Test Set Size Research Articles

Maximizing glaucoma detection accuracy through high confidence AI predictions: A RIM One database study

Purpose: This study aimed to assess an artificial intelligence (AI) model's performance in glaucoma detection using the RIM One public database. Specifically, we examined the impact of considering only high‐confidence predictions from the AI model on overall diagnostic accuracy.Methods: The RIM One database provided 485 images, including 313 normal and 172 glaucomatous cases. Of these, 248 images were for training the EfficientNetV2B0 model, 63 for validation, and 174 for testing. Initially, the model's performance was evaluated on the entire test set, yielding an AUC of 96%. Subsequently, we investigated the model's performance when limiting predictions to those with a certainty probability over 95%, reducing the test set to 155 images.Results: The AI model achieved an AUC of 96% when considering all predictions. However, focusing solely on high‐confidence predictions ( > 95% certainty) increased the AUC to 100%. This adjustment reduced the test set size from 174 to 155 images. Statistical analysis using the De Long Test revealed a significant difference between the two AUC values, highlighting the efficacy of high‐confidence predictions.Conclusion: Our findings demonstrate that by considering only AI predictions with a certainty probability exceeding 95%, we can significantly enhance glaucoma detection's diagnostic accuracy. This approach resulted in a perfect AUC, indicating robust performance in identifying glaucomatous cases. The reduction in test set size suggests that 19 cases may require further evaluation through methods such as visual field testing or optical coherence tomography (OCT) to confirm or exclude the diagnosis of glaucoma. This selective approach not only improves the efficiency of glaucoma diagnosis but also streamlines clinical decision‐making by directing resources to cases with the highest predictive certainty.

Bounding Random Test Set Size with Computational Learning Theory

Bounding Random Test Set Size with Computational Learning Theory

Deep Learning-Based Automated Detection of Retinal Breaks and Detachments on Fundus Photography.

The purpose of this study was to develop a deep learning algorithm, to detect retinal breaks and retinal detachments on ultra-widefield fundus (UWF) optos images using artificial intelligence (AI). Optomap UWF images of the database were annotated to four groups by two retina specialists: (1) retinal breaks without detachment, (2) retinal breaks with retinal detachment, (3) retinal detachment without visible retinal breaks, and (4) a combination of groups 1 to 3. The fundus image data set was split into a training set and an independent test set following an 80% to 20% ratio. Image preprocessing methods were applied. An EfficientNet classification model was trained with the training set and evaluated with the test set. A total of 2489 UWF images were included into the dataset, resulting in a training set size of 2008 UWF images and a test set size of 481 images. The classification models achieved an area under the receiver operating characteristic curve (AUC) on the testing set of 0.975 regarding lesion detection, an AUC of 0.972 for retinal detachment and an AUC of 0.913 for retinal breaks. A deep learning system to detect retinal breaks and retinal detachment using UWF images is feasible and has a good specificity. This is relevant for clinical routine as there can be a high rate of missed breaks in clinics. Future clinical studies will be necessary to evaluate the cost-effectiveness of applying such an algorithm as an automated auxiliary tool in a large practices or tertiary referral centers. This study demonstrates the relevance of applying AI in diagnosing peripheral retinal breaks in clinical routine in UWF fundus images.

Distributed Bayesian Inference for Large-Scale IoT Systems

In this work, we present a Distributed Bayesian Inference Classifier for Large-Scale Systems, where we assess its performance and scalability on distributed environments such as PySpark. The presented classifier consistently showcases efficient inference time, irrespective of the variations in the size of the test set, implying a robust ability to handle escalating data sizes without a proportional increase in computational demands. Notably, throughout the experiments, there is an observed increase in memory usage with growing test set sizes, this increment is sublinear, demonstrating the proficiency of the classifier in memory resource management. This behavior is consistent with the typical tendencies of PySpark tasks, which witness increasing memory consumption due to data partitioning and various data operations as datasets expand. CPU resource utilization, which is another crucial factor, also remains stable, emphasizing the capability of the classifier to manage larger computational workloads without significant resource strain. From a classification perspective, the Bayesian Logistic Regression Spark Classifier consistently achieves reliable performance metrics, with a particular focus on high specificity, indicating its aptness for applications where pinpointing true negatives is crucial. In summary, based on all experiments conducted under various data sizes, our classifier emerges as a top contender for scalability-driven applications in IoT systems, highlighting its dependable performance, adept resource management, and consistent prediction accuracy.

