Complex Features Research Articles

Using topological data analysis and machine learning to predict customer churn

Customer churn is a common problem faced by many industries, including telecommunication industries. This has resulted in the development of advanced techniques for the prediction and prevention of customer churn. The availability of stored customer data in the form of big data, together with the use of advanced and tuned machine learning (ML) algorithms, have paved the way for the realisation and extraction of useful features associated with customer behaviour and consequently the prediction of customer churn. An effective way to further improve churn prediction capability of different ML algorithms is through the employment of topological data analysis (TDA). TDA is a framework that applies topological methods to uncover the underlying hidden structural features in complex, high-dimensional data. Here, a TDA summary of 0- and 1-dimensional holes of the data, called barcode statistics, was extracted and used as an additional feature to the preprocessed customer data. To address issues such as the effective preprocessing and analysis of large customer datasets and the effective tuning of ML hyperparameters, we implement an advanced data preprocessing technique that consists of different stages such as handling of missing data, feature engineering, encoding of categorical features using the hashing encoding method, and feature selection. Without including barcode statistics in the model, the XGBoost algorithm with tuned hyperparameters achieved the best results, with accuracy of 92.71%, precision of 85.95%, recall of 92.71%, and F-measure of 89.20%. Including barcode statistics as an additional feature, the XGBoost algorithm with tuned hyperparameters achieved the best and much improved results, with accuracy of 98.50%, precision of 98.50%, recall of 98.50%, and F-measure of 98.50%. The use of TDA barcode statistics significantly improved the churn prediction capability of the ML algorithms. In addition, hyperparameter tuning is not needed when an effective data preprocessing technique is used, or when barcode statistics is used. The best accuracy of 98.5% from this work was in line with the best accuracy of 98.7% from a related work, but interestingly, the best precision of 98.5% from this work was superior to the 94.3% precision from the same related work with higher accuracy.

Bio-assembling and Bioprinting for Engineering Microvessels from the Bottom Up.

Blood vessels are essential in transporting nutrients, oxygen, metabolic wastes, and maintaining the homeostasis of the whole human body. Mass of engineered microvessels is required to deliver nutrients to the cells included in the constructed large three-dimensional (3D) functional tissues by diffusion. It is a formidable challenge to regenerate microvessels and build a microvascular network, mimicking the cellular viabilities and activities in the engineered organs with traditional or existing manufacturing techniques. Modular tissue engineering adopting the “bottom-up” approach builds one-dimensional (1D) or two-dimensional (2D) modular tissues in micro scale first and then uses these modules as building blocks to generate large tissues and organs with complex but indispensable microstructural features. Building the microvascular network utilizing this approach could be appropriate and adequate. In this review, we introduced existing methods using the “bottom-up” concept developed to fabricate microvessels including bio-assembling powered by different micromanipulation techniques and bioprinting utilizing varied solidification mechanisms. We compared and discussed the features of the artificial microvessels engineered by these two strategies from multiple aspects. Regarding the future development of engineering the microvessels from the bottom up, potential directions were also concluded.

Mutations in glioblastoma proteins do not disrupt epitope presentation and recognition, maintaining a specific CD8 T cell immune response potential

Antigen-specific cytotoxic CD8 T cells are extremely effective in controlling tumor growth and have been the focus of immunotherapy approaches. We leverage in silico tools to investigate whether the occurrence of mutations in proteins previously described as immunogenic and highly expressed by glioblastoma multiforme (GBM), such as Epidermal Growth Factor Receptor (EGFR), Isocitrate Dehydrogenase 1 (IDH1), Phosphatase and Tensin homolog (PTEN) and Tumor Protein 53 (TP53), may be contributing to the differential presentation of immunogenic epitopes. We recovered Class I MHC binding information from wild-type and mutated proteins using the Immune Epitope Database (IEDB). After that, we built peptide-MHC (pMHC-I) models in HLA-arena, followed by hierarchical clustering analysis based on electrostatic surface features from each complex. We identified point mutations that are determinants for the presentation of a set of peptides from TP53 protein. We point to structural features in the pMHC-I complexes of wild-type and mutated peptides, which may play a role in the recognition of CD8 T cells. To further explore these features, we performed 100 ns molecular dynamics simulations for the peptide pairs (wt/mut) selected. In pursuit of novel therapeutic targets for GBM treatment, we selected peptides where our predictive results indicated that mutations would not disrupt epitope presentation, thereby maintaining a specific CD8 T cell immune response. These peptides hold potential for future GBM interventions, including peptide-based or mRNA vaccine development applications.

