Text Box: Volume 2, Number 1, July 2020 e-ISSN: 2797-6068 and p-ISSN: 2777-0915

 


ML-BASED MEDICAL IMAGE ANALYSIS FOR ANOMALY DETECTION IN CT SCANS, X-RAYS, AND MRIS

 

Moazzam Siddiq

Independent Researcher, Manchester, United Kingdom

Email: [email protected]

 

KEYWORDS

medical image analysis; healthcare; types of medical images; CT scans; X-rays; MRI's; anomaly detection; machine learning algorithms

 

 

ABSTRACT

The area of medical image analysis is examined in this review article along with its potential to revolutionize healthcare. The article starts off by going through the different kinds of medical imaging, such as CT scans, X-rays, and MRIs, as well as the difficulties in analyzing these images. The discussion then switches to the use of machine learning algorithms to identify anomalies in medical pictures, emphasising the value of high-quality data and inter professional cooperation. The study also discusses the difficulties in analyzing medical images, including the need for more reliable machine learning algorithms and standardized techniques for image acquisition. The significance of creating clear norms and laws is also emphasized, along with concerns about patient privacy and data security. A discussion of recent developments and potential future paths in medical image analysis is included in the article's conclusion. Personalised medicine, the utilization of augmented reality and virtual reality technology, and the creation of visible and comprehensible machine learning algorithms are a few of these. The essay gives a thorough summary of the state of medical image analysis now and how it could revolutionize healthcare as a whole.

 

 

INTRODUCTION

The interpretation of medical pictures including CT scans, X-rays, and MRIs is automated using machine learning techniques in the rapidly expanding field of medical image analysis [1, 2]. The complexity of medical images, the growing amount of medical data collected daily, and the shortcomings of human interpretation all point to the necessity for automated medical image analysis [3, 4]. By making disease diagnosis, treatment planning, and monitoring quicker and more accurate, medical image analysis has the potential to completely transform healthcare.

Medical Image Modalities: There are many ways to acquire medical photographs, each with their own benefits and drawbacks. X-rays are used in CT scans to create precise images of the inside organs and tissues of the body. X-rays are suitable for regular diagnosis of a variety of disorders such as fractures, tumors, and lung ailments since they are rapid and non-invasive [5]. A powerful magnetic field and radio waves are used in MRI to provide precise images of the body's soft tissues. MRIs are more thorough than CT scans and are frequently used to identify tumors, joint issues, and cancers [6]. Using high-frequency sound waves, ultrasound can provide images of the internal organs of the body. Monitoring fetal development, identifying gallbladder and liver conditions, and assisting with biopsies can all be done with the safe, non-invasive use of ultrasound. The best modality to choose will depend on the clinical question at hand as well as the patient's health. Each modality has its advantages and disadvantages.

Figure 1 Medical image of Lungs

Preprocessing and Feature Extraction Preprocessing and feature extraction procedures are necessary to get the data ready for analysis before applying machine learning algorithms to medical pictures. To enhance the image quality and eliminate artefacts, preprocessing entails a number of processes such noise reduction, image registration, and normalization [7]. Finding pertinent aspects in a medical image that can be used for analysis is called feature extraction. Intensity-based features like mean and standard deviation, texture-based features like co-occurrence matrix and wavelet transform, and shape-based features like geometric features and curvature are all examples of feature extraction techniques [8]. Depending on the modality, the clinical question, and the type of analysis, a feature extraction approach is chosen.

Machine Learning Algorithms for Medical Image Analysis: By automating the process of finding patterns and anomalies in medical images, machine learning algorithms play a crucial role in image analysis. Supervised, unsupervised, and deep learning algorithms are three broad categories of machine learning algorithms. Support vector machines and random forests are two examples of supervised learning techniques that employ labelled training data to create a model that can classify fresh images into established categories [9]. Unsupervised learning methods, including clustering and principal component analysis, find patterns and structures in the data instead of using labelled training data. By enabling automated feature extraction and delivering cutting-edge performance on a variety of medical image analysis tasks, deep learning techniques, such as convolutional neural networks and recurrent neural networks, have revolutionized the field of medical image analysis.