Overestimated prediction using polygenic prediction derived from summary statistics

BackgroundWhen polygenic risk score (PRS) is derived from summary statistics, independence between discovery and test sets cannot be monitored. We compared two types of PRS studies derived from raw genetic data (denoted as rPRS) and the summary statistics for IGAP (sPRS).ResultsTwo variables with the high heritability in UK Biobank, hypertension, and height, are used to derive an exemplary scale effect of PRS. sPRS without APOE is derived from International Genomics of Alzheimer’s Project (IGAP), which records ΔAUC and ΔR2 of 0.051 ± 0.013 and 0.063 ± 0.015 for Alzheimer’s Disease Sequencing Project (ADSP) and 0.060 and 0.086 for Accelerating Medicine Partnership - Alzheimer’s Disease (AMP-AD). On UK Biobank, rPRS performances for hypertension assuming a similar size of discovery and test sets are 0.0036 ± 0.0027 (ΔAUC) and 0.0032 ± 0.0028 (ΔR2). For height, ΔR2 is 0.029 ± 0.0037.ConclusionConsidering the high heritability of hypertension and height of UK Biobank and sample size of UK Biobank, sPRS results from AD databases are inflated. Independence between discovery and test sets is a well-known basic requirement for PRS studies. However, a lot of PRS studies cannot follow such requirements because of impossible direct comparisons when using summary statistics. Thus, for sPRS, potential duplications should be carefully considered within the same ethnic group.

Test case simplification based on coupling metrics in software bug location

Software test cases are one of the most critical aspects of software testing in the product development process. As software products are updated several times, the same test requirement may be covered by multiple test cases, so this aspect is often redundant, yet the approximate test case set has an impact on its error detection rate. This study proposes the idea of using redundant test cases in software error location, introduces a coupling metric, analyses its program slicing and establishes a second coverage criterion in order to balance the relationship between the reduced test suite and the false detection rate the test case set. The results show that the size of test set and the number of error detection by the Ruby On Rails (ROR) method used in this study are larger than those of other commonly used reduction algorithms. The test suite has the lowest error detection loss rate, with an average of 17.96 % across the six test case sets. The highest error detection capability of individual test cases was found in the reduced test set, with a mean value of 90.63 % in the test set. The method also has the highest average reduction efficiency of 91.05 %. Compared with other simplification methods, the research method has a better balance between the size and false detection rate of the reduced test suite and the advantages of simplification.

Clinical knowledge embedded method based on multi-task learning for thyroid nodule classification with ultrasound images

Objective. Thyroid nodules are common glandular abnormality that need to be diagnosed as benign or malignant to determine further treatments. Clinically, ultrasonography is the main diagnostic method, but it is highly subjective with severe variability. Recently, many deep-learning-based methods have been proposed to alleviate subjectivity and achieve good results yet, these methods often neglect important guidance from clinical knowledge. Our objective is to utilize such guidance for accurate and reliable thyroid nodule classification. Approach. In this study, a multi-task learning model embedded with clinical knowledge of ACR Thyroid Imaging, Reporting and Data System guideline is proposed. The clinical features defined in the guideline have strong correlations with malignancy and they were modeled as tasks alongside the pathological type. Multi-task learning was utilized to exploit the correlations to improve diagnostic performance. To alleviate the impact of noisy labels on clinical features, a loss-weighting strategy was proposed. Five-fold cross-validation was applied to an internal training set of size 4989, and an external test set of size 243 was used for evaluation. Main results. The proposed multi-task learning model achieved an average AUC of 0.901 and an ensemble AUC of 0.917 on the test set, which significantly outperformed the single-task baseline models. Significance. The results indicated that multi-task learning of clinical features can effectively classify thyroid nodules and reveal the possibility of using clinical indicators as auxiliary tasks to improve performance when diagnosing other diseases.

