Computer-aided Diagnosis Research Articles

Training-Free Image Style Alignment for Domain Shift on Handheld Ultrasound Devices.

Handheld ultrasound devices face usage limitations due to user inexperience and cannot benefit from supervised deep learning without extensive expert annotations. Moreover, the models trained on standard ultrasound device data are constrained by training data distribution and perform poorly when directly applied to handheld device data. In this study, we propose the Training-free Image Style Alignment (TISA) to align the style of handheld device data to those of standard devices. The proposed TISA eliminates the demand for source data, and can transform the image style while preserving spatial context during testing. Furthermore, our TISA avoids continuous updates to the pre-trained model compared to other test-time methods and is suited for clinical applications. We show that TISA performs better and more stably in medical detection and segmentation tasks for handheld device data than other test-time adaptation methods. We further validate TISA as the clinical model for automatic measurements of spinal curvature and carotid intima-media thickness, and the automatic measurements agree well with manual measurements made by human experts. We demonstrate the potential for TISA to facilitate automatic diagnosis on handheld ultrasound devices and expedite their eventual widespread use. Code is available at https://github.com/zenghy96/TISA.

Clinician-artificial intelligence collaboration: A win-win solution for efficiency and reliability in atrial fibrillation diagnosis.

Clinician-artificial intelligence collaboration: A win-win solution for efficiency and reliability in atrial fibrillation diagnosis.

Computer-Aided Diagnosis Models for Breast Cancer Detection Decision Support Systems

Computer-aided diagnosis (CADx) technology has demonstrated enhanced efficacy in breast cancer (BC) detection, which is particularly crucial during primary care physicians' initial evaluation of patients. The work aims to develop novel, user-friendly auxiliary computer aids accessible directly to frontline medical practitioners, eliminating the need for costly computer systems. The scientific novelty lies in the fact that we have devised a non-relational database (DB) of factual data designed to house the results of patient studies, which can be harnessed for machine learning in computer-aided BC diagnosis systems. The DB encompasses a heterogeneous vector of primary measurements (metadata, DICOM standard files, alongside other images and data) for each patient, facilitating the construction of a neural network for tumor recognition and preliminary classification. We have populated the database with new, region-specific data pertinent to women in Ukraine amidst severe stress induced by the ongoing war. Additionally, we have developed a new system for concurrent monitoring of ultrasound, computed tomography, and mammography results, complemented by a decision support system for simultaneous cross-verification of neoplasm diagnoses based on density and spatial correlation.

A novel network-level fused deep learning architecture with shallow neural network classifier for gastrointestinal cancer classification from wireless capsule endoscopy images

Deep learning has significantly contributed to medical imaging and computer-aided diagnosis (CAD), providing accurate disease classification and diagnosis. However, challenges such as inter- and intra-class similarities, class imbalance, and computational inefficiencies due to numerous hyperparameters persist. This study aims to address these challenges by presenting a novel deep-learning framework for classifying and localizing gastrointestinal (GI) diseases from wireless capsule endoscopy (WCE) images. The proposed framework begins with dataset augmentation to enhance training robustness. Two novel architectures, Sparse Convolutional DenseNet201 with Self-Attention (SC-DSAN) and CNN-GRU, are fused at the network level using a depth concatenation layer, avoiding the computational costs of feature-level fusion. Bayesian Optimization (BO) is employed for dynamic hyperparameter tuning, and an Entropy-controlled Marine Predators Algorithm (EMPA) selects optimal features. These features are classified using a Shallow Wide Neural Network (SWNN) and traditional classifiers. Experimental evaluations on the Kvasir-V1 and Kvasir-V2 datasets demonstrate superior performance, achieving accuracies of 99.60% and 95.10%, respectively. The proposed framework offers improved accuracy, precision, and computational efficiency compared to state-of-the-art models. The proposed framework addresses key challenges in GI disease diagnosis, demonstrating its potential for accurate and efficient clinical applications. Future work will explore its adaptability to additional datasets and optimize its computational complexity for broader deployment.

