacecraft are presented. Traditional threshold-based and modern deep learning algorithms were investigated for the task of boundary detection. Sections of Cassini’s orbits were pre-selected based on empirical BS and MP boundary models, and from outlier detection in magnetic field data using an autoencoder neural network. The threshold method was applied to pre-selected magnetic field and plasma data independently to compute parameters to which a threshold was applied to determine the presence of a boundary. The deep learning method used a type of convolutional neural network (CNN) called ResNet on images of magnetic field time series data and electron energy-time spectrograms to classify the presence of boundaries. 2012 data were held out of the training data to test and compare the algorithms on unseen data. The comparison showed that the CNN method applied to plasma data outperformed the threshold method. A final multiclass CNN classifier trained on plasma data obtained F1 scores of 92.1% ± 1.4% for BS crossings and 84.7% ± 1.9% for MP crossings on a corrected test dataset (from use of a bootstrap method). Reliable automated detection of boundary crossings could enable future spacecraft experiments like the PEP instrument on the upcoming JUICE spacecraft mission to dynamically adapt the best observing mode based on rapid classification of the boundary crossings as soon as it appears, thus yielding higher quality data and improved potential for scientific discovery.

Event-Based Adaptive Neural Control of Nonlinear Systems With Deferred Constraint

In this article, the problem of constant yet deferred output constraint for uncertain strict-feedback nonlinear systems is studied. By “deferred output constraint,” we mean that the system output is free/released from any constraint in the initial interval and then preserves within a bounded region right after a finite time. Due to such a form of output constraint, the normally employed Barrier Lyapunov Function (BLF)-based results become invalid because the corresponding BLF is undefined in the initial period. The problem will be rather complicated yet challenging if computation and communication constraints are taken into account. By developing an error-based nonlinear function and constructing a prescribed-time scaling function, together with the approximate ability of neural networks, a varying threshold-based event-triggering adaptive neural control algorithm is presented such that not only the deferred output constraint can be ensured and the network resources can be saved but also the tracking error is able to converge to a pregiven region in a prescribed time. Simulations are provided to demonstrate the effectiveness of the proposed control.

A Comparison among Different Strategies to Detect Potential Unstable Behaviors in Postural Sway.

Assistive Technology helps to assess the daily living and safety of frail people, with particular regards to the detection and prevention of falls. In this paper, a comparison is provided among different strategies to analyze postural sway, with the aim of detecting unstable postural status in standing condition as precursors of potential falls. Three approaches are considered: (i) a time-based features threshold algorithm, (ii) a time-based features Neuro-Fuzzy inference system, and (iii) a Neuro-Fuzzy inference fed by Discrete-Wavelet-Transform-based features. The analysis was performed across a wide dataset and exploited performance indexes aimed at assessing the accuracy and the reliability of predictions provided by the above-mentioned strategies. The results obtained demonstrate valuable performances of the three considered strategies in correctly distinguishing among stable and unstable postural status. However, the analysis of robustness against noisy data highlights better performance of Neuro-Fuzzy inference systems with respect to the threshold-based algorithm.

Optimal Weighted Logarithmic Transformation Converted HMOG Features for Automatic Smart Phone Authentication

This paper intends to develop an automatic behavior-based smart phone authentication model by three major phases: feature extraction, weighted logarithmic transformation, and classification. Initially, from the data related to the touches/gesture of the smartphone user, hand movement, orientation, and grasp (HMOG), features are extracted with the aid of grasp resistance and grasp stability. These extracted features are mapped within the particular range by normalizing HMOG. These normalized data are multiplied with the weights followed by logarithmic transformation in the weighted logarithmic transformation phase. As a novelty, the decision-making process related to the logarithmic and weight selection is based on the improved optimization algorithm, called modified threshold-based whale optimization algorithm (MT-WOA). The final feature vectors are fed to DBN for recognizing the authorized users. Finally, a performance-based evaluation is performed between the MT-WOA+DBN and the existing models in terms of various relevant performance measures.