A novel ultrasound image diagnostic method for thyroid nodules

The incidence of thyroid nodules is increasing year by year. Accurate determination of benign and malignant nodules is an important basis for formulating treatment plans. Ultrasonography is the most widely used methodology in the diagnosis of benign and malignant nodules, but diagnosis by doctors is highly subjective, and the rates of missed diagnosis and misdiagnosis are high. To improve the accuracy of clinical diagnosis, this paper proposes a new diagnostic model based on deep learning. The diagnostic model adopts the diagnostic strategy of localization-classification. First, the distribution laws of the nodule size and nodule aspect ratio are obtained through data statistics, a multiscale localization network structure is a priori designed, and the nodule aspect ratio is obtained from the positioning results. Then, uncropped ultrasound images and nodule area image are correspondingly input into a two-way classification network, and an improved attention mechanism is used to enhance the feature extraction performance. Finally, the deep features, the shallow features, and the nodule aspect ratio are fused, and a fully connected layer is used to complete the classification of benign and malignant nodules. The experimental dataset consists of 4021 ultrasound images, where each image has been labeled under the guidance of doctors, and the ratio of the training set, validation set, and test set sizes is close to 3:1:1. The experimental results show that the accuracy of the multiscale localization network reaches 93.74%, and that the accuracy, specificity, and sensitivity of the classification network reach 86.34%, 81.29%, and 90.48%, respectively. Compared with the champion model of the TNSCUI 2020 classification competition, the accuracy rate is 1.52 points higher. Therefore, the network model proposed in this paper can effectively diagnose benign and malignant thyroid nodules.

AdaTest: Reinforcement Learning and Adaptive Sampling for On-chip Hardware Trojan Detection

This paper proposes AdaTest, a novel adaptive test pattern generation framework for efficient and reliable Hardware Trojan (HT) detection. HT is a backdoor attack that tampers with the design of victim integrated circuits (ICs) . AdaTest improves the existing HT detection techniques in terms of scalability and accuracy of detecting smaller Trojans in the presence of noise and variations. To achieve high trigger coverage, AdaTest leverages Reinforcement Learning (RL) to produce a diverse set of test inputs. Particularly, we progressively generate test vectors with high ‘reward’ values in an iterative manner. In each iteration, the test set is evaluated and adaptively expanded as needed. Furthermore, AdaTest integrates adaptive sampling to prioritize test samples that provide more information for HT detection, thus reducing the number of samples while improving the samples’ quality for faster exploration. We develop AdaTest with a Software/Hardware co-design principle and provide an optimized on-chip architecture solution. AdaTest’s architecture minimizes the hardware overhead in two ways: (i) Deploying circuit emulation on programmable hardware to accelerate reward evaluation of the test input; (ii) Pipelining each computation stage in AdaTest by automatically constructing auxiliary circuit for test input generation, reward evaluation, and adaptive sampling. We evaluate AdaTest’s performance on various HT benchmarks and compare it with two prior works that use logic testing for HT detection. Experimental results show that AdaTest engenders up to two orders of test generation speedup and two orders of test set size reduction compared to the prior works while achieving the same level or higher Trojan detection rate.

Effects of randomness in the development of machine learning models in neuroimaging studies of schizophrenia

Numerous studies have used machine learning with neuroimaging data for identifying individuals with a schizophrenia diagnosis. However, inconsistent results have limited the ability of the psychiatric community to objectively judge and accept the value of this approach. One factor that has contributed to the inconsistency, but has long been ignored, is randomness in the practice of machine learning. This is manifest when executing the same machine learning pipeline multiple times on the same dataset but getting different results. In the current study, a dataset of anatomical MRI scans from 158 patients with first-episode medication-naïve schizophrenia and 166 matched controls was used to investigate the effect of randomness on classifier performance estimates under different algorithm complexity and data splitting ratios. The maximum discriminatory accuracy that could be reached was 62.6 % ± 4.7 % (43.5 %–79.3 %) obtained when using extra-trees classifiers without feature normalization. Regions contributing to discrimination were located at bilateral temporal lobes and right frontal lobe. The results show that randomness has a significant impact on the precision of model performance estimates, especially when the size of test set is small. Current neuroimaging feature engineering combined with machine learning still falls short of being able to make diagnoses in the clinical context, but has value in revealing patterns of regional brain alteration associated with the illness. The current results indicate that effects of randomness on model performance should be reported and considered in interpreting model utility and it is necessary to evaluate models on large test sets to obtain valid estimates of model performance.