Melanoma Skin Cancer Detection with the Integration of a Conversational Chatbot

Skin cancer, specifically melanoma, results from abnormal melanocytic cell growth and can be fatal. It typically appears as dark lesions due to UV exposure and genetic factors. Early detection is crucial for treatment. The conventional method, biopsy, is invasive, painful, and slow, as it requires lab analysis. To address these issues, a non-invasive computer-aided diagnosis (CAD) system is proposed, using dermoscopy images. This system preprocesses the images, segments the lesion, extracts unique features, and then classifies the skin as normal or cancerous using a support vector machine (SVM). The SVM with a linear kernel demonstrates optimal accuracy. CAD eliminates the need for physical contact, reducing pain and improving efficiency in melanoma detection through advanced image processing techniques.

Automatic Detection of Genetic Diseases in Pediatric Age Using Pupillometry

Automatic Detection of Genetic Diseases in Pediatric Age Using Pupillometry

Pinpoint the Aliment of Skin Disease With AI

Abstract: Dermatology is the field of science responsible for diagnosis and management of skin disorders. The dermatological disorders are wide spread and vary geographically, and usually due to the change in temperature, humidity and other environmental factors. Human skin presents one of the most complex systems to mechanically image and analyze due to its unevenness, color, presence of hair and other soothing features. Although, numerous researches are being done to find out and prove human skin As exploited (Computer Vision techniques), very few concentrated around the medical paradigm of the issue. Owing to lack of medical facilities available in the remote regions, patients often ignore early warning signs which could later worsen the situation over time. Thus, there is always a growing demand for automatic skin disease detection system with high accuracy of diagnosis. Therefore, we propose a system for automatic detection and classification of skin diseases for building a multi-class deep learning model to tell the difference between Healthy Skin and skin suffering from a disease and categorization of skin diseases into the following groups: Melanocytic Nevi, Melanoma, Keratosis like lesions, Basal cell Carcinoma, Actinic Keratoses, Vascular tumor, and Dermatofibroma. We have used Profound Learning to prepare our show. Profound Learning could be a subset of Machine Learning which uses a much bigger dataset than the traditional approach, thus the number of classifiers is reduced significantly. The machine is capable of learning all by itself, it sorts the provided data into different levels of prediction and very quickly, gives the precise results, thus helping and fostering the development of Dermatology. The algorithm that we have used is Convolutional Neural Arrange (CNN) because it is one of the foremost preferred calculation for picture classification

Disease Recognition Using X-Ray Plates Using Deep Learning

Chest X-rays are essential in medical imaging, serving as a reliable diagnostic tool for various thoracic diseases. However, despite their critical role in healthcare, a vast amount of imaging data remains underutilized within Picture Archiving and Communication Systems (PACS) in hospitals and medical institutions. These stored images, along with their associated diagnoses, hold immense potential for training deep learning models, which require large datasets to enhance automated disease detection. This project aims to bridge that gap by utilizing the Chest Xray 8 dataset, a large-scale collection of labelled chest X-ray images covering multiple diseases, including pneumonia, tuberculosis, and COVID-19. By integrating this dataset with a deep learning models, Custom CNN, VGG19, ResNet50, DenseNet121, MobileNet, we aim to develop an AI-driven diagnostic model capable of identifying and classifying chest diseases with good accuracy. This approach can significantly enhance early detection, assist radiologists in decision-making, and improve healthcare accessibility, especially in regions with limited medical expertise. The project represents a step toward harnessing AI's power to optimize medical imaging, automate diagnostics, and revolutionize disease detection in clinical settings

Automatic Blob Detection Method for Cancerous Lesions in Unsupervised Breast Histology Images.

