Explainable deep learning models for diagnosing lumbar spine disorders from medical imaging

Explainable deep learning diagnostic system for prediction of lung disease from medical images

Comprehensive Study for Breast Cancer Using Deep Learning and Traditional Machine Learning

Chest diseases, including COVID-19, have caused a global pandemic, resulting in a large number of deaths. In some countries, the medical system has become overwhelmed with a shortage of doctors and medical supplies, making it difficult to accommodate all patients. Deep learning has been widely used to offer smart solutions using medical images, such as chest X-rays (CXR) to identify the disease. However, it has been noticed that the majority of current studies have been based on relatively small datasets of medical images, due to the recent emergence of chest disease and the ongoing process of gathering and publishing corresponding datasets. This limited number of COVID-19 medical images may be insufficient to build robust and accurate deep learning models. To address this problem, this study proposed GANCHEST, a framework that generates Chest CXR images based on two different generative adversarial networks (GANs): the basic GAN (GAN) and the conditional GAN (CGAN). The generated images are then validated automatically through four deep transfer learning models, namely GoogleNet, InceptionV3, SqueezNet, and VGG16. The two GAN architectures' images serve as the basis for training the models, and a test set of actual chest images serves to evaluate their performance. According to the experiments, the CGAN outperformed the GAN in creating images that were more similar to the original images. Specifically, the highest classification accuracy of the CGAN achieved 92.47\\% with the VGG16 model, while the highest classification accuracy of the GAN achieved 69.27\\%. The GANCHEST framework, as proposed in this study, can have a wider range of applications beyond chest CXR images. It can be applied to other domains that lack datasets, such as other medical imaging modalities, where the same problem of limited dataset availability exists. The framework can be adapted to generate synthetic images that can be used to augment existing datasets and improve the performance of deep learning models. Additionally, the proposed approach of using deep transfer learning models for validation can also be applied to other fields where the need for efficient and accurate image validation arises. The GANCHEST framework is a novel approach that can be useful for several fields where the lack of datasets is a bottleneck for the performance of machine learning models.

Focus issue: Artificial intelligence in medical physics.

BackgroundOsteoporosis (OP) is projected to be a major issue significantly impacting the well-being of middle-aged and old populations. Machine learning (ML) and deep learning (DL) models developed based on medical imaging have enhanced clinicians’ diagnostic accuracy and work efficiency. However, the diagnostic performance of different types of medical imaging for OP has not been systematically assessed.ObjectiveBy summarizing related literature, this study aims to elucidate the role of DL models based on different medical imaging modalities in OP detection.MethodsPubMed, Embase, the Cochrane Library, and Web of Science were systematically searched for studies using ML for the diagnosis of OP based on medical imaging. The final search was conducted on May 16, 2024. The risk of bias in the included studies was assessed using the Quality Assessment of Diagnostic Accuracy Studies-2 tool. A bivariate mixed-effects model was applied to perform meta-analyses of sensitivity (SEN) and specificity (SPC), stratified by imaging modality (x-ray, computed tomography [CT], magnetic resonance imaging [MRI]). In addition, subgroup analyses were carried out based on the type of ML algorithm, the method of validation dataset generation, and the anatomical site of assessment.ResultsA total of 60 studies comprising 66,195 participants were encompassed in this systematic review and meta-analysis. Among these, 22 studies used x-ray imaging, 37 applied CT imaging, and 3 used MRI for ML-based OP diagnosis. For x-ray–based models, the pooled SEN and SPC for studies focusing on the appendicular skeleton were 0.97 (95% CI 0.83‐0.99) and 0.90 (95% CI 0.75‐0.96), respectively. For studies using the mandible as the target site, SEN and SPC were 0.94 (95% CI 0.89‐0.97) and 0.80 (95% CI 0.56‐0.93), respectively. For those focusing on the lumbar spine, the pooled SEN and SPC were 0.87 (95% CI 0.77‐0.93) and 0.82 (95% CI 0.75‐0.87), respectively. For CT-based models, studies targeting the hip joint reported a pooled SEN and SPC of 0.87 (95% CI 0.83‐0.90) and 0.92 (95% CI 0.81‐0.96), respectively. For the thoracic spine, SEN and SPC were 0.91 (95% CI 0.86‐0.94) and 0.94 (95% CI 0.92‐0.95), respectively, while for the lumbar spine, they were 0.91 (95% CI 0.87‐0.94) and 0.92 (95% CI 0.86‐0.95), respectively.ConclusionsML based on medical imaging demonstrates high diagnosis accuracy for OP, particularly DL models using x-ray and CT modalities. However, this study included only a limited number of original studies using MRI-based ML, and there remains a lack of adequate external validation across studies, which poses interpretative limitations. Future research should aim to develop artificial intelligence tools with broader applicability and enhanced diagnostic precision.

