Learning-based Convolutional Neural Network Research Articles

Analyzing the Deep Learning-Based Mobile Environment in Educational Institutions

More efficient instructional models that encourage students to engage more actively in their education are essential in today’s world. How information can be found and exchanged, as well as how explanations are delivered, have been shaped by technology. The outcomes demonstrate that mobile learning (m-learning) has enhanced collaboration among instructors and learners, provided immediate feedback, increased student participation and engagement, allowed for authentic learning and evaluation, and supported learning communities in higher education institutions. There was also a change in the teachers’ instructional methods. Students will benefit from simple learning activities, convenient communication, and coaching provided through mobile devices. This study recommends using a learning-based convolutional neural network (CNN) technique to improve students’ English-speaking fluency. In addition, communication between teachers and children in remote areas becomes easier through an interactive system, making both sides more accessible. As a result, it is advisable that educational institutions continually develop innovative teaching techniques that bridge face-to-face and formal-informal learning in the educational setting. The article discusses the background of mobile learning and how it might enhance e-learning as a whole. The m-learning approach is described in this paper as a future version of e-learning. In comparison to the next generation of educational systems, it will be extensively offered and simple to operate for any individual interested in learning. The article also discusses the benefits and potential drawbacks of mobile learning in current educational institutions.

Multimodal radiotherapy dose prediction using a multi-task deep learning model.

In radiation therapy (RT), accelerated partial breast irradiation (APBI) has emerged as an increasingly preferred treatment modality over conventional whole breast irradiation due to its targeted dose delivery and shorter course of treatment. APBI can be delivered through various modalities including Cobalt-60-based systems and linear accelerators with C-arm, O-ring, or robotic arm design. Each modality possesses distinct features, such as beam energy or the degrees of freedom in treatment planning, which influence their respective dose distributions. These modality-specific considerations emphasize the need for a quantitative approach in determining the optimal dose delivery modality on a patient-specific basis. However, manually generating treatment plans for each modality across every patient is time-consuming and clinically impractical. We aim to develop an efficient and personalized approach for determining the optimal RT modality for APBI by training predictive models using two different deep learning-based convolutional neural networks. The baseline network performs a single-task (ST), predicting dose for a single modality. Our proposed multi-task (MT) network, which is capable of leveraging shared information among different tasks, can concurrently predict dose distributions for various RT modalities. Utilizing patient-specific input data, such as a patient's computed tomography (CT) scan and treatment protocol dosimetric goals, the MT model predicts patient-specific dose distributions across all trained modalities. These dose distributions provide patients and clinicians quantitative insights, facilitating informed and personalized modality comparison prior to treatment planning. The dataset, comprising 28 APBI patients and their 92 treatment plans, was partitioned into training, validation, and test subsets. Eight patients were dedicated to the test subset, leaving 68 treatment plans across 20 patients to divide between the training and validation subsets. ST models were trained for each modality, and one MT model was trained to predict doses for all modalities simultaneously. Model performance was evaluated across the test dataset in terms of Mean Absolute Percent Error (MAPE). We conducted statistical analysis of model performance using the two-tailed Wilcoxon signed-rank test. Training times for five ST models ranged from 255 to 430min per modality, totaling 1925min, while the MT model required 2384min. MT model prediction required an average of 1.82s per patient, compared to ST model predictions at 0.93s per modality. The MT model yielded MAPE of 1.1033±0.3627% as opposed to the collective MAPE of 1.2386±0.3872% from ST models, and the differences were statistically significant (p=0.0003, 95% confidence interval=[-0.0865, -0.0712]). Our study highlights the potential benefits of a MT learning framework in predicting RT dose distributions across various modalities without notable compromises. This MT architecture approach offers several advantages, such as flexibility, scalability, and streamlined model management, making it an appealing solution for clinical deployment. With such a MT model, patients can make more informed treatment decisions, physicians gain more quantitative insight for pre-treatment decision-making, and clinics can better optimize resource allocation. With our proposed goal array and MT framework, we aim to expand this work to a site-agnostic dose prediction model, enhancing its generalizability and applicability.