The early detection of cancerous lesions is a challenging task given the cancer biology and the variability in tissue characteristics, thus rendering medical image analysis tedious and time-inefficient. In the past, conventional computer-aided diagnosis (CAD) and detection methods have heavily relied on the visual inspection of medical images, which is ineffective, particularly for large and visible cancerous lesions in such images. Additionally, conventional methods face challenges in analyzing objects in large images due to overlapping/intersecting objects and the inability to resolve their image boundaries/edges. Nevertheless, the early detection of breast cancer lesions is a key determinant for diagnosis and treatment. In this study, we present a deep learning-based technique for breast cancer lesion detection, namely blob detection, which automatically detects hidden and inaccessible cancerous lesions in unsupervised human breast histology images. Initially, this approach prepares and pre-processes data through various augmentation methods to increase the dataset size. Secondly, a stain normalization technique is applied to the augmented images to separate nucleus features from tissue structures. Thirdly, morphology operation techniques, namely erosion, dilation, opening, and a distance transform, are used to enhance the images by highlighting foreground and background pixels while removing overlapping regions from the highlighted nucleus objects in the image. Subsequently, image segmentation is handled via the connected components method, which groups highlighted pixel components with similar intensity values and assigns them to their relevant labeled components (binary masks). These binary masks are then used in the active contours method for further segmentation by highlighting the boundaries/edges of ROIs. Finally, a deep learning recurrent neural network (RNN) model automatically detects and extracts cancerous lesions and their edges from the histology images via the blob detection method. This proposed approach utilizes the capabilities of both the connected components method and the active contours method to resolve the limitations of blob detection. This detection method is evaluated on 27,249 unsupervised, augmented human breast cancer histology dataset images, and it shows a significant evaluation result in the form of a 98.82% F1 accuracy score.

Melanoma Skin Cancer Detection with the Integration of a Conversational Chatbot

Skin cancer, specifically melanoma, results from abnormal melanocytic cell growth and can be fatal. It typically appears as dark lesions due to UV exposure and genetic factors. Early detection is crucial fortreatment. The conventional method, biopsy, is invasive, painful, and slow, as it requires lab analysis. To address these issues, a non- invasive computer-aided diagnosis (CAD) system is proposed, using dermoscopy images. This system preprocesses the images, segments the lesion, extracts unique features, and then classifies the skin as normal or cancerous using a support vector machine (SVM). The SVM with a linear kernel demonstrates optimal accuracy. CAD eliminates the need for physical contact, reducing pain and improving efficiency in melanoma detection through advanced image processing techniques.

Application of the efficientnet model for detecting ischemic heart disease

This article examines a neural network model that facilitates the early detection of is-chemic heart disease on chest X-rays, playing a crucial role in the diagnosis and treatment of cardiovascular diseases. The study explores the application of convolutional neural net-works(CNNs) for the automatic detection of ischemic heart disease on chest X-rays. A dataset containing chest X-rays annotated by an experienced cardiologist was used in the study. A CNN model was developed and trained to detect signs of ischemic heart disease based on chest X-rays. The model was trained on a small dataset (700 images) and tested on an independent set of test images. The research results demonstrated that the convolutional neural network effectively recognizes signs of ischemic heart disease on chest X-rays with high accuracy and reliability. This could significantly improve the capabilities for early diagnosis of ischemic heart disease and help clinicians make informed decisions regarding patient treatment. The findings of this study confirm the potential of using convolutional neural networks in medical diagnostics and represent a new step forward in the automated detection of is-chemic heart disease based on chest X-rays. This paves the way for improving the accuracy and efficiency of cardiovascular disease diagnostics and reducing the burden on medical per-sonnel.

Deep Learning Based on Ultrasound Images Differentiates Parotid Gland Pleomorphic Adenomas and Warthin Tumors.