A comprehensive comparison of accuracy and practicality of different types of algorithms for pre-impact fall detection using both young and old adults

This study aims to comprehensively compare the accuracy and practicality of three different types of algorithms for pre-impact fall detection using both young and old subjects. Threshold-based, conventional machine learning (SVM) and deep learning (ConvLSTM) algorithms were compared. Results showed that ConvLSTM had an accuracy of 99.16 % (sensitivity: 99.32 %, specificity: 99.01 %) and an averaged lead time of 403 ms on young subjects, which outperformed SVM (97.16 %, 385 ms) and much superior to the threshold-based algorithm (89.06 %, 333 ms). In addition, latency tests on an embedded device showed that the Lite model of ConvLSTM had a low latency of 2.1 ms, which was comparable to the threshold-based algorithm (<1 ms) but much lower than SVM (86.9 ms). The feasibility and effectiveness of applying algorithms trained on young subjects to old subjects were also validated. These findings suggested that ConvLSTM has great potential for detecting pre-impact falls and preventing fall-related injuries.

Automated detection and removal of flat line segments and large amplitude fluctuations in neonatal electroencephalography.

BackgroundArtefact removal in neonatal electroencephalography (EEG) by visual inspection generally depends on the expertise of the operator, is time consuming and is not a consistent pre-processing step to the pipeline for the automated EEG analysis. Therefore, there is the need for the automated detection and removal of artefacts in neonatal EEG, especially of distinct and predominant artefacts such as flat line segments (mainly caused by instrumental error where contact between electrodes and head box is lost) and large amplitude fluctuations (related to neonatal movements).MethodA threshold-based algorithm for the automated detection and removal of flat line segments and large amplitude fluctuations in neonatal EEG of infants at term-equivalent age is developed. The algorithm applies thresholds to the absolute second difference, absolute amplitude, absolute first difference and the ratio between the frequency content above 50 Hz and the frequency content across all frequencies.ResultsThe algorithm reaches a median accuracy of 0.91, a median hit rate of 0.91 and a median false discovery rate of 0.37. Also, a significant improvement (≈10%) in the performance of a four-stage sleep classifier is observed after artefact removal with the proposed algorithm as compared to before its application.SignificanceAn automated artefact removal method contributes to the pipeline of automated EEG analysis. The proposed algorithm has shown to have good performance and to be effective in neonatal EEG applications.

Nonlinear Control Design for Regenerative and Hybrid Antilock Braking in Electric Vehicles

The use of regenerative braking and recent electrohydraulic or electromechanical brake-by-wire systems with enhanced dynamics enables improved antilock braking control performance in electric vehicles. However, the possible benefit is often limited by suboptimal control architectures with separate distributed controllers for electric motor (EM) and friction brake torque requests. Furthermore, oversimplifications in synthesis models for brake control design constrain the achievable performance. Instead, to date, threshold-based algorithms are widely used in modern production vehicles despite having been designed for slower hydraulic brake systems originally. To exploit the full potential, this work proposes a continuous nonlinear control design using input–output linearization for robust wheel speed tracking control. The design uses a unified controller for the redundant actuators. In addition, it explicitly considers drivetrain oscillations induced by regenerative braking using onboard EMs in the synthesis model. The stability for the resulting zero dynamics is shown through the Lyapunov analysis. The nonlinear controller is augmented by a subsequent control allocation for splitting the control effort on the redundant actuators. The allocation design ensures consistent control performance for both regenerative braking and hybrid braking while simultaneously aiming for high-energy recovery. The overall antilock braking control design is implemented in an electric test vehicle. Its tracking performance, disturbance attenuation, and robustness are experimentally validated through various emergency braking maneuvers on different road surfaces.