Vehicle trajectory prediction based on attention optimized with real-scene sampling

Advancements in autonomous vehicles and deep learning have notably improved vehicle trajectory prediction accuracy. However, extracting interaction features in complex driving scenarios, such as vehicle-to-vehicle interactions and lane constraints, presents challenges. Deep learning-based methods struggle to achieve optimal predictive performance under limited computational resources. This study introduces a global attention mechanism to enhance feature extraction from driving scene encodings, focusing the decoder on interactive behaviours and boosting long-term prediction performance. An adaptive scheduled sampling model is employed, using actual driving scenarios probabilistically for training, addressing slow learning of actual driving behaviours and lack of initial feature correction. This method increases attention to actual interactions, reducing reliance on natural scenes and improving model generalizability. On the NGSIM dataset, sampling attention encoder-decoder (SAED) achieves a 1–5 s average displacement error (ADE) of 1.34 m, with 4 s and 5 s final displacement errors (FDEs) of 1.64 and 2.06 m, respectively. Compared to methods based on long short-term memory (LSTM), SAED reduces the model's storage space by 24.68% under the same network layer count. That demonstrates its effectiveness in extracting interactive behaviours in complex scenarios and enhances the accuracy of long-term predictions.

Scaled Position Consensus of High-Order Uncertain Multiagent Systems Over Switching Directed Graphs.

We investigate the scaled position consensus of high-order multiagent systems with parametric uncertainties over switching directed graphs, where the agents' position states reach a consensus value with different scales. The intricacy arises from the asymmetry inherent in information interaction. Achieving scaled position consensus in high-order multiagent systems over directed graphs remains a significant challenge, particularly when confronted with the following complex features: 1) uniformly jointly connected switching directed graphs; 2) complex agent dynamics with unknown inertias, unknown control directions, parametric uncertainties, and external disturbances; 3) interacting with each other via only relative scaled position information (without high-order derivatives of relative position); and 4) fully distributed in terms of no shared gains and no global gain dependency. To address these challenges, we propose a distributed adaptive algorithm based on a acrlong MRACon scheme, where a linear high-order reference model is designed for every individual agent employing relative scaled position information as input. A new transformation is proposed which converts the scaled position consensus of high-order linear reference models to that of first-order ones. Theoretical analysis is presented where agents' positions achieve the scaled consensus over switching directed graphs. Numerical simulations are performed to validate the efficacy of our algorithm and some collective behaviors on traditional consensus, bipartite consensus, and cluster consensus are shown by precisely choosing the scales of the agents.

On the issue of blood circulation features in the fetoplacental complex in pregnant women with thyroid diseases

Studying the issues of pregnancy management with extragenital diseases is a relevant and important task of modern obstetrics. The issue of the influence of thyroid diseases on the formation and function of the placenta remains not fully studied. Purpose of the study: to study the characteristics of the uteroplacental and fetal placental circulation in pregnant women with certain diseases of the thyroid gland. Material and research methods. 170 pregnant women were under observation. The 1st group (64 people) was formed by patients with autoimmune thyroiditis (AIT), the 2nd group (76 people) was formed by patients with diffuse toxic goiter (DTG), the 3rd (control) group was made up of 30 women with normal pregnancy. The study of feto-placental blood circulation was carried out by analyzing blood flow velocity curves (BSV) in the uterine arteries and in the umbilical cord artery. Statistical processing was carried out using the StatSoft program (Russia). The significance of the difference between the compared groups was determined using the Student's t test. Research results. An increase in indices of peripheral vascular resistance of the uterine arteries and a decrease in the systolic-diastolic ratio in the umbilical arteries in pregnant women with thyroid diseases were established, while more pronounced changes were observed in patients with autoimmune thyroiditis. Ultrasound data indicated the formation of fetoplacental insufficiency during pregnancy in women with thyroid diseases. Conclusion. To predict placental dysfunction and the formation of a risk group for the formation of perinatal complications, it is necessary to take into account the identified sonographic changes that are observed during pregnancy in patients with thyroid diseases.