Exploring the clinical significance of employing deep learning methodologies on ultrasound images for the development of an automated model to accurately identify pleomorphic adenomas and Warthin tumors in salivary glands. A retrospective study was conducted on 91 patients who underwent ultrasonography examinations between January 2016 and December 2023 and were subsequently diagnosed with pleomorphic adenoma or Warthin's tumor based on postoperative pathological findings. A total of 526 ultrasonography images were collected for analysis. Convolutional neural network (CNN) models, including ResNet18, MobileNetV3Small, and InceptionV3, were trained and validated using these images for the differentiation of pleomorphic adenoma and Warthin's tumor. Performance evaluation metrics such as receiver operating characteristic (ROC) curves, area under the curve (AUC), sensitivity, specificity, positive predictive value, and negative predictive value were utilized. Two ultrasound physicians, with varying levels of expertise, conducted independent evaluations of the ultrasound images. Subsequently, a comparative analysis was performed between the diagnostic outcomes of the ultrasound physicians and the results obtained from the best-performing model. Inter-rater agreement between routine ultrasonography interpretation by the two expert ultrasonographers and the automatic identification diagnosis of the best model in relation to pathological results was assessed using kappa tests. The deep learning models achieved favorable performance in differentiating pleomorphic adenoma from Warthin's tumor. The ResNet18, MobileNetV3Small, and InceptionV3 models exhibited diagnostic accuracies of 82.4% (AUC: 0.932), 87.0% (AUC: 0.946), and 77.8% (AUC: 0.811), respectively. Among these models, MobileNetV3Small demonstrated the highest performance. The experienced ultrasonographer achieved a diagnostic accuracy of 73.5%, with sensitivity, specificity, positive predictive value, and negative predictive value of 73.7%, 73.3%, 77.8%, and 68.8%, respectively. The less-experienced ultrasonographer achieved a diagnostic accuracy of 69.0%, with sensitivity, specificity, positive predictive value, and negative predictive value of 66.7%, 71.4%, 71.4%, and 66.7%, respectively. The kappa test revealed strong consistency between the best-performing deep learning model and postoperative pathological diagnoses (kappa value: .778, p-value < .001). In contrast, the less-experienced ultrasonographer demonstrated poor consistency in image interpretations (kappa value: .380, p-value < .05). The diagnostic accuracy of the best deep learning model was significantly higher than that of the ultrasonographers, and the experienced ultrasonographer exhibited higher diagnostic accuracy than the less-experienced one. This study demonstrates the promising performance of a deep learning-based method utilizing ultrasonography images for the differentiation of pleomorphic adenoma and Warthin's tumor. The approach reduces subjective errors, provides decision support for clinicians, and improves diagnostic consistency.

Brain tumor classification: a blend of ensemble learning and fine-tuned pre-trained models

In medicinal Computer Assisted Diagnosis (CAD) systems, the automatic classification of brain tumors plays a substantial role, as misclassifications can have considerable implications for a patient’s chances of survival. To enhance the performance of classification analysis, we suggested a brain tumor classification model based on Convolutional Neural Network (CNN) pre-trained models by including some additional layers for feature extraction with three diverse activation functions (Rectified Linear Unit (ReLU), Parametric Rectified Linear Unit (PReLU), and Swish).We analyzed seven pre-trained models such as VGG19, InceptionV3, ResNet50V2, InceptionResNetV2, DenseNet201, MobileNetV2, and EfficientNetB7, with additional layers for feature extraction. In order to get more precise outcomes and consistent results, we designed an ensemble algorithm using a majority voting scheme.We trained and tested our proposed architecture using the ’Brain MRI Images for Brain Tumor Detection’ dataset. Our model attained a 99.34% classification accuracy on the Brain Magnetic Resonance Imaging (MRI) images for Brain Tumor Detection dataset and an Area Under Curve (AUC), Precision, Recall, and F1-score of 0.9841, 1.0, 0.9843 and 0.9921 respectively. These results illustrate the suggested system’s efficiency in enhancing the classification rate and reducing the misclassification rate. It also demonstrates that the proposed model is a promising method for automatically classifying brain tumors. The increased accuracy and performance metrics suggest its potential usefulness in real-world medical applications, ultimately leading to enhanced patient outcomes.

3D segmentation combining spatial and multi-scale features for intracranial aneurysm.

Traditionally, the diagnosis of intracranial aneurysms has relied on the experience of the doctor in assessing the scanning results of radiological imaging technology, which is subjective and inefficient, and it is also limited by the physical strength and energy of the doctor. In order to improve the diagnostic efficiency of doctors and reduce the rate of misdiagnosis and missed diagnosis as much as possible. We propose a 3D segmentation network, SMNet, based on the U-Net architecture that combines spatial and multi-scale features. The network can better solve the problem of intracranial aneurysm segmentation on magnetic resonance angiography (MRA) scanning sequences. Specifically, semantic information of different dimensions is extracted at each stage of the encoder by the multi-scale feature extraction block (MSE Block) and the strip volumetric pooling block (SVP Block), respectively. Then, after the fusion of adjacent scale features extracted by the decoder, the weight of features is further redistributed by the quaternary spatial attention block (QSA Block). While focusing on the important features, the ability to discriminate different foregrounds is improved. Experiments show that the proposed three modules improve the segmentation performance to different degrees. Dice and MIoU have increased by 16.7% and 28% compared to the baseline in the private dataset, and the results of the Aneurysm Detection And segMentation (ADAM) public dataset are 0.482 and 0.389, respectively. It has shown better segmentation quality than 3D medical image segmentation mainstream models. Our model greatly improves the segmentation results of intracranial aneurysms with MRA images, and makes a certain contribution to the clinical intervention of computer-assisted diagnosis and treatment in this field.