A wheel slip control scheme for aeronautical braking applications based on neural network estimation

Due to safety reasons, anti-skid braking control strategies in aircraft can rely only on sensors that are integral to the landing gear, and these currently are limited to wheel rotational speed and braking pressure. Thus, up to now anti-lock-braking systems (ABS) are developed as threshold-based algorithms that regulate the wheel deceleration. However, as it is known from the Automotive field, slip control offers superior closed-loop properties and robustness, paired with a much easier and time-saving tuning phase. This work aims at investigating first of all if slip control does indeed offer the same performance advancement also in the aeronautical braking context and, secondly, whether the wheel slip can be effectively estimated and employed in a closed-loop braking controller without the need of additional sensors on the landing gear. To do this, we propose a wheel slip control scheme where a data-driven approach using a neural network architecture is used to solve the problem of wheel slip estimation. The proposed control scheme is tested within a very realistic simulation setting, capturing all the non-linear effects peculiar of the aeronautical context, and compared with a standard deceleration-based ABS in multiple braking maneuvers carried out within a large operational envelope. The results show that superior performance and robustness can indeed be achieved with the proposed control approach, paving the way for the adoption of wheel slip abs strategies also for aircraft braking.

Design and Optimization of a Decision Support System for Sports Training Based on Data Mining Technology

This paper presents an in-depth study and analysis of sports training decision support systems through data mining techniques, and designs a corresponding sports training decision support system for application in practical situations. It investigates the application of data mining techniques in solving the problem of optimal analysis of sports training indicators. The Apriori algorithm is used to find out the hidden correlations behind the data, while the correlation study applicable to college students’ physical fitness indicators is proposed and practiced, and the potential hidden relationships between the indicator data are found. The analysis results are further verified by the improved AP algorithm, FP algorithm, which provides important opinions for college physical education curriculum reform and youth physical culture development. Then, the Shi-Tomasi algorithm is used for feature point tracking and the support vector machine approach for human motion posture recognition. Considering that the collected action data contain a large amount of stationary action data, a threshold-based action segmentation algorithm is designed to extract useful action segments; after extracting the action segments, common action features are extracted from both statistical features, such as mean, variance, skewness, kurtosis, and physical features, such as acceleration and plantar pressure, to give a specific description of human daily actions. Thus, the fatigue of different postures is judged according to their number of occurrences and a preliminary assessment result is obtained. Then, the surface EMG data collected by the sensors are pre-processed to reduce noise, and the surface EMG signals are fused at the feature level based on the wavelet change algorithm, and the feature vectors are trained with BP neural networks to evaluate the fatigue of human muscle movements and obtain a preliminary evaluation result. Finally, by studying the application of information fusion technology at the decision level, the assessment results of the comprehensive human movement analysis are obtained using D-S evidence theory. In addition, a web platform is built on this basis to display and manage human exercise data and assessment results, and to facilitate the storage and query of historical exercise records.

Nanoflare distributions over solar cycle 24 based on SDO/AIA differential emission measure observations

Aims. Nanoflares in quiet-Sun regions during solar cycle 24 are studied with the best available plasma diagnostics to derive their energy distribution and contribution to coronal heating during different levels of solar activity. Methods. Extreme ultraviolet filters of the Atmospheric Imaging Assembly (AIA) onboard the Solar Dynamics Observatory (SDO) were used. We analyzed 30 AIA/SDO image series between 2011 and 2018, each covering a 400″ × 400″ quiet-Sun field-of-view of over two hours with a 12-s cadence. Differential emission measure (DEM) analysis was used to derive the emission measure (EM) and temperature evolution for each pixel. We detected nanoflares as EM enhancements using a threshold-based algorithm and derived their thermal energy from the DEM observations. Results. Nanoflare energy distributions follow power laws that show slight variations in steepness (α = 2.02–2.47), but no correlation to the solar activity level. The combined nanoflare distribution of all data sets covers five orders of magnitude in event energies (1024 − 1029 erg) with a power-law index α = 2.28 ± 0.03. The derived mean energy flux of (3.7 ± 1.6)×104 erg cm−2 s−1 is one order of magnitude smaller than the coronal heating requirement. We found no correlation between the derived energy flux and solar activity. Analysis of the spatial distribution reveals clusters of high energy flux (up to 3 × 105 erg cm−2 s−1) surrounded by extended regions with lower activity. Comparisons with magnetograms from the Helioseismic and Magnetic Imager demonstrate that high-activity clusters are preferentially located in the magnetic network and above regions of enhanced magnetic flux density. Conclusions. The steep power-law slope (α &gt; 2) suggests that the total energy in the flare energy distribution is dominated by the smallest events, that is to say nanoflares. We demonstrate that in the quiet-Sun, the nanoflare distributions and their contribution to coronal heating does not vary over the solar cycle.