Incremental Testing in Software Product Lines—An Event Based Approach

One way of developing fast, effective, and high-quality software products is to reuse previously developed software components and products. In the case of a product family, the software product line (SPL) approach can make reuse more effective. The goal of SPLs is faster development of low-cost and high-quality software products. This paper proposes an incremental model-based approach to test products in SPLs. The proposed approach utilizes event-based behavioral models of the SPL features. It reuses existing event-based feature models and event-based product models along with their test cases to generate test cases for each new product developed by adding a new feature to an existing product. Newly introduced featured event sequence graphs (FESGs) are used for behavioral feature and product modeling; thus, generated test cases are event sequences. The paper presents evaluations with three software product lines to validate the approach and analyze its characteristics by comparing it to the state-of-the-art ESG-based testing approach. Results show that the proposed incremental testing approach highly reuses the existing test sets as intended. Also, it is superior to the state-of-the-art approach in terms of fault detection effectiveness and test generation effort but inferior in terms of test set size and test execution effort.

Feasibility of a Deep Learning approach to estimate Shear Wave Speed using the framework of Reverberant Shear Wave Elastography: A numerical simulation study.

Reverberant Shear Wave Elastography (RSWE) is an ultrasound elastography technique that offers great advantages, however, current estimators generate underestimations and time-consuming issues. As well, the involvement of Deep Learning into the medical imaging field with new tools to assess complex problems, makes it a great candidate to serve as a new approach for a RSWE estimator. This work addresses the application of a Deep Neural Network (DNN) for the estimation of Shear Wave Speed (SWS) maps from particle velocity using numerically simulated data. The architecture of the proposed network is based on a U-Net, which works with a custom loss function specifically adopted for the reconstruction task. Four DNNs were trained using four different databases: clean, noisy, acquired at variable frequency, and noisy and acquired at variable frequency data. After the training of the DNNs, the predicted SWS maps were evaluated based on different metrics related to segmentation, regression and similarity of images. The model for clean data showed better results with a Mean Absolute Error (MAE) of 0.011, Mean Square Error(MSE) of 0.001, modified Intersection over Union (mIoU) of 98.4%, Peak Signal to Noise Ratio (PSNR) of 32.925 and a Structural Similarity Index Measure (SSIM) of 0.99, for 250 (size of Testing Sets); while the other models delivered SSIM in the range of 0.87 to 0.96. It was concluded that noisy and clean data could be effectively handled by the model, while the other ones still need enhancement. Clinical Relevance- This work is focused on the application of a Deep Learning approach to accurately asses the Shear Wave Speed in numerical simulations of Reverberant Shear Wave Elastography approach. This novel estimator could be useful for future clinical experiments specially with real time applications to determine the status of living tissue such as detection of malignant or benign tumors located in breast cervix prostate or skin and in the diagnosis of other pathologies such us liver fibrosis.

DFT with Universal Test Set for All Missing Gate Faults in Reversible Circuits

The design of reversible logic circuits has received considerable attention in recent times for their potential use in implementing quantum computers. A fault model, namely, the Missing-Gate Fault (MGF) model, has been found to be more suitable for modeling defects in quantum circuits as compared to the classical fault models used for testing conventional CMOS circuits. In this paper, we propose two design-for-testability techniques with universal test set to detect MGFs in a reversible circuit. In these techniques, an [Formula: see text] reversible circuit implemented with [Formula: see text]-CNOT gates is transformed to a testable design with addition of two extra inputs lines and some additional gates, which uses a universal test set of size [Formula: see text] to detect the MGFs. The proposed DFT techniques are compared with earlier works and it has been shown that these techniques outperform the earlier techniques with respect to quantum cost overhead and percentage detection of the different MGFs.

Effectiveness Assessment of the Search-Based Statistical Structural Testing

Search-based statistical structural testing (SBSST) is a promising technique that uses automated search to construct input distributions for statistical structural testing. It has been proved that a simple search algorithm, for example, the hill-climber is able to optimize an input distribution. However, due to the noisy fitness estimation of the minimum triggering probability among all cover elements (Tri-Low-Bound), the existing approach does not show a satisfactory efficiency. Constructing input distributions to satisfy the Tri-Low-Bound criterion requires an extensive computation time. Tri-Low-Bound is considered a strong criterion, and it is demonstrated to sustain a high fault-detecting ability. This article tries to answer the following question: if we use a relaxed constraint that significantly reduces the time consumption on search, can the optimized input distribution still be effective in fault-detecting ability? In this article, we propose a type of criterion called fairness-enhanced-sum-of-triggering-probability (p-L1-Max). The criterion utilizes the sum of triggering probabilities as the fitness value and leverages a parameter p to adjust the uniformness of test data generation. We conducted extensive experiments to compare the computation time and the fault-detecting ability between the two criteria. The result shows that the 1.0-L1-Max criterion has the highest efficiency, and it is more practical to use than the Tri-Low-Bound criterion. To measure a criterion’s fault-detecting ability, we introduce a definition of expected faults found in the effective test set size region. To measure the effective test set size region, we present a theoretical analysis of the expected faults found with respect to various test set sizes and use the uniform distribution as a baseline to derive the effective test set size region’s definition.