A New Pes Planus Automatic Diagnosis Method: ViT-OELM Hybrid Modeling.

Background/Objectives: Pes planus (flat feet) is a condition characterized by flatter than normal soles of the foot. In this study, a Vision Transformer (ViT)-based deep learning architecture is proposed to automate the diagnosis of pes planus. The model analyzes foot images and classifies them into two classes, as "pes planus" and "not pes planus". In the literature, models based on Convolutional neural networks (CNNs) can automatically perform such classification, regression, and prediction processes, but these models cannot capture long-term addictions and general conditions. Methods: In this study, the pes planus dataset, which is openly available on the Kaggle database, was used. This paper suggests a ViT-OELM hybrid model for automatic diagnosis from the obtained pes planus images. The suggested ViT-OELM hybrid model includes an attention mechanism for feature extraction from the pes planus images. A total of 1000 features obtained for each sample image from this attention mechanism are used as inputs for an Optimum Extreme Learning Machine (OELM) classifier using various activation functions, and are classified. Results: In this study, the performance of this suggested ViT-OELM hybrid model is compared with some other studies, which used the same pes planus database. These comparison results are given. The suggested ViT-OELM hybrid model was trained for binary classification. The performance metrics were computed in testing phase. The model showed 98.04% accuracy, 98.04% recall, 98.05% precision, and an F-1 score of 98.03%. Conclusions: Our suggested ViT-OELM hybrid model demonstrates superior performance compared to those of other studies, which used the same dataset, in the literature.

A Novel Framework for Detection of Facial Paralysis Using Cascaded Convolutional Neural Networks

Early detection and accurate diagnosis of facial paralysis are vital because of timely medical treatment and improved patient outcomes. Traditional diagnostic techniques are based on subjective evaluations, thus leading to unnecessary delays in diagnosis. This work attempts to solve this challenge by introducing a cascaded convolutional neural network (CNN) for the automatic diagnosis of facial paralysis signs from recorded facial images in real-time. Our proposed system uses advanced image preprocessing and feature extraction techniques to classify facial paralysis symptoms with great accuracy. The model was trained on a dataset composed of diverse facial expressions; it achieved a training accuracy of 98% and a testing accuracy of 99.86%. The cascaded CNN architecture is capable of detecting very effectively by combining many feature layers for correct classification. This system has enormous applicability in real-time telemedicine, remote diagnostics, and in continuous monitoring of patients. Thus, the project will tackle a relevant gap between advanced machine learning technology and health by providing a more scalable and efficient solution, accessible to many.

Use of computer-assisted detection (CADe) colonoscopy in colorectal cancer screening and surveillance: European Society of Gastrointestinal Endoscopy (ESGE) Position Statement.

This statement conveys the European Society of Gastrointestinal Endoscopy (ESGE) position on the use of computer-aided detection (CADe) with artificial intelligence (AI) during colonoscopy for colorectal cancer (CRC) screening or surveillance. The ESGE position is informed by the BMJ Rapid Recommendation initiative and the approach of the MAGIC Evidence Ecosystem Foundation; these include systematic reviews of currently available evidence, supplemented by microsimulation modeling and patient values and preferences, for the benefits and harms of AI CADe devices during colonoscopy.ESGE convened a panel of European experts for this Position Statement. On December 18, 2024, panel members voted on their preferred recommendation between two choices about CADe during colonoscopy for indications of CRC screening or polyp surveillance. Out of 19 eligible votes, 13 (68.4%) voted to recommend CADe for colonoscopy, and six panel members (31.6%) voted against. Therefore, the current ESGE statement is: RECOMMENDATION: The panel believes that most well-informed patients who have already decided to undergo colonoscopy for screening or surveillance would favor CADe assistance during colonoscopy. This is due to the potential benefits, although limited, of reduction in colorectal cancer incidence and mortality.This recommendation is weak, because the evidence is limited with considerable uncertainty of the evidence estimates, the absolute benefits for colorectal cancer incidence and mortality are small, and there is a patient burden associated with CADe (more polyp overdiagnosis and more colonoscopy surveillance).