Human Activity Recognition Based on the Fading Characteristics of the On-Body Channel

Several solutions using wearable devices for continuous Human Activity Recognition (HAR) have been proposed up to now, and HAR sensor-based methods are by far the most adopted. However, many data transmission obstacles prevent the viable operation of these systems, such as those related to deep channel fading, the shadowing of body links during movements, packet loss resulting from unpredictable and dynamic body movements, and environmental variations. In this work, a novel HAR approach based solely on on-body Received Signal Strength Indicator (RSSI) measurements of different body channels is presented to discriminate various human activities. In this perspective, a novel hierarchical threshold-based algorithm is proposed to classify RSSI data into two classes. The first one encompasses static movements. The second one groups dynamic movements. In this paper, the features provided to machine learning algorithms are unconventionnaly based on the statistical characterization parameters of the body channel in an indoor environment. These parameters include the average of the slow fading channel component and the fast fading component&#x2019;s cumulative distribution function (CDF). In addition, the impact of the sliding window size adopted for the extraction of the slow fading component is investigated, as well as the impingement of the placement of the radio nodes on the body for the selection of the best on-body link for the transmission. It is shown that the proposed scheme provides higher accuracy compared to the state of the art radio-based HAR techniques and that the mislabeling problem is reduced by distinguishing between static and dynamic activities. Indeed, an approximate F1 score-based accuracy of 99&#x0025; is obtained with a demeaning window size of 50 ms, considering the most informative channel.

Detection of Grassland Mowing Events for Germany by Combining Sentinel-1 and Sentinel-2 Time Series

Grasslands cover one-third of the agricultural area in Germany and play an important economic role by providing fodder for livestock. In addition, they fulfill important ecosystem services, such as carbon storage, water purification, and the provision of habitats. These ecosystem services usually depend on the grassland management. In central Europe, grasslands are grazed and/or mown, whereby the management type and intensity vary in space and time. Spatial information on the mowing timing and frequency on larger scales are usually not available but would be required in order to assess the ecosystem services, species composition, and grassland yields. Time series of high-resolution satellite remote sensing data can be used to analyze the temporal and spatial dynamics of grasslands. Within this study, we aim to overcome the drawbacks identified by previous studies, such as optical data availability and the lack of comprehensive reference data, by testing the time series of various Sentinel-2 (S2) and Sentinal-1 (S1) parameters and combinations of them in order to detect mowing events in Germany in 2019. We developed a threshold-based algorithm by using information from a comprehensive reference dataset of heterogeneously managed grassland parcels in Germany, obtained by RGB cameras. The developed approach using the enhanced vegetation index (EVI) derived from S2 led to a successful mowing event detection in Germany (60.3% of mowing events detected, F1-Score = 0.64). However, events shortly before, during, or shortly after cloud gaps were missed and in regions with lower S2 orbit coverage fewer mowing events were detected. Therefore, S1-based backscatter, InSAR, and PolSAR features were investigated during S2 data gaps. From these, the PolSAR entropy detected mowing events most reliably. For a focus region, we tested an integrated approach by combining S2 and S1 parameters. This approach detected additional mowing events, but also led to many false positive events, resulting in a reduction in the F1-Score (from 0.65 of S2 to 0.61 of S2 + S1 for the focus region). According to our analysis, a majority of grasslands in Germany are only mown zero to two times (around 84%) and are probably additionally used for grazing. A small proportion is mown more often than four times (3%). Regions with a generally higher grassland mowing frequency are located in southern, south-eastern, and northern Germany.