Assessment features for complexity of educational texts for schoolchildren

It is extremely important to choose an appropriate material for the students learning a foreign language. Hence the issues of studying the complexity of academic texts, their readability and their compliance with the level of language proficiency is of outmost importance. At this point, one can note the inevitable interconnection between linguistics and teaching methods. An educational text is a special type of secondary text that has a number of important functions, among which, first of all, there are informative, cognitive and cultural-logical ones. This type of text both contains new information and has an educational and developmental effect on students.The difficulty of text as well as its readability and complexity are crucial for text understanding. The evaluation of academic texts difficulty is closely connected with consideration of other characteristics. However, there are no widely accepted definitions of such concepts as “difficulty”, “complexity” and “readability” in contemporary linguistics. They are unreasonably used as synonymic notions. That is why the purpose of this article is to more clearly distinguish between these characteristics of the text. The features of complexity assessment of educational texts are revealed on the basis of their key characteristics.The study of educational texts’ complexity was conducted by using secondary school textbooks on physics. The text analysis was performed with the help of the automated analyzer Rulingva. This study demonstrates the capabilities of this program and presents the results of the analysis, which suggest that the use of Rulingva analyzer could be beneficial not only for linguists, but also for educators, measurement and control materials developers and those preparing for state exams in various academic disciplines.

Contrastive Unfolding Deraining Network.

Due to the fact that the degradation of image quality caused by rain usually affects outdoor vision tasks, image deraining becomes more and more important. Focusing on the single image deraining (SID) task, in this article, we propose a novel Contrastive Unfolding DEraining Network (CUDEN), which combines the traditional iterative algorithm and deep network, exhibiting excellent performance and nice interpretability. CUDEN transforms the challenge of locating rain streaks into discovering rain features and defines the relationship between the image and feature domains in terms of mapping pairs. To obtain the mapping pairs efficiently, we propose a dynamic multidomain translation (DMT) module for decomposing the original mapping into sub-mappings. To enhance the feature extraction capability of networks, we also propose a new serial multireceptive field fusion (SMF) block, which extracts complex and variable rain features with convolution kernels of different receptive fields. Moreover, we are the first to introduce contrastive learning to the SID task and combine it with perceptual loss to propose a new contrastive perceptual loss (CPL), which is quite generalized and greatly helpful in identifying the appropriate gradient descent direction during training. Extensive experiments on synthetic and real-world datasets demonstrate that our proposed CUDEN outperforms the state-of-the-art (SOTA) deraining networks.

Precision-Boosted Forest Fire Target Detection via Enhanced YOLOv8 Model

Forest fires present a significant challenge to ecosystems, particularly due to factors like tree cover that complicate fire detection tasks. While fire detection technologies, like YOLO, are widely used in forest protection, capturing diverse and complex flame features remains challenging. Therefore, we propose an enhanced YOLOv8 multiscale forest fire detection method. This involves adjusting the network structure and integrating Deformable Convolution and SCConv modules to better adapt to forest fire complexities. Additionally, we introduce the Coordinate Attention mechanism in the Detection module to more effectively capture feature information and enhance model accuracy. We adopt the WIoU v3 loss function and implement a dynamically non-monotonic mechanism to optimize gradient allocation strategies. Our experimental results demonstrate that our model achieves a mAP of 90.02%, approximately 5.9% higher than the baseline YOLOv8 network. This method significantly improves forest fire detection accuracy, reduces False Positive rates, and demonstrates excellent applicability in real forest fire scenarios.