Applications of Medical Image Analysis: There are many uses for medical image analysis in healthcare, from diagnosis to planning and monitoring of treatments. Medical image analysis can be used to diagnose bone fractures, locate lung nodules, find brain tumors in the early stages, and track fetal development [10]. Analysis of medical images can be used to measure the evolution of a condition, forecast how a therapy will go, and gauge how well it worked. The promising subject of medical image analysis has the potential to transform healthcare by facilitating quicker and more precise disease diagnosis, treatment planning, and disease monitoring. Medical imaging, machine learning, and clinical applications are just a few of the fields that demand knowledge while analyzing medical images. Medical image analysis will be more crucial as the amount of medical data grows as it will improve patient outcomes and advance healthcare.

Machine Learning Algorithms in Medical Image Analysis:

Due to their capacity to automate image analysis and boost diagnostic precision, machine learning algorithms have found extensive usage in the study of medical images [11]. Due to the availability of extensive medical image datasets and the advancements in deep learning algorithms, the application of machine learning algorithms in medical image analysis has risen quickly in recent years.

Supervised Learning Algorithms: Algorithms for supervised learning are frequently employed in classification and segmentation tasks in medical image analysis. For these algorithms to develop a model that can identify or segment fresh images, training data must be labelled. Support vector machines (SVM), random forests, and artificial neural networks are examples of popular supervised learning techniques used in medical image processing. While random forests are frequently utilized for multi-class classification problems like tissue categorization, SVMs are frequently employed for binary classification tasks like tumor detection [12]. Convolutional and recurrent neural networks, two types of artificial neural networks, have attained cutting-edge performance on a variety of medical image processing tasks such image classification, segmentation, and registration.

Unsupervised Learning Algorithms: In tasks involving clustering and feature extraction in medical image analysis, unsupervised learning methods are applied. These algorithms discover patterns and structures in the data rather than requiring labelled training material. Principal component analysis (PCA), hierarchical clustering, and k-means clustering are examples of widely used unsupervised learning methods in medical image analysis [13]. While PCA is frequently used for feature extraction tasks like lowering the dimensionality of high-dimensional medical pictures, K-means clustering is frequently utilized for segmentation tasks like identifying tumor boundaries.

Deep Learning Algorithms: By enabling automated feature extraction and attaining cutting-edge performance on a variety of medical image analysis tasks, deep learning algorithms have completely changed the field of medical image analysis. In comparison to conventional machine learning algorithms, deep learning algorithms, such as CNNs and RNNs, perform better and can automatically learn hierarchical features from raw medical pictures [14]. In medical image analysis, CNNs are frequently employed for tasks like image classification, segmentation, and registration [14]. RNNs are employed in jobs involving sequence data, such as time-series data and medical reports, in medical image analysis.

Applications of Machine Learning Algorithms in Medical Image Analysis: Image segmentation, classification, registration, and reconstruction are just a few of the medical image analysis tasks for which machine learning techniques have been applied. Early-stage cancer detection, the identification of brain tumors, the detection of lung nodules, the diagnosis of bone fractures, and the monitoring of fetal development have all been accomplished using machine learning algorithms [15]. The quantification of illness development, the prediction of therapeutic outcomes, and the evaluation of therapeutic response have all been accomplished using machine learning algorithms.

Challenges in Machine Learning Algorithms in Medical Image Analysis: Despite the enormous advancements made in machine learning algorithms for medical image interpretation, there are still a number of issues that need to be resolved. Lack of extensive annotated datasets, which restricts the development and assessment of machine learning algorithms, is one of the key issues. The interpretability of deep learning algorithms is another issue that prevents their use in therapeutic contexts [16]. To ensure that machine learning algorithms work well across a range of patient populations and imaging modalities, robustness and generalizability must also be increased.

By enabling automated picture analysis and increasing diagnosis accuracy, machine learning algorithms have shown considerable promise in the field of medical image analysis. Due to the accessibility of big medical image datasets and the advancements in deep learning algorithms, the use of machine learning techniques in medical image analysis has risen quickly in recent years [17]. Despite the tremendous advancements gained, a number of issues still need to be resolved in order to guarantee the secure and efficient application of machine learning algorithms in clinical settings.

Types of Medical Images:

A wide variety of medical disorders can be diagnosed and treated with the use of medical imaging. A number of modalities are used in medical imaging to take pictures of the body's internal organs, tissues, and structures. These pictures can be used to detect anomalies and make medical diagnoses [18]. Medical imaging can take many various forms, each with specific features and advantages, such as CT scans, X-rays, and MRIs.