An Effective and Fast Model for Characterization of Cardiac Arrhythmia and Congestive Heart Failure.

Background/Objectives: Cardiac arrhythmia (ARR) and congestive heart failure (CHF) are heart diseases that can cause dysfunction of other body organs and possibly death. This paper describes a fast and accurate detection system to distinguish between ARR and normal sinus (NS), and between CHF and NS. Methods: the proposed automatic detection system uses the higher amplitude coefficients (HAC) of the discrete cosine transform (DCT) of the electrocardiogram (ECG) as discriminant features to distinguish ARR and CHF signals from NS. The approach is tested with three statistical classifiers, of which the k-nearest neighbors (k-NN) algorithm. Results: the DCT provides fast compression of the ECG signal, and statistical tests show that the obtained HACs are different from ARR and NS, and for CHF and NS. The latter achieved highest accuracy under ten-fold cross-validation in comparison to Naïve Bayes (NB) and nonlinear support vector machine (SVM). The kNN yielded 97% accuracy, 99% sensitivity, 90% specificity and 0.63 s processing time when classifying ARR against NS, and it yielded 99% accuracy, 99.7% sensitivity, and 99.2% specificity, and 0.27 seconds processing time when classifying HCF against NS. In addition to a fast response, the DCT-kNN system yields higher accuracy in comparison to recent works. Conclusions: it is concluded that using the DCT based HACs as biomarkers of ARR and CHF can lead an efficient computer aided diagnosis (CAD) system which is fast, accurate and does not require ECG signal pre-processing and segmentation. The proposed system is promising for applications in clinical milieu.

Feasibility study of AI-assisted multi-parameter MRI diagnosis of prostate cancer

Distinguishing between benign and malignant prostate lesions in magnetic resonance imaging (MRI) poses challenges that affect prostate cancer screening accuracy. We propose a novel computer-aided diagnosis (CAD) system to extract cancerous lesions from the prostate in multi-parametric MRI (mp-MRI), assessing the feasibility of using artificial intelligence for detecting clinically significant prostate cancer (PCa). A retrospective study was conducted on 106 patients who underwent mp-MRI from 2021 to 2024 at a single center. We analyzed three sequences (T2W, DCE, and DWI) and collected 137 mp-MRI images corresponding to pathological sections. From these, we obtained 274 sets of ROI data, using 206 for training and validation, and 68 for testing. A feature extractor was developed using a pre-trained ResNet50 model combined with a multi-head attention mechanism to fuse modality-specific features and perform classification. The experimental results indicate that our model demonstrates high classification performance, achieving an AUC of 0.89. The PR curve shows high precision across most recall values, with an AUC of 0.91. We developed a novel CAD system based on deep learning ResNet50 models to assess the risk of prostate malignancy in mpMRI images. High classification ability is achieved, and combining the attention mechanism or optimization strategy can improve the performance of the model in medical imaging.

Computer-Aided Diagnosis System for Classifying the Liver Lesions from Multiphase CT Images

Background and ObjectiveLiver cancer is one of the lethal cancers, with a high mortality rate. This study aims to develop a Computer-Aided Diagnosis (CADx) system for identifying benign, malignant and metastatic tumors using multi-phase 3D Computed Tomography (CT) data.Materials and MethodsThe proposed study uses 601 retrospective cases from an internal institutional database consisting of benign (n = 208), malignant (n = 200) and metastases (n = 193) and 105 Hepatocellular Carcinoma (HCC) cases from a public dataset. The liver is segmented automatically using a Deep Learning (DL) model based on SegNet and atrous spatial pyramid pooling module. Features are extracted from the segmented liver volume using histogram, texture, wavelet and DL methods for characterizing the three categories. The relevant features are then fed to the standard classifiers for comparative analyses.ResultsThe proposed DL-based liver segmentation method performed better than the standard DL methods. Support vector machine gave the best results for both test sets among the classifiers. The average classification accuracies achieved were 80 and 81.9% for the internal and public datasets.ConclusionThe proposed CADx system has good clinical potential in distinguishing liver lesions from multi-phase CT images. The promising results obtained for internal and public datasets prove the model’s generalizability.