Artificial intelligence (AI) and machine learning (ML), especially deep learning, have generated tremendous impacts throughout our society, including the tomographic medical imaging field. In contrast to computer vision and image analysis, which have been major application examples of deep learning and deal with existing images, tomographic medical imaging mainly produces cross-sectional or volumetric images of internal structures from sensor measurements. Recently, deep learning has started being actively developed worldwide for medical imaging, including both tomographic reconstruction and image analysis. While medical imaging is a well-established field, in which extensive teaching experience has been accumulated over the past few decades, updating the medical imaging course to reflect AI/ML influence is a new challenge given the rapidly changing landscape of AI-based medical imaging, particularly deep tomographic imaging. In the 2019 fall semester, the medical imaging course at Rensselaer Polytechnic Institute was modified to include an AI framework with positive feedback from students. Encouragingly, many students showed a strong motivation to learn AI in classes and hands-on projects, as confirmed in their survey reports. In the 2020 fall semester, we improved this course further, incorporating new advances. This article describes our teaching philosophy, practice, and considerations with respect to integrating deep learning, tomographic imaging, and hands-on programming to redefine the medical imaging course. Furthermore, given the persistent pandemic, online teaching and examination have become an integral part of higher education. These needs will be addressed as well, with the hope of developing an open course in the future.

Deep learning models have become the province of cutting-edge machine learning models that are widely used in medical imaging in multiple kinds of ranging from image recognition to natural language processing. This article presents a brief overview on the recent developments and some relevant issues in medical image processing and image analysis as they related to machine learning. The proposed article will not cover the entire application landscape because it is becoming a very large and rapidly growing field, but will instead focus primarily on deep learning theory and techniques in medical imaging. Our goal is threefold: (i) To provide such a brief overview to deep medical imaging learning; (ii) To represent different approaches as well as comparisons of how deep learning has been accomplished, from image retrieval to segmentation to disease prediction; (iii) Provide the current medical imaging-related deep learning applications.

Simple SummaryIn this paper, we introduce a new technique for enhancing medical image diagnosis through transfer learning (TL). The approach addresses the issue of limited labelled images by pre-training deep learning models on similar medical images and then refining them with a small set of annotated medical images. Our method demonstrated excellent results in classifying the humerus and wrist, surpassing previous methods, and showing greater robustness in various experiments. Furthermore, we demonstrate the adaptability of the approach with a CT case, which showed improvements in the results.Medical image classification poses significant challenges in real-world scenarios. One major obstacle is the scarcity of labelled training data, which hampers the performance of image-classification algorithms and generalisation. Gathering sufficient labelled data is often difficult and time-consuming in the medical domain, but deep learning (DL) has shown remarkable performance, although it typically requires a large amount of labelled data to achieve optimal results. Transfer learning (TL) has played a pivotal role in reducing the time, cost, and need for a large number of labelled images. This paper presents a novel TL approach that aims to overcome the limitations and disadvantages of TL that are characteristic of an ImageNet dataset, which belongs to a different domain. Our proposed TL approach involves training DL models on numerous medical images that are similar to the target dataset. These models were then fine-tuned using a small set of annotated medical images to leverage the knowledge gained from the pre-training phase. We specifically focused on medical X-ray imaging scenarios that involve the humerus and wrist from the musculoskeletal radiographs (MURA) dataset. Both of these tasks face significant challenges regarding accurate classification. The models trained with the proposed TL were used to extract features and were subsequently fused to train several machine learning (ML) classifiers. We combined these diverse features to represent various relevant characteristics in a comprehensive way. Through extensive evaluation, our proposed TL and feature-fusion approach using ML classifiers achieved remarkable results. For the classification of the humerus, we achieved an accuracy of 87.85%, an F1-score of 87.63%, and a Cohen’s Kappa coefficient of 75.69%. For wrist classification, our approach achieved an accuracy of 85.58%, an F1-score of 82.70%, and a Cohen’s Kappa coefficient of 70.46%. The results demonstrated that the models trained using our proposed TL approach outperformed those trained with ImageNet TL. We employed visualisation techniques to further validate these findings, including a gradient-based class activation heat map (Grad-CAM) and locally interpretable model-independent explanations (LIME). These visualisation tools provided additional evidence to support the superior accuracy of models trained with our proposed TL approach compared to those trained with ImageNet TL. Furthermore, our proposed TL approach exhibited greater robustness in various experiments compared to ImageNet TL. Importantly, the proposed TL approach and the feature-fusion technique are not limited to specific tasks. They can be applied to various medical image applications, thus extending their utility and potential impact. To demonstrate the concept of reusability, a computed tomography (CT) case was adopted. The results obtained from the proposed method showed improvements.