CT Scans: X-rays are used in Computed Tomography (CT) scans to produce finely detailed pictures of the inside organs. The detection of bone fractures and other injuries, as well as the detection of anomalies in soft tissues and organs including the liver, lungs, and brain, are among the many uses for CT scans. CT scans can also be used to keep track of how well diseases like cancer and heart disease are being treated. The fact that CT scans are non-invasive and deliver high-quality images with a very quick scanning period is one of their advantages. However, ionizing radiation, which can be dangerous in large levels, is exposed to patients during CT scans [19]. As a result, it's crucial to carefully consider the advantages and disadvantages of CT scans for each patient.

X-rays: X-rays employ electromagnetic radiation to produce images of the inside organs and tissues of the body. The most frequent applications of X-rays are the detection of lung and chest abnormalities, as well as the detection of bone fractures and other traumas. X-rays are frequently used in dentistry to detect tooth decay and other problems with the mouth's health. X-rays are rapid and non-invasive, just like CT scans. Ionizing radiation, which can be dangerous at high doses, is further exposed to patients during X-rays [20]. As a result, it's crucial to utilize X-rays sparingly and to carefully weigh the advantages and disadvantages for each patient.

MRI's: A powerful magnetic field and radio waves are used in magnetic resonance imaging (MRI) to provide precise pictures of the inside organs of the body. The brain, liver, and heart are examples of soft tissues and organs where MRIs are very helpful in detecting problems. MRIs can also be used to find injuries, tumors, and other types of illnesses. The fact that MRIs don't subject patients to ionizing radiation makes them a safer option for individuals who may need several scans. MRIs cost more money and take longer to complete than other kinds of medical pictures, though. Additionally, due to safety concerns, people with certain medical disorders, such as pacemakers, may not be allowed to have MRIs. In conclusion, medical imaging is essential for the detection and management of a variety of medical problems. Three of the most popular medical picture kinds are CT scans, X-rays, and MRIs, each with special characteristics and advantages [21]. Healthcare professionals should carefully weigh the advantages and disadvantages of each form of medical imaging before selecting the best modality for each patient.

METHOD RESEARCH

In this study researchers used a qualitative approach with the type of research case study research (case study) and is descriptive. According to Denzin and Lincoln qualitative research is research that uses a natural setting, with the intention of interpreting phenomena that occur and are carried out by involving various ways existing methods. The qualitative approach is an important one to understand a social phenomenon and the individual perspective studied. The qualitative approach is also the research procedure produce descriptive data in the form of written words or verbal from the behavior of people who are silent.

 

RESULT AND DISCUSSION

Anomaly Detection in Medical Images

Machine learning algorithms are crucial for detecting anomalies in medical imaging, which can help doctors identify and treat a variety of medical diseases. Early and more precise diagnoses can improve patient outcomes and perhaps save lives thanks to machine learning algorithms' capacity to spot patterns and anomalies in medical pictures. The enormous volume of data that needs to be processed presents one of the main difficulties in medical picture analysis. Large, high-resolution files typical of medical imaging can require a great amount of time and resources to analyse. By automatically identifying abnormalities and emphasising regions of interest in medical imaging, machine learning algorithms can help interpret this data [22]. Numerous medical diseases, including as cancer, cardiovascular illness, and neurological disorders, can be detected using anomaly detection. In order to help medical personnel create more successful treatment regimens, machine learning algorithms, for instance, can be trained to recognize patterns of abnormal growth in cancerous tumors. The early identification and prevention of heart disease can be helped by machine learning algorithms that can be trained to spot anomalies in cardiovascular pictures like blockages or plaque buildup.

The ability of machine learning algorithms to learn and adapt over time is one advantage of using them in medical picture analysis. Machine learning algorithms can increase their accuracy and effectiveness at identifying anomalies by being trained on vast datasets of medical images. This may result in earlier and more precise diagnosis, improving patient outcomes and maybe saving lives. However, there are a number of difficulties and factors to be thought about when applying machine learning algorithms to the detection of anomalies in medical imaging. The requirement to guarantee that algorithms are objective and accurate is one of the main problems. Machine learning algorithms are susceptible to biases and mistakes, which might produce results that are incorrect or discriminating [23]. Consequently, it is crucial to properly build algorithms and to confirm their efficacy.

The necessity to safeguard patient information and privacy is another factor in medical image analysis. Sensitive patient data is contained in medical photographs, so it's critical to protect this data from unauthorized access or exposure. De-identification methods and secure storage protocols can be used to achieve this. In conclusion, early and accurate diagnoses made possible by anomaly detection in medical pictures using machine learning algorithms have the potential to greatly improve patient outcomes. The need to ensure algorithm fairness and accuracy, safeguard patient privacy and data security, and address regulatory compliance are just a few of the issues and factors that need to be taken into account [24]. Machine learning algorithms can be used to help medical personnel in diagnosing and treating a variety of medical disorders by carefully constructing algorithms and putting in place relevant protocols.