Effectiveness of a Smartwatch App in Detecting Induced Falls: Observational Study.

BackgroundOlder adults are at an increased risk of falls with the consequent impacts on the health of the individual and health expenditure for the population. Smartwatch apps have been developed to detect a fall, but their sensitivity and specificity have not been subjected to blinded assessment nor have the factors that influence the effectiveness of fall detection been fully identified.ObjectiveThis study aims to assess accuracy metrics for a novel fall detection smartwatch algorithm.MethodsWe performed a cross-sectional study of 22 healthy adults comparing the detection of induced forward, side (left and right), and backward falls and near falls provided by a smartwatch threshold-based algorithm, with a video record of induced falls serving as the gold standard; a blinded assessor compared the two. Three different smartwatches with two different operating systems were used. There were 226 falls: 64 were backward, 51 forward, 55 left sided, and 56 right sided.ResultsThe overall smartwatch app sensitivity for falls was 77%, the specificity was 99%, the false-positive rate was 1.7%, and the false-negative rate was 16.4%. The positive and negative predictive values were 98% and 84%, respectively, while the accuracy was 89%. There were 249 near falls: the sensitivity was 89%, the specificity was 100%, there were no false positives, 11% were false negatives, the positive predictive value was 100%, the false-negative predictive value was 83%, and the accuracy was 93%.ConclusionsFalls were more likely to be detected if the fall was on the same side as the wrist with the smartwatch. There was a trend toward some smartwatches and operating systems having superior sensitivity, but these did not reach statistical significance. The effectiveness data and modifying factors pertaining to this smartwatch app can serve as a reference point for other similar smartwatch apps.

A Novel Radiomics-Based Tumor Volume Segmentation Algorithm for Lung Tumors in FDG-PET/CT after 3D Motion Correction-A Technical Feasibility and Stability Study.

Positron emission tomography (PET) provides important additional information when applied in radiation therapy treatment planning. However, the optimal way to define tumors in PET images is still undetermined. As radiomics features are gaining more and more importance in PET image interpretation as well, we aimed to use textural features for an optimal differentiation between tumoral tissue and surrounding tissue to segment-target lesions based on three textural parameters found to be suitable in previous analysis (Kurtosis, Local Entropy and Long Zone Emphasis). Intended for use in radiation therapy planning, this algorithm was combined with a previously described motion-correction algorithm and validated in phantom data. In addition, feasibility was shown in five patients. The algorithms provided sufficient results for phantom and patient data. The stability of the results was analyzed in 20 consecutive measurements of phantom data. Results for textural feature-based algorithms were slightly worse than those of the threshold-based reference algorithm (mean standard deviation 1.2%—compared to 4.2% to 8.6%) However, the Entropy-based algorithm came the closest to the real volume of the phantom sphere of 6 ccm with a mean measured volume of 26.5 ccm. The threshold-based algorithm found a mean volume of 25.0 ccm. In conclusion, we showed a novel, radiomics-based tumor segmentation algorithm in FDG-PET with promising results in phantom studies concerning recovered lesion volume and reasonable results in stability in consecutive measurements. Segmentation based on Entropy was the most precise in comparison with sphere volume but showed the worst stability in consecutive measurements. Despite these promising results, further studies with larger patient cohorts and histopathological standards need to be performed for further validation of the presented algorithms and their applicability in clinical routines. In addition, their application in other tumor entities needs to be studied.