Enhancing medical image classification with generative AI using latent denoising diffusion probabilistic model and wiener filtering approach

The need for time and attention given by the doctor to the patient, due to the increased volume of medical data to be interpreted and filtered for diagnostic and therapeutic purposes has encouraged the development of the option to support, constructively and effectively, deep learning (DL) models for applications in the interpretation of medical images. Imaging physicians combine data from different stages and medical experiences, as opposed to DL models that incorporate the same types and modes of artisanal features. Access to big-databased medical imaging can be considered a benefit to the performance of DL models in interpreting medical imaging but similar or superior performance has been achieved with small, multi-feature and well-categorized databases that have improved annotation and labelling. The major contribution of this paper is primarily to highlight the impact of data quality, type and volume used by deep learning models in medical image analysis accompanied by updated characterization of the components of the deep learning process from data to medical applications. Second, it describes the specific correlations between the components of the deep learning process. Finally, it presents problems and directions for future research.

Simple SummaryFor automated cancer diagnosis on medical imaging, explainable artificial intelligence technology uses advanced image analysis methods like deep learning to make a diagnosis and analyze medical images, as well as provide a clear explanation for how it arrived at its diagnosis. The objective of XAI is to provide patients and doctors with a better understanding of the system’s decision-making process and to increase transparency and trust in the diagnosis method. The manual classification of cancer using medical images is a tedious and tiresome process, which necessitates the design of automated tools for the decision-making process. In this study, we explored the significant application of explainable artificial intelligence and an ensemble of deep-learning models for automated cancer diagnosis. To demonstrate the enhanced performance of the proposed model, a widespread comparison study is made with recent models, and the results exhibit the significance of the proposed model on benchmark test images. Therefore, the proposed model has the potential as an automated, accurate, and rapid tool for supporting the detection and classification process of cancer.Explainable Artificial Intelligence (XAI) is a branch of AI that mainly focuses on developing systems that provide understandable and clear explanations for their decisions. In the context of cancer diagnoses on medical imaging, an XAI technology uses advanced image analysis methods like deep learning (DL) to make a diagnosis and analyze medical images, as well as provide a clear explanation for how it arrived at its diagnoses. This includes highlighting specific areas of the image that the system recognized as indicative of cancer while also providing data on the fundamental AI algorithm and decision-making process used. The objective of XAI is to provide patients and doctors with a better understanding of the system’s decision-making process and to increase transparency and trust in the diagnosis method. Therefore, this study develops an Adaptive Aquila Optimizer with Explainable Artificial Intelligence Enabled Cancer Diagnosis (AAOXAI-CD) technique on Medical Imaging. The proposed AAOXAI-CD technique intends to accomplish the effectual colorectal and osteosarcoma cancer classification process. To achieve this, the AAOXAI-CD technique initially employs the Faster SqueezeNet model for feature vector generation. As well, the hyperparameter tuning of the Faster SqueezeNet model takes place with the use of the AAO algorithm. For cancer classification, the majority weighted voting ensemble model with three DL classifiers, namely recurrent neural network (RNN), gated recurrent unit (GRU), and bidirectional long short-term memory (BiLSTM). Furthermore, the AAOXAI-CD technique combines the XAI approach LIME for better understanding and explainability of the black-box method for accurate cancer detection. The simulation evaluation of the AAOXAI-CD methodology can be tested on medical cancer imaging databases, and the outcomes ensured the auspicious outcome of the AAOXAI-CD methodology than other current approaches.