Fast prototype and rapid construction of three-dimensional and multi-scaled pitcher for controlled drainage by systematic biomimicry

Biomimetic materials that use natural wisdom to solve practical problems are developing rapidly. The trend for systematic biomimicry is towards in-situ characterization of natural creatures with high spatial resolutions. Furthermore, rapid reconstruction of digital twin models with the same complex features as the prototype is indispensable. However, it faces bottlenecks and limits in fast characterization and fabrication, precise parameter optimization, geometric deviations control, and quality prediction. To solve these challenges, here, we demonstrate a state-of-the-art method taking advantage of micro-computed tomography and three-dimensional printing for the fast characterization of the pitcher plant Nepenthes x ventrata and fabrication of its biomimetic model to obtain a superior drainage controller with multiscale structures with precise surface morphology optimization and geometric deviation control. The film-rupture-based drainage dynamic and mechanisms are characterized by x-ray and high-speed videography, which determines the crucial structures for unique directional drainage. Then the optimized artificial pitchers are further developed into sustained drainage devices with novel applications, such as detection, reaction, and smoke control.

Evaluation of the Potential for CO2 Storage and Saline Water Displacement in Huaiyin Sag, Subei Basin, East China

CO2 geological storage combined with deep saline water recovery technology (CO2-EWR) is one of the most effective ways to reduce carbon emissions. Due to the complex structural features, it is difficult to use CO2-EWR technology in Huaiyin Sag, Subei basin, East China. In this study, the multi-source information superposition evaluation technology of GIS was utilized for the selection of CO2 storage sites and water displacement potential target areas in this area, which mainly focused on the sandstone reservoirs of Cretaceous Pukou Formation. Based on the results, a three-dimensional injection–extraction model was established. Various scenarios with different production/injection well ratios (PIR) were simulated. Research has shown that the suitability of the surrounding site of Huaiyin Power Plant can be divided into two levels: relatively suitable and generally suitable; the area in the generally suitable level accounts for more than 80%. At a PIR of 1, CO2 is distributed asymmetrically, whereas at PIRs of 2 or 4, CO2 is distributed symmetrically. When the number of production wells is constant, a higher injection rate results in a faster expansion rate of the CO2 plume. This means that the time taken for the CO2 plume to reach the production wells is shorter. Reservoir pressure increases rapidly after more than 60 years of CO2 injection at lower PIR values, while at higher PIRs, reservoir pressure eventually stabilizes. Higher PIR values correspond to higher gas saturation, indicating a greater capacity for CO2 sequestration with more producing wells. When PIR = 4, the total CO2 injection increased by 55.73% compared to PIR = 1. However, the extraction of saline decreases with an increase in the number of producing wells, resulting in a decrease in replacement efficiency. This study provides a theoretical basis and technical support for the implementation of large-scale CO2-EWR engineering and technology demonstration in this region.

Minimizing Dimensional Defects in FFF Using a Novel Adaptive Slicing Method Based on Local Shape Complexity

Additive Manufacturing (AM) has emerged as an innovative technology that gives designers several advantages, such as geometric freedom of design and less waste. However, the quality of the parts produced is affected by different design and manufacturing parameters, such as the part orientation, the nozzle temperature and speed, the support material, and the layer thickness. In this context, the layer thickness is considered an important AM parameter affecting the part quality and accuracy. Thus, in this paper, a new adaptative slicing method based on the cusp vector and the surface deviation is proposed with the aim of minimizing the dimensional defects of FFF printed parts and investigate the impact on the dimensional part tolerancing. An algorithm is developed to automatically extract data from the STL file, select the build orientation, and detect intersection points between the initial slicing and the STL mesh. The innovation of this algorithm is exhibited via adapting the slicing according to the surface curvature based on two factors: the cusp vector and the surface deviation. The suggested slicing technique guarantees dimensional accuracy, especially for complex feature shapes that are challenging to achieve using a uniform slicing approach. Finally, a preview of the slicing is displayed, and the G-code is generated to be used by the FFF machine. The case study consists of the dimensional tolerance inspection of prototypes manufactured using the conventional and adaptive slicing processes. The proposed method’s effectiveness is investigated using RE and CMM processes. The method demonstrates its reliability through the observed potential for accuracy improvements exceeding 0.6% and cost savings of up to 4.3% in specific scenarios. This reliability is substantiated by comparing the resulting dimensional tolerances and manufacturing costs.