A New ViT-Based Augmentation Framework for Wafer Map Defect Classification to Enhance the Resilience of Semiconductor Supply Chains

Wafer map defect classification plays a crucial role in sustaining the semiconductor supply chain during industrial disruptions by ensuring continuity, resilience, and efficiency while aligning with sustainability principles. Up to 30 percent of production costs is lost to chip testing and yield losses for semiconductor manufacturing supply chain. In the semiconductor practice, wafer map defects recognition plays a linchpin role in the front-end-of-line stage. Defect pattern recognition can directly pinpoint the assignable causes and provide the domain experts with actionable insights. However, various wafer defect types occur differently from each other, which makes the collected wafer map dataset to be highly imbalanced in defect classes. In the face of data imbalance, the classification model usually cannot provide satisfactory classification performance. In this aspect, this paper intends to investigate the wafer defect map classification problem by using Vision Transformer (ViT) as an alternative data augmentation approach. The primary purpose is to alleviate the class-imbalance issue and then the performance of a deep learning-based convolutional neural network for wafer defect classification can be effectively improved. The experimental results demonstrate that the proposed data augmentation method by using ViT proves to be a potential generative model for improving wafer map defect classification, particularly robust on the individual minority class. In a word, the proposed augmentation framework for wafer map defect classification facilitates a more targeted and efficient approach to quality control, resource utilization, and production optimization in maintaining a sustainable semiconductor supply chain, particularly in times of industrial disruption.

Construction of deep learning-based convolutional neural network model for automatic detection of fluid hysteroscopic endometrial micropolyps in infertile women with chronic endometritis

Objective(s)Chronic endometritis (CE) is a localized mucosal inflammatory disorder associated with female infertility of unknown etiology, endometriosis, tubal factors, repeated implantation failure, and recurrent pregnancy loss, along with atypical uterine bleeding and iron deficiency anemia. Diagnosis of CE has traditionally relied on endometrial biopsy and detection of CD138(+) endometrial stromal plasmacytes. To develop a less invasive diagnostic system for CE, we aimed to construct a deep learning-based convolutional neural network (CNN) model for the automatic detection of endometrial micropolyps (EMiP), a fluid hysteroscopy (F-HSC) finding recognized as tiny protrusive lesions that are closely related to this disease. Study designThis is an in silico study using archival images of F-HSC performed at an infertility center in a private clinic. A total of 244 infertile women undergoing F-HSC on the days 6–12 of the menstrual cycle between April 2019 and December 2021 with histopathologically-confirmed CE with the aid of immunohistochemistry for CD138 were utilized. ResultsThe archival F-HSC images of 208 women (78 with EMiP and 130 without EMiP) who met the inclusion criteria were finally subjected to analysis. Following preprocessing of the images, half a set was input into a CNN architecture for training, whereas the remaining images were utilized as the test set to evaluate the performance of the model, which was compared with that of the experienced gynecologists. The sensitivity, specificity, accuracy, precision, and F1-score of the CNN model-aided diagnosis were 93.6 %, 92.3 %, 92.8 %, 88.0 %, and 0.907, respectively. The area under the receiver operating characteristic curves of the CNN model-aided diagnosis (0.930) was at a similar level (p > .05) to the value of conventional diagnosis by three experienced gynecologists (0.927, 0.948, and 0.906). ConclusionThese findings indicate that our deep learning-based CNN is capable of recognizing EMiP in F-HSC images and holds promise for further development of the computer-aided diagnostic system for CE.