 

Challenges in Medical Image Analysis

Due to the complexity of medical images and the crucial significance of accurate diagnoses, there are considerable challenges in medical image analysis. Medical image analysis using machine learning algorithms has showed considerable promise, but there are still a number of issues that need to be resolved before their full potential can be realised. The demand for huge, high-quality datasets is one of the main obstacles in medical image analysis. For learning and making precise predictions, machine learning algorithms rely on enormous volumes of data. This implies that in the case of medical imaging, algorithms must be trained on big datasets of excellent photos. However, due to privacy issues and the high expense of medical imaging operations, such datasets are frequently hard to find and expensive to purchase.

The requirement for interpretable models presents another difficulty in medical picture analysis. Medical picture patterns and anomalies can be recognized by machine learning algorithms, but the mechanisms by which they do so are frequently opaque and challenging to understand [25]. This can make it difficult to validate data and create efficient treatment strategies based on them. The requirement for explain ability and openness in machine learning algorithms is a similar issue. It is crucial to comprehend how these algorithms make choices and to make sure they are doing so in an ethical and objective manner as machine learning algorithms are increasingly integrated into healthcare decision-making. This necessitates creating models that can concisely and clearly describe their decision-making processes. The requirement to address issues of bias and impartiality presents another difficulty in medical image analysis. Biases may exist in the data used to train machine learning algorithms as well as in the algorithms' design. Particularly for populations who are already marginalized, these biases might produce biased or erroneous results [25]. Careful assessment of the data used to train algorithms and the criteria used to assess their performance is necessary to address these difficulties.

Regulatory and legal issues might also provide tough problems for medical image analysis. Strict guidelines for the management and preservation of medical data, including medical photographs, are set down in laws like HIPAA and GDPR. Implementing suitable security measures and ensuring patient privacy is maintained are required for compliance with these regulations. New methods and strategies for medical image analysis must be developed as part of continuous research to meet these concerns. The creation of federated learning approaches, which enable several institutions to work together on machine learning tasks without sharing patient data, is one potential field of research. This can make it possible to build bigger, more varied datasets while still protecting patient privacy. The creation of interpretability and explain ability strategies for machine learning algorithms is another area of research. It may be possible to increase trust and confidence in machine learning findings by creating models that can clearly and understandably describe their decision-making processes [26]. Overall, despite the fact that there are numerous obstacles to overcome in the field of medical image analysis, continuing research and development present promising directions for overcoming these obstacles and realizing the full potential of machine learning algorithms in the healthcare industry. Machine learning algorithms have the ability to greatly enhance patient outcomes and revolutionize the field of medical imaging by trying to construct more interpretable, transparent, and fair models and by addressing concerns with data quality and privacy.

 

Advancements and Future Directions in Medical Image Analysis

The area of healthcare could be revolutionized by improvements in diagnosis and treatment results for a variety of medical disorders thanks to current developments and planned directions in medical image analysis. Medical image analysis has undergone substantial advancements in recent years, thanks to improvements in machine learning methods and the accessibility of big, high-quality datasets [27]. The application of deep learning algorithms is a significant area of development for medical picture analysis. Artificial neural networks are used by deep learning algorithms, a subset of machine learning algorithms, to recognize and learn from data patterns [28]. In the fields of segmentation, classification, and anomaly detection, they have demonstrated considerable promise in the study of medical images.

Deep learning algorithms have been successfully used in the interpretation of medical images on multiple occasions in recent years. For instance, deep learning algorithms have been used to accurately identify breast cancer in mammograms and identify lung nodules in CT scans of the lungs [29]. Deep learning algorithms have also been used to track the evolution of Alzheimer's disease and detect brain tumors in MRI scans. The application of multi-modal imaging methods is another area of development in medical image analysis. To get a more full view of the patient's condition, multi-modal imaging uses several imaging modalities, including CT scans, MRI scans, and PET scans [30]. Combining information from many imaging modalities may make diagnosis more accurate and reveal irregularities that weren't previously noticed.