Part-Of-Speech Tagging for Mizo Language Using Conditional Random Field

Part of speech (POS) tagging assigns a class or tag to each token in a sentence. The tag allocated to a word is mainly its part of speech or any other class of interest. Several applications of Natural Language Processing (NLP) require it as a prerequisite. The development of part-of-speech tagging for the under-resourced Mizo language is presented in this study, which makes use of a stochastic model known as Conditional Random Field (CRF). The CRF is a discriminative probabilistic classifier that considers both the context of a given word and the tag transition probabilities in the training dataset. A corpus of approximately 30,000 words was collected and manually annotated with the proposed tagset for system evaluation. On various sizes of training and test sets, the tagger achieved 89.46 % accuracy, 89.3 % F1-score, 89.42 % precision, and 89.48 % recall.

Towards radiologist-level cancer risk assessment in CT lung screening using deep learning.

Lung cancer is the leading cause of cancer mortality in the US, responsible for more deaths than breast, prostate, colon and pancreas cancer combined and large population studies have indicated that low-dose computed tomography (CT) screening of the chest can significantly reduce this death rate. Recently, the usefulness of Deep Learning (DL) models for lung cancer risk assessment has been demonstrated. However, in many cases model performances are evaluated on small/medium size test sets, thus not providing strong model generalization and stability guarantees which are necessary for clinical adoption. In this work, our goal is to contribute towards clinical adoption by investigating a deep learning framework on larger and heterogeneous datasets while also comparing to state-of-the-art models. Three low-dose CT lung cancer screening datasets were used: National Lung Screening Trial (NLST, n = 3410), Lahey Hospital and Medical Center (LHMC, n = 3154) data, Kaggle competition data (from both stages, n = 1397 + 505) and the University of Chicago data (UCM, a subset of NLST, annotated by radiologists, n = 132). At the first stage, our framework employs a nodule detector; while in the second stage, we use both the image context around the nodules and nodule features as inputs to a neural network that estimates the malignancy risk for the entire CT scan. We trained our algorithm on a part of the NLST dataset, and validated it on the other datasets. Special care was taken to ensure there was no patient overlap between the train and validation sets. The proposed deep learning model is shown to: (a) generalize well across all three data sets, achieving AUC between 86% to 94%, with our external test-set (LHMC) being at least twice as large compared to other works; (b) have better performance than the widely accepted PanCan Risk Model, achieving 6 and 9% better AUC score in our two test sets; (c) have improved performance compared to the state-of-the-art represented by the winners of the Kaggle Data Science Bowl 2017 competition on lung cancer screening; (d) have comparable performance to radiologists in estimating cancer risk at a patient level.

Construction of complete minimal test set for single intra-level bridging and stuck-at faults in reversible circuits

Reversible logic computing is gaining enough interest in the field of low-power circuit design technology. Moreover, the concept of reversible computing is widely used in quantum circuit computation. Therefore, it has motivated researchers to explore reversible logic as a circuit design alternative. Several problems of synthesis and optimisation of the reversible circuits have been reported in the present literature. In this context, the testing plays an important role in the effective performance of these circuits. For detecting all possible faults in the reversible circuits, numerous fault models have been proposed, where many of the proposed fault models common to conventional logic circuits. In this paper, we consider the problem of testing in the NCT or GT library-based reversible circuits with respect to an efficient test set generation for single intra-level bridging faults and single stuck-at faults. The proposed test generation method has been developed based on the concept of one-to-one mapping mechanism (reversible circuit property). Finally, experimental results show that the generated test is capable of 100% fault coverage for both the fault models and also the results are compared with the existing methods for analysing the test set size.