Deep learning-based CNN for multiclassification of ocular diseases using transfer learning

ABSTRACT Effective and timely diagnosis and treatment of ocular diseases is essential for swift recovery of the patients. Among ocular diseases, cataract and glaucoma are the most prevalent globally and need adequate attention. The present paper aims to develop an optimised deep learning based convolutional neural network (CNN) for the multi-classification of ocular diseases (normal, glaucoma and cataract). Three pre-trained CNNs (SqueezeNet, Darknet-53, EfficientNet-b0) were optimised concerning batch size (6/8/10) & optimiser type (SGDM, RMSProp, Adam) for obtaining maximum possible accuracy in the detection of multiple ocular diseases (cataract & glaucoma). Darknet-53 (batch size-6, optimiser type-Adam) gave the highest accuracy of 99.4% for a test sample of 1000 images. The performance metrics of Darknet-53 have been computed using a confusion matrix. Confusion matrix is also applied to calculate accuracy, sensitivity, specificity, f1 score and receiver operating curve (ROC). Through comparative performance analysis of the three CNNs, SqueezeNet, Darknet-53 and EfficientNet-b0 achieved the highest accuracy of 95%, 99.4% and 90%, respectively. The results indicate the importance of batch size and optimiser type on the performance of CNN models.

SC-Track: a robust cell-tracking algorithm for generating accurate single-cell lineages from diverse cell segmentations.

Computational analysis of fluorescent timelapse microscopy images at the single-cell level is a powerful approach to study cellular changes that dictate important cell fate decisions. Core to this approach is the need to generate reliable cell segmentations and classifications necessary for accurate quantitative analysis. Deep learning-based convolutional neural networks (CNNs) have emerged as a promising solution to these challenges. However, current CNNs are prone to produce noisy cell segmentations and classifications, which is a significant barrier to constructing accurate single-cell lineages. To address this, we developed a novel algorithm called Single Cell Track (SC-Track), which employs a hierarchical probabilistic cache cascade model based on biological observations of cell division and movement dynamics. Our results show that SC-Track performs better than a panel of publicly available cell trackers on a diverse set of cell segmentation types. This cell-tracking performance was achieved without any parameter adjustments, making SC-Track an excellent generalized algorithm that can maintain robust cell-tracking performance in varying cell segmentation qualities, cell morphological appearances and imaging conditions. Furthermore, SC-Track is equipped with a cell class correction function to improve the accuracy of cell classifications in multiclass cell segmentation time series. These features together make SC-Track a robust cell-tracking algorithm that works well with noisy cell instance segmentation and classification predictions from CNNs to generate accurate single-cell lineages and classifications.

Multi-channel microseismic signals classification with convolutional neural networks

Identifying and classifying microseismic signals is essential to warn of mines’ dangers. Deep learning has replaced traditional methods, but labor-intensive manual identification and varying deep learning outcomes pose challenges. This paper proposes a transfer learning-based convolutional neural network (CNN) method called microseismic signals-convolutional neural network (MS-CNN) to automatically recognize and classify microseismic events and blasts. The model was instructed on a limited sample of data to obtain an optimal weight model for microseismic waveform recognition and classification. A comparative analysis was performed with an existing CNN model and classical image classification models such as AlexNet, GoogLeNet, and ResNet50. The outcomes demonstrate that the MS-CNN model achieved the best recognition and classification effect (99.6% accuracy) in the shortest time (0.31 s to identify 277 images in the test set). Thus, the MS-CNN model can efficiently recognize and classify microseismic events and blasts in practical engineering applications, improving the recognition timeliness of microseismic signals and further enhancing the accuracy of event classification.