In medical images, lesion segmentation is used to locate and isolate abnormal structures. It is an essential part of medical image analysis for precise diagnosis and care. However, obstacles exist in medical image lesion segmentation owing to things like image noise, shape and size fluctuation, and poor image quality. Automated lesion segmentation methods include conventional image processing techniques, deep learning (DL) models and machine learning (ML) algorithms to name a few. Thresholding, region growth, and active contour models are examples of conventional methods, while decision trees, random forests, and support vector machines are examples of ML techniques. DL models particularly convolutional neural networks (CNNs), have shown extraordinary performance in lesion segmentation because to their innate potential to autonomously collect high-level characteristics. The objective of the research is to study lesion segmentation in medical images and explore different methods for accurate and precise diagnosis and care.The research focuses on the obstacles faced in lesion segmentation in medical images, such as image noise, shape and size fluctuation, and poor image quality. The research also highlights the need for evaluation metrics, such as sensitivity, specificity, Dice coefficient, and Hausdorff distance, to assess the performance of lesion segmentation algorithms. Additionally, the research emphasizes the importance of annotated datasets for training and evaluating the performance of segmentation algorithms.

Traditional medical image segmentation methods have problems such as low segmentation accuracy and low adaptive ability. Therefore, many scholars have proposed a medical image segmentation method based on deep learning, which has achieved good results in the field of medical image segmentation. However, this type of method has the following problems in the application process: (1) Medical image segmentation target boundary positioning problem. Constrained by factors such as medical image contrast, heterogeneity, and boundary resolution, existing convolution models still cannot accurately locate boundaries. (2) Deep adaptability of deep learning network structure to medical images. Because medical images have more distinct and different feature information than natural images, the current deep learning-based medical segmentation methods have not fully considered this feature. In view of this, this paper proposes a multi-level boundary-aware RUNet segmentation model. The network structure consists of a U-Net-based segmentation network and a multi-level boundary detection network. It can solve the problem of boundary positioning. At the same time, in order to solve the problem of poor adaptability of deep learning network structures to medical images, this paper proposes to introduce a new interactive self-attention module into deep learning models. It can make the feature map get global information, and realize the effective extraction of medical image feature information. It solves the problem of weak matching between the deep learning network structure and medical images. Based on the above ideas, this paper proposes an image segmentation algorithm based on a multi-layer boundary perception-self-attention mechanism deep learning model. This method and other mainstream segmentation algorithms are used to perform experiments on related medical databases. The results show that the proposed method not only improves the segmentation effect significantly compared with traditional machine learning methods, but also improves it to a certain extent compared with other deep learning methods.

Dual adversarial attacks on Explainable Deep Learning in medical image classification.

Deep learning revolutionized biological signal processing and visualization. Deep learning has transformed patient analysis, diagnosis, and treatment by automatically spotting subtleties in large datasets. This article reviews recent deep learning advances in biomedical signal processing and medical imaging. Deep learning models handle electrical heart, brain, and muscle activity in biomedical signal processing. Deep neural networks have improved medical diagnosis and patient monitoring via signal denoising, feature extraction, classification, and anomaly detection. Deep learning improves medical imaging accuracy and efficiency. CNNs excel in segmenting MRI, CT, X-ray, and ultrasound images; identifying lesions; and classifying illnesses. Deep learning and medical imaging have improved diagnosis and personalized therapy. Transfer learning and data augmentation allow pre-trained deep learning models to adapt to biological signal and medical imaging applications with minimum training data. Because of this, deep learning approaches may be applied in therapeutic contexts with limited annotated datasets. Despite advancements, deep learning model interpretability, training data biases, and clinical deployment rules continue. Before ethically integrating deep learning into healthcare, researchers and practitioners must resolve these concerns. Deep learning in biomedical signal processing and medical imaging improves diagnosis and therapy. Deep learning-based technologies might revolutionize healthcare and improve patient outcomes.