Improving precipitation estimates for Turkey with multimodel ensemble: a comparison of nonlinear artificial neural network method with linear methods

AbstractEnsemble analysis is proven to provide advantages in climate change impact assessment based on outputs from climate models. Ensembled series are shown to outperform single-model assessments through increased consistency and stability. This study aims to test the improvement of precipitation estimates through the use of ensemble analysis for south and southwestern Turkey which is known to have complex climatic features due to varying topography and interacting climate forcings. The analysis covers an evaluation of the performance of eight regional climate models (RCMs) from the EUR-11 domain available from the CORDEX database. The historical outputs are evaluated for their representativeness of the current climate of the Mediterranean region and its surroundings in Turkey through a comparison with long-term monthly precipitation time series obtained from ground-based precipitation observations by the use of statistical performance indicators and Taylor diagrams. This is followed by a comparative evaluation of three ensemble methodologies, simple average of the models, multiple linear regression for superensemble, and artificial neural networks (ANN). The analysis results show that the overall performance of ensembled time series is better compared to individual RCMs. ANN generally provided the best performance when all RCMs are used as inputs. Improvement in the performance of ensembling due to the use of nonlinear models is further confirmed by fuzzy inference systems (FIS). Both ANN and FIS generated monthly precipitation time series with higher correlations with those of observations. However, extreme events are poorly represented in the ensembled time series, and this may result in inefficiency in the design of various water structures such as spillways and storm water drainage systems that are based on high return period events.

LUMBAR syndrome-OEIS complex overlap: A case series and review.

We present three new and six published infants with overlapping features of LUMBAR syndrome (lower body hemangioma, urogenital anomalies, spinal cord malformations, bony deformities, anorectal/arterial anomalies and renal anomalies) and OEIS complex (omphalocele, exstrophy, imperforate anus, and spinal defects), also known as cloacal exstrophy. OEIS is included under the recently proposed umbrella coined recurrent constellations of embryonic malformations (RCEMs). The RCEMs represent a phenotypically overlapping spectrum of rare disorders of caudal dysgenesis with unknown cause but likely shared pathogenesis. It has recently been proposed that LUMBAR be considered an RCEM. This report of infants with combined features of OEIS and LUMBAR is the first to demonstrate an overlap between LUMBAR and another RCEM, which supports LUMBAR's inclusion within the RCEM spectrum.

Recent Advances in Antibacterial Coatings to Combat Orthopedic Implant-Associated Infections.

Implant-associated infections (IAIs) represent a major health burden due to the complex structural features of biofilms and their inherent tolerance to antimicrobial agents and the immune system. Thus, the viable options to eradicate biofilms embedded on medical implants are surgical operations and long-term and repeated antibiotic courses. Recent years have witnessed a growing interest in the development of robust and reliable strategies for prevention and treatment of IAIs. In particular, it seems promising to develop materials with anti-biofouling and antibacterial properties for combating IAIs on implants. In this contribution, we exclusively focus on recent advances in the development of modified and functionalized implant surfaces for inhibiting bacterial attachment and eventually biofilm formation on orthopedic implants. Further, we highlight recent progress in the development of antibacterial coatings (including self-assembled nanocoatings) for preventing biofilm formation on orthopedic implants. Among the recently introduced approaches for development of efficient and durable antibacterial coatings, we focus on the use of safe and biocompatible materials with excellent antibacterial activities for local delivery of combinatorial antimicrobial agents for preventing and treating IAIs and overcoming antimicrobial resistance.