Land cover mapping of mixed classes using 2D CNN with multi-frequency SAR data

Synthetic aperture radar (SAR) data obtained at multiple frequencies and polarizations offers valuable complementary information for classifying mixed classes that exhibit similar backscattering response. Although deep learning-based convolutional neural networks (CNNs) effectively extract features from multi-frequency SAR data, the arbitrary ordering of SAR features may hinder optimal convolution of the best feature sub-space for a specific class and underutilize available multi-frequency data. To address this, a novel CNN transforming SAR feature-space from 1-D to 2-D and employing varied dilation-rate convolutions is introduced. This transformation maximizes unique and localized feature combinations, efficiently utilizing the available feature sub-spaces and extracting discriminative features for accurate classifications, addressing the challenge of arbitrary band neighborhoods. Utilizing dual-polarization SAR data from ALOS-2 PALSAR-2 and Sentinel-1 sensors, the proposed CNN achieves an average f-score of 0.97 and a kappa coefficient of 0.97, an improvement of 11 %, 7 % and 3 % in OA compared to the 1-D, 2-D and 3-D CNN classifiers, without feature transformation. The classifier's generalization ability is evaluated using ground truth knowledge of various heterogeneous classes, and the proposed CNN classifier outperforms others in terms of accuracy metrics and generalization ability.

Pattern Recognition in Embedded Systems for Event Occurrences

Abstract: PR (pattern recognition) typically includes interaction with humans and other complicated processes in the real world, embedded systems are ideal candidates. A typical PR application often considered the more perceptual branch of AI, responds to external events that the system detects through physical sensors or input devices by activating actuators or displaying relevant information. To explore the embedded recognition system and apply the deep learning algorithm to face detection, the deep learning-based Convolutional Neural Network (CNN)suggests two deep face detection methods. These are presented to use the deep learning algorithm.This was done to make it possible for us to use the deep learning algorithm for face detection. Because of this, to analyze the built-in face recognition system and applied the deep learning algorithm to the process of identifying faces. In addition, to do both things simultaneously. OMTCNN's training accuracy is 85.14%, higher than the unimproved algorithm. Accuracy of the recognition and calculation acceleration modules boosts embedded system face detection and identification performance. Embedded deep learning recognition is helpful.

Auto-segmentation of Adult-Type Diffuse Gliomas: Comparison of Transfer Learning-Based Convolutional Neural Network Model vs. Radiologists.

Segmentation of glioma is crucial for quantitative brain tumor assessment, to guide therapeutic research and clinical management, but very time-consuming. Fully automated tools for the segmentation of multi-sequence MRI are needed. We developed and pretrained a deep learning (DL) model using publicly available datasets A (n = 210) and B (n = 369) containing FLAIR, T2WI, and contrast-enhanced (CE)-T1WI. This was then fine-tuned with our institutional dataset (n = 197) containing ADC, T2WI, and CE-T1WI, manually annotated by radiologists, and split into training (n = 100) and testing (n = 97) sets. The Dice similarity coefficient (DSC) was used to compare model outputs and manual labels. A third independent radiologist assessed segmentation quality on a semi-quantitative 5-scale score. Differences in DSC between new and recurrent gliomas, and between uni or multifocal gliomas were analyzed using the Mann-Whitney test. Semi-quantitative analyses were compared using the chi-square test. We found that there was good agreement between segmentations from the fine-tuned DL model and ground truth manual segmentations (median DSC: 0.729, std-dev: 0.134). DSC was higher for newly diagnosed (0.807) than recurrent (0.698) (p < 0.001), and higher for unifocal (0.747) than multi-focal (0.613) cases (p = 0.001). Semi-quantitative scores of DL and manual segmentation were not significantly different (mean: 3.567 vs. 3.639; 93.8% vs. 97.9% scoring ≥ 3, p = 0.107). In conclusion, the proposed transfer learning DL performed similarly to human radiologists in glioma segmentation on both structural and ADC sequences. Further improvement in segmenting challenging postoperative and multifocal glioma cases is needed.

Identifying optical microscope images of CVD-grown two-dimensional MoS2 by convolutional neural networks and transfer learning.