Attention-Guided 3D-CNN Framework for Glaucoma Detection and Structural-Functional Association Using Volumetric Images.

The direct analysis of 3D Optical Coherence Tomography (OCT) volumes enables deep learning models (DL) to learn spatial structural information and discover new bio-markers that are relevant to glaucoma. Downsampling 3D input volumes is the state-of-art solution to accommodate for the limited number of training volumes as well as the available computing resources. However, this limits the network's ability to learn from small retinal structures in OCT volumes. In this paper, our goal is to improve the performance by providing guidance to DL model during training in order to learn from finer ocular structures in 3D OCT volumes. Therefore, we propose an end-to-end attention guided 3D DL model for glaucoma detection and estimating visual function from retinal structures. The model consists of three pathways with the same network architecture but different inputs. One input is the original 3D-OCT cube and the other two are computed during training guided by the 3D gradient class activation heatmaps. Each pathway outputs the class-label and the whole model is trained concurrently to minimize the sum of losses from three pathways. The final output is obtained by fusing the predictions of the three pathways. Also, to explore the robustness and generalizability of the proposed model, we apply the model on a classification task for glaucoma detection as well as a regression task to estimate visual field index (VFI) (a value between 0 and 100). A 5-fold cross-validation with a total of 3782 and 10,370 OCT scans is used to train and evaluate the classification and regression models, respectively. The glaucoma detection model achieved an area under the curve (AUC) of 93.8% compared with 86.8% for a baseline model without the attention-guided component. The model also outperformed six different feature based machine learning approaches that use scanner computed measurements for training. Further, we also assessed the contribution of different retinal layers that are relevant to glaucoma. The VFI estimation model achieved a Pearson correlation and median absolute error of 0.75 and 3.6%, respectively, for a test set of size 3100 cubes.

The Tox21 10K Compound Library: Collaborative Chemistry Advancing Toxicology.

Since 2009, the Tox21 project has screened ∼8500 chemicals in more than 70 high-throughput assays, generating upward of 100 million data points, with all data publicly available through partner websites at the United States Environmental Protection Agency (EPA), National Center for Advancing Translational Sciences (NCATS), and National Toxicology Program (NTP). Underpinning this public effort is the largest compound library ever constructed specifically for improving understanding of the chemical basis of toxicity across research and regulatory domains. Each Tox21 federal partner brought specialized resources and capabilities to the partnership, including three approximately equal-sized compound libraries. All Tox21 data generated to date have resulted from a confluence of ideas, technologies, and expertise used to design, screen, and analyze the Tox21 10K library. The different programmatic objectives of the partners led to three distinct, overlapping compound libraries that, when combined, not only covered a diversity of chemical structures, use-categories, and properties but also incorporated many types of compound replicates. The history of development of the Tox21 “10K” chemical library and data workflows implemented to ensure quality chemical annotations and allow for various reproducibility assessments are described. Cheminformatics profiling demonstrates how the three partner libraries complement one another to expand the reach of each individual library, as reflected in coverage of regulatory lists, predicted toxicity end points, and physicochemical properties. ToxPrint chemotypes (CTs) and enrichment approaches further demonstrate how the combined partner libraries amplify structure–activity patterns that would otherwise not be detected. Finally, CT enrichments are used to probe global patterns of activity in combined ToxCast and Tox21 activity data sets relative to test-set size and chemical versus biological end point diversity, illustrating the power of CT approaches to discern patterns in chemical–activity data sets. These results support a central premise of the Tox21 program: A collaborative merging of programmatically distinct compound libraries would yield greater rewards than could be achieved separately.

Efficient static compaction of test patterns using partial maximum satisfiability

Static compaction methods aim at finding unnecessary test patterns to reduce the size of the test set as a post-process of test generation. Techniques based on partial maximum satisfiability are often used to track many hard problems in various domains, including artificial intelligence, computational biology, data mining, and machine learning. We observe that part of the test patterns generated by the commercial Automatic Test Pattern Generation (ATPG) tool is redundant, and the relationship between test patterns and faults, as a significant information, can effectively induce the test patterns reduction process. Considering a test pattern can detect one or more faults, we map the problem of static test compaction to a partial maximum satisfiability problem. Experiments on ISCAS89, ISCAS85, and ITC99 benchmarks show that this approach can reduce the initial test set size generated by TetraMAX18 while maintaining fault coverage.