Exploring Emergent Properties in Enzymatic Reaction Networks: Design and Control of Dynamic Functional Systems.

The intricate and complex features of enzymatic reaction networks (ERNs) play a key role in the emergence and sustenance of life. Constructing such networks in vitro enables stepwise build up in complexity and introduces the opportunity to control enzymatic activity using physicochemical stimuli. Rational design and modulation of network motifs enable the engineering of artificial systems with emergent functionalities. Such functional systems are useful for a variety of reasons such as creating new-to-nature dynamic materials, producing value-added chemicals, constructing metabolic modules for synthetic cells, and even enabling molecular computation. In this review, we offer insights into the chemical characteristics of ERNs while also delving into their potential applications and associated challenges.

A Transformer-Based Knowledge Distillation Network for Cortical Cataract Grading.

Cortical cataract, a common type of cataract, is particularly difficult to be diagnosed automatically due to the complex features of the lesions. Recently, many methods based on edge detection or deep learning were proposed for automatic cataract grading. However, these methods suffer a large performance drop in cortical cataract grading due to the more complex cortical opacities and uncertain data. In this paper, we propose a novel Transformer-based Knowledge Distillation Network, called TKD-Net, for cortical cataract grading. To tackle the complex opacity problem, we first devise a zone decomposition strategy to extract more refined features and introduce special sub-scores to consider critical factors of clinical cortical opacity assessment (location, area, density) for comprehensive quantification. Next, we develop a multi-modal mix-attention Transformer to efficiently fuse sub-scores and image modality for complex feature learning. However, obtaining the sub-score modality is a challenge in the clinic, which could cause the modality missing problem instead. To simultaneously alleviate the issues of modality missing and uncertain data, we further design a Transformer-based knowledge distillation method, which uses a teacher model with perfect data to guide a student model with modality-missing and uncertain data. We conduct extensive experiments on a dataset of commonly-used slit-lamp images annotated by the LOCS III grading system to demonstrate that our TKD-Net outperforms state-of-the-art methods, as well as the effectiveness of its key components. Codes are available at https://github.com/wjh892521292/Cataract_TKD-Net.

Shots segmentation-based optimized dual-stream framework for robust human activity recognition in surveillance video

Nowadays, for controlling crime, surveillance cameras are typically installed in all public places to ensure urban safety and security. However, automating Human Activity Recognition (HAR) using computer vision techniques faces several challenges such as lowlighting, complex spatiotemporal features, clutter backgrounds, and inefficient utilization of surveillance system resources. Existing attempts in HAR designed straightforward networks by analyzing either spatial or motion patterns resulting in limited performance while the dual streams methods are entirely based on Convolutional Neural Networks (CNN) that are inadequate to learning the long-range temporal information for HAR. To overcome the above-mentioned challenges, this paper proposes an optimized dual stream framework for HAR which mainly consists of three steps. First, a shots segmentation module is introduced in the proposed framework to efficiently utilize the surveillance system resources by enhancing the lowlight video stream and then it detects salient video frames that consist of human. This module is trained on our own challenging Lowlight Human Surveillance Dataset (LHSD) which consists of both normal and different levels of lowlighting data to recognize humans in complex uncertain environments. Next, to learn HAR from both contextual and motion information, a dual stream approach is used in the feature extraction. In the first stream, it freezes the learned weights of the backbone Vision Transformer (ViT) B-16 model to select the discriminative contextual information. In the second stream, ViT features are then fused with the intermediate encoder layers of FlowNet2 model for optical flow to extract a robust motion feature vector. Finally, a two stream Parallel Bidirectional Long Short-Term Memory (PBiLSTM) is proposed for sequence learning to capture the global semantics of activities, followed by Dual Stream Multi-Head Attention (DSMHA) with a late fusion strategy to optimize the huge features vector for accurate HAR. To assess the strength of the proposed framework, extensive empirical results are conducted on real-world surveillance scenarios and various benchmark HAR datasets that achieve 78.6285%, 96.0151%, and 98.875% accuracies on HMDB51, UCF101, and YouTube Action, respectively. Our results show that the proposed strategy outperforms State-of-the-Art (SOTA) methods. The proposed framework gives superior performance in HAR, providing accurate and reliable recognition of human activities in surveillance systems.