In Complementary Metal-Oxide Semiconductor (CMOS) technology, scaling down has been a key strategy to improve chip performance and reduce power losses. However, challenges such as sub-threshold leakage and gate leakage, resulting from short-channel effects, contribute to an increase in distributed static power. Two-dimensional transition metal dichalcogenides (2D TMDs) emerge as potential solutions, serving as channel materials with steep sub-threshold swings and lower power consumption. However, the production and development of these 2-dimensional materials require some time-consuming tasks. In order to employ them in different fields, including chip technology, it is crucial to ensure that their production meets the required standards of quality and uniformity; in this context, deep learning techniques show significant potential. This research introduces a transfer learning-based deep convolutional neural network (CNN) to classify chemical vapor deposition (CVD) grown molybdenum disulfide (MoS2) flakes based on their uniformity or the occurrence of defects affecting electronic properties. Acquiring and labeling a sufficient number of microscope images for CNN training may not be realistic. To address this challenge, artificial images were generated using Fresnel equations to pre-train the CNN. Subsequently, accuracy was improved through fine-tuning with a limited set of real images. The proposed transfer learning-based CNN method significantly improved all measurement metrics with respect to the ordinary CNNs. The initial CNN, trained with limited data and without transfer learning, achieved 68% average accuracy for binary classification. Through transfer learning and artificial images, the same CNN achieved 85% average accuracy, demonstrating an average increase of approximately 17%. While this study specifically focuses on MoS2 structures, the same methodology can be extended to other 2-dimensional materials by simply incorporating their specific parameters when generating artificial images.

Classification of Progressive Wear on a Multi-Directional Pin-on-Disc Tribometer Simulating Conditions in Human Joints-UHMWPE against CoCrMo Using Acoustic Emission and Machine Learning

Human joint prostheses experience wear failure due to the complex interactions between Ultra-High-Molecular-Weight Polyethylene (UHMWPE) and Cobalt-Chromium-Molybdenum (CoCrMo). This study uses the wear classification to investigate the gradual and progressive abrasive wear mechanisms in UHMWPE. Pin-on-disc tests were conducted under simulated in vivo conditions, monitoring wear using Acoustic Emission (AE). Two Machine Learning (ML) frameworks were employed for wear classification: manual feature extraction with ML classifiers and a contrastive learning-based Convolutional Neural Network (CNN) with ML classifiers. The CNN-based feature extraction approach achieved superior classification performance (94% to 96%) compared to manual feature extraction (81% to 89%). The ML techniques enable accurate wear classification, aiding in understanding surface states and early failure detection. Real-time monitoring using AE sensors shows promise for interventions and improving prosthetic joint design.

Deep Learning Method for Heliostat Instance Segmentation

Heliostat instance segmentation (HST-IS) is a crucial component of the heliostat tracking system at Heliogen’s Lancaster test facility. The system estimates the mirror normal of each heliostat by performing a nonlinear optimization-based fitting strategy using approximations of the non-shaded, non-blocked sunlit pixels on each heliostat, and the tracking system uses these estimates to improve performance.&#x0D; HST-IS is fundamentally challenging due to variability in lighting conditions and heliostat size relative to the capturing camera. Deep learning-based convolutional neural networks (CNN) have emerged in recent years by demonstrating noteworthy precision in tasks such as object recognition, detection, and segmentation. CNN-based methods offer a robust augmentation to HST-IS methods as they capture a context-less hierarchy of image features.&#x0D; In this study, we developed deep learning models to automatically segment heliostat instances from elevated images taken from the field. We study various image parameters and architectural customizations to optimize for scalability, robustness, and accuracy in our predictions. We perform robust evaluations of our best model to quantify gaps between model development and real-world deployment and provide evidence for utility in the field.

A comparative analysis of CNN-based deep learning architectures for early diagnosis of bone cancer using CT images

Bone cancer is a rare in which cells in the bone grow out of control, resulting in destroying the normal bone tissue. A benign type of bone cancer is harmless and does not spread to other body parts, whereas a malignant type can spread to other body parts and might be harmful. According to Cancer Research UK (2021), the survival rate for patients with bone cancer is 40% and early detection can increase the chances of survival by providing treatment at the initial stages. Prior detection of these lumps or masses can reduce the risk of death and treat bone cancer early. The goal of this current study is to utilize image processing techniques and deep learning-based Convolution neural network (CNN) to classify normal and cancerous bone images. Medical image processing techniques, like pre-processing (e.g., median filter), K-means clustering segmentation, and, canny edge detection were used to detect the cancer region in Computer Tomography (CT) images for parosteal osteosarcoma, enchondroma and osteochondroma types of bone cancer. After segmentation, the normal and cancerous affected images were classified using various existing CNN-based models. The results revealed that AlexNet model showed a better performance with a training accuracy of 98%, validation accuracy of 98%, and testing accuracy of 100%.

Efficient Diagnosis of Bacterial Leaf Spot in Tomato Plants using Deep Learning CNN Models

Latest field surveys show that the bacterial spot caused by Xanthomonas campestris pv. vesicatoria (Xcv) has devastated commercial tomato output all over the world. The purpose of this study is to more quickly and accurately diagnose bacterial leaf spot problems in tomato plants. To do this, a deep learning-based Convolutional Neural Network (CNN) named Xantho_Net has been proposed to classify the bacterial and non-bacterial tomato leafs. Various state-of-art pre-trained models i.e., VGG16, MobileNet, DenseNet_201, Inception-ResNet_v2 were selected for comparison purposes. Experiments proved that the proposed CNN model is more accurate than other models with 99% accuracy.

Integration of harvester trajectory and satellite imagery for large-scale winter wheat mapping using deep positive and unlabeled learning

Limited accurate ground truth labels are the primary constraint for data-driven modeling analysis of large-scale crop mapping. Existing labeling methods largely rely on field surveys, visual interpretation, and historical ground information. These labor-intensive approaches are often limited by spatiotemporal heterogeneity of crop distribution and encounter the challenge of gathering extensive crop labels. The massive operating trajectories of agricultural machinery contain precise location information of the crop fields, providing a new source for accurate crop labels at a large spatial scale. This study develops a large-scale crop mapping workflow through widespread harvester trajectory and 10 m Sentinel-2 imagery. The trajectory-based automatic labeling method is developed to generate 287,533 winter wheat labels by jointly using harvester coordinates and satellite images. These generated one-class ground labels are further used to develop positive and unlabeled learning based deep learning models for winter wheat mapping. The Positive and Unlabeled Learning-based Convolutional Neural Network (PUL-CNN) outperforms the other four one-class based classifiers with an F1 score of 94.4 % at 12 study sites. The estimated county-level winter wheat acreage agrees well with census data with R2 of 0.86 in the overall study region. The interpretation analysis based on the Shapley Additive Explanation method shows the heading and greening stages are the critical periods for wheat mapping, aligning well with the separability in Normalized Difference Vegetation Index (NDVI) curves. The results of winter wheat mapping demonstrate the integration of harvester trajectory and remote sensing data facilitates large-scale winter wheat mapping. To the best of our knowledge, this is the first study that fuses operating trajectories of agricultural machinery and satellite images for large-scale crop mapping based on the deep positive and unlabeled learning approach. This study could be possibly applied for better understanding the land cover and land use changes.

Detection of the separated endodontic instrument on periapical radiographs using a deep learning-based convolutional neural network algorithm.

The study evaluated the diagnostic performance of an artificial intelligence system to detect separated endodontic instruments on periapical radiograph radiographs. Three hundred seven periapical radiographs were collected and divided into 222 for training and 85 for testing to be fed to the Mask R-CNN model. Periapical radiographs were assigned to the training and test set and labelled on the DentiAssist labeling platform. Labelled polygonal objects had their bounding boxes automatically generated by the DentiAssist system. Fractured instruments were classified and segmented. As a result of the proposed method, the mean average precision (mAP) metric was 98.809%, the precision value was 95.238, while the recall reached 98.765 and the f1 score 96.969%. The threshold value of 80% was chosen for the bounding boxes working with the Intersection over Union (IoU) technique. The Mask R-CNN distinguished separated endodontic instruments on periapical radiographs.