RED-Net: Residual and Enhanced Discriminative Network for Image Steganalysis in the Internet of Medical Things and Telemedicine.

Internet of Medical Things (IoMT) and telemedicine technologies utilize computers, communications, and medical devices to facilitate off-site exchanges between specialists and patients, specialists, and medical staff. If the information communicated in IoMT is illegally steganography, tampered or leaked during transmission and storage, it will directly impact patient privacy or the consultation results with possible serious medical incidents. Steganalysis is of great significance for the identification of medical images transmitted illegally in IoMT and telemedicine. In this article, we propose a Residual and Enhanced Discriminative Network (RED-Net) for image steganalysis in the internet of medical things and telemedicine. RED-Net consists of a steganographic information enhancement module, a deep residual network, and steganographic information discriminative mechanism. Specifically, a steganographic information enhancement module is adopted by the RED-Net to boost the illegal steganographic signal in texturally complex high-dimensional medical image features. A deep residual network is utilized for steganographic feature extraction and compression. A steganographic information discriminative mechanism is employed by the deep residual network to enable it to recalibrate the steganographic features and drop high-frequency features that are mistaken for steganographic information. Experiments conducted on public and private datasets with data hiding payloads ranging from 0.1bpp/bpnzac-0.5bpp/bpnzac in the spatial and JPEG domain led to RED-Net's steganalysis error PE in the range of 0.0732-0.0010 and 0.231-0.026, respectively. In general, qualitative and quantitative results on public and private datasets demonstrate that the RED-Net outperforms 8 state-of-art steganography detectors.

Lung Cancer Detection and Severity Analysis with a 3D Deep Learning CNN Model Using CT-DICOM Clinical Dataset

Objectives: To propose a new AI based CAD model for early detection and severity analysis of pulmonary (lung) cancer disease. A deep learning artificial intelligence-based approach is employed to maximize the discrimination power in CT images and minimize the dimensionality in order to boost detection accuracy. Methods: The AI-based 3D Convolutional Neural Network (3D-DLCNN) method is employed to learn complex patterns and features in a robust way for efficient detection and classification. The pulmonary nodules are identified by 3D Mask-R-CNN at the initial level, and classification is done by 3D-DLCNN. Kernel Density Estimation (KDE) is used to discover the error data points in the extracted features for early removal before candidate screening. The study uses the CT-DICOM dataset, which includes 355 instances and 251135 CT-DICOM images with target attributes of cancer, healthy, and severity condition (if cancer is positive). Statistical outlier detection is utilized to measure the z-score of each feature to reduce the data point deviation. The intensity and pixel masking of CT-DOCIM is measured by using the ER-NCN method to identify the severity of the disease. The performance of the 3D-DLCNN model is done using the MATLAB R2020a tool and comparative analysis is done with prevailing detection and classification approaches such as GA-PSO, SVM, KNN, and BPNN. Findings: The suggested pulmonary detection 3D-DLCNN model outperforms the prevailing models with promising results of 93% accuracy rate, 92.7% sensitivity, 93.4% specificity, 0.8 AUC-ROC, 6.6% FPR, and 0.87 C-Index, which helps the pulmonologists detect the PC and identify the severity for early diagnosis. Novelty: The novel hybrid 3D-DLCNN approach has the ability to detect pulmonary disease and analyze the severity score of the patient at an early stage during the screening process of candidates. It overcomes the limitations of the prevailing machine learning models, GA-PSO, SVM, KNN, and BPNN. Keywords: Artificial Intelligence, Disease Prediction, Lung Cancer, Deep Learning, Cancer Detection, Computational Model, 3D-DLCNN