Geographical discrimination of Asian red pepper powders using 1H NMR spectroscopy and deep learning-based convolution neural networks

This study investigated an innovative approach to discriminate the geographical origins of Asian red pepper powders by analyzing one-dimensional 1H NMR spectra through a deep learning-based convolution neural network (CNN). 1H NMR spectra were collected from 300 samples originating from China, Korea, and Vietnam and used as input data. Principal component analysis − linear discriminant analysis and support vector machine models were employed for comparison. Bayesian optimization was used for hyperparameter optimization, and cross-validation was performed to prevent overfitting. As a result, all three models discriminated the origins of the test samples with over 95 % accuracy. Specifically, the CNN models achieved a 100 % accuracy rate. Gradient-weighted class activation mapping analysis verified that the CNN models recognized the origins of the samples based on variations in metabolite distributions. This research demonstrated the potential of deep learning-based classification of 1H NMR spectra as an accurate and reliable approach for determining the geographical origins of various foods.

Dental bitewing radiographs segmentation using deep learning-based convolutional neural network algorithms.

Dental radiographs, particularly bitewing radiographs, are widely used in dental diagnosis and treatment Dentalimage segmentation is difficult for various reasons, such as intricate structures, low contrast, noise, roughness, andunclear borders, resulting in poor image quality. Recent developments in deep learning models have improvedperformance in analyzing dental images. In this research, our primary objective is to determine the most effectivesegmentation technique for bitewing radiographs based on different metrics: accuracy, training time, and thenumber of training parameters as a reflection of architectural cost. In this research, we employ several deep learning models, namely Resnet-18, Resnet-50, Xception, InceptionResnet v2, and Mobilenetv2, to segment bitewing radiographs. The process begins by importing the radiographsinto MATLAB®(MathWorks Inc), where the images are first improved, then segmented using the graph cut methodbased on regions to produce a binary mask that distinguishes the background from the original X-ray. The deep learning models were trained on 298 and 99 radiograph training and validation sets and were evaluatedusing 99 images from the testing set. We also compare the segmentation model using several criteria, includingaccuracy, speed, and size, to determine which network is superior. Furthermore, we compare our findings with priorresearch to provide a comprehensive understanding of the advancements made in dental image segmentation. Theaccurate segmentation achieved was 93.67% and 94.42% by the Resnet-18 and Resnet-50 models, respectively. This research advances dental image analysis and facilitates more accurate diagnoses and treatment planning bydetermining the best segmentation technique. The outcomes of this study can guide researchers and practitionersin selecting appropriate segmentation methods for practical dental image analysis.

COMPARISON OF VISION TRANSFORMERS AND CONVOLUTIONAL NEURAL NETWORKS FOR SKIN DISEASE CLASSIFICATION

Skin diseases are one of the most common diseases in humans. Due to its many various symptoms and types, computer vision studies have been frequently applied to its diagnosis and classification. Previous studies have frequently used machine learning methods and deep learning-based Convolutional Neural Networks (CNN) for skin disease diagnosis. Although deep learning-based applications have achieved great success in terms of detection accuracy, research continues to ensure the desired performance. However, Vision Transformer (ViT), recently proposed as a competitive alternative to CNNs, is gaining increasing popularity. This paper compares ResNet18 and ResNet50 networks, important CNN models, with ViT for classifying skin disease. The comparison is applied on a dataset containing a small number of samples. In the application performed on a dataset containing skin disease images, ViT provides 68.93% classification accuracy, while ResNet18 and ResNet50 classification accuracy is 61.65% and 61.17%, respectively. Other metrics calculated along with the accuracies also prove the superiority of ViT over ResNet models. However, ViT has a big disadvantage in terms of training time.