The development of medical image analysis has also been aided by improvements in computer hardware and software. Medical image analysis is now more quick and accurate than ever thanks to the development of high-performance computing systems like graphics processing units (GPUs) and field-programmable gate arrays (FPGAs), which allow for the real-time processing of large amounts of data [31]. Additionally, it is now simpler to create and use machine learning algorithms for medical image analysis thanks to the emergence of specialized software platforms like Tensor Flow and Torch. There are various promising future prospects for the development of medical image analysis. The creation of federated learning techniques is one area of emphasis [32]. Federated learning enables the collaboration of different institutions on machine learning tasks without disclosing patient information, allowing for the development of larger and more varied datasets while protecting patient privacy. The problem of data availability and quality in medical image analysis may be helped by this.

The creation of comprehensible AI methods is another area of emphasis. Machine learning models that can clearly and understandably explain their decision-making processes are referred to as "explainable AI" [33]. This can assist in addressing the problem of interpretability in medical image analysis and increase confidence and trust in the outcomes of machine learning. A great amount of attention is being paid to creating machine learning algorithms that are reliable and resistant to adversarial attacks. Cyber-attacks known as adversarial attacks include the purposeful manipulation of input data in attempt to deceive machine learning systems. The accuracy and dependability of medical image analysis may be increased by creating algorithms that are resistant to such attacks [34]. Medical image analysis innovations have the potential to revolutionize the healthcare industry by enhancing the outcomes of diagnosis and treatment for a variety of medical disorders. The future of medical image analysis is promising if machine learning algorithms are developed and improved further, as well as issues like data quality and privacy are addressed. We may anticipate further advancements in this fascinating and quickly developing sector with ongoing research and development [35].

CONCLUSION

Healthcare could be transformed by the rapidly developing field of medical image analysis. There have been significant developments in recent years that are changing how doctors diagnose and treat patients, from the detection of anomalies in medical pictures to the creation of sophisticated machine learning algorithms. The significance of collaboration between medical practitioners and computer scientists is one of the main lessons to be learned from this conversation. Together, these two teams can create and improve machine learning algorithms that are specifically suited to the requirements of the medical industry. The accuracy, dependability, and safety of machine learning algorithms for application in medical contexts can be improved with the aid of this partnership. The requirement for ongoing study and advancement in medical image analysis is another crucial conclusion. Even if there have been substantial improvements recently, there are still a lot of problems that need to be solved. For gathering medical photographs, for instance, there is a need for higher quality data and more standardized processes. Additionally, stronger machine learning techniques are required in order to manage the complexity and diversity of medical imagery. Additionally, concerns about patient privacy and data security must be addressed. Patient data may be hacked or exploited for unauthorized reasons as medical image analysis becomes more common. Establishing precise rules and regulations for the use of patient data in medical image analysis is crucial to addressing this issue. There are numerous intriguing avenues for medical image analysis in the future. For instance, the usage of augmented reality (AR) and virtual reality (VR) technology may offer medical practitioners new methods to see and interact with medical imaging. Additionally, these technologies might make it easier for medical specialists to work together across geographical boundaries. Personalised medicine, which involves adapting medical treatments to each patient's particular needs based on their unique traits, such as their genetics or medical history, is another promising area of growth. Personalised medicine and medical image analysis could be combined to create medicines that are more efficient and have fewer negative effects. Finally, it is important to keep creating visible and comprehensible machine learning algorithms. Medical personnel need to be able to comprehend how machine learning algorithms make judgments and suggestions as medical image analysis becomes increasingly prevalent. This will ensure that these algorithms are being used in a secure and efficient manner and assist to increase trust and confidence in them. The subject of medical image analysis is fast developing and has the potential to revolutionize how we identify and treat medical diseases. We can anticipate further development in this fascinating area by enhancing machine learning algorithms, addressing issues with data quality and patient privacy, and investigating new research trajectories.

REFERENCES

1.      R. Domingues, M. Filippone, P. Michiardi, and J. Zouaoui, “A comparative evaluation of outlier detection algorithms: Experiments andanalyses,” Pattern Recognition, vol. 74, pp. 406–421, 2018.

2.      F. E. Grubbs, “Procedures for detecting outlying observations in samples,” Technometrics, vol. 11, no. 1, pp. 1–21, 1969.

3.      P. Vincent, H. Larochelle, I. Lajoie, Y. Bengio, P.-A. Manzagol, andL. Bottou, “Stacked denoising autoencoders: Learning useful representations in a deep network with a local denoising criterion.” Journal ofmachine learning research, vol. 11, no. 12, 2010.

4.      J. Sun, X. Wang, N. Xiong, and J. Shao, “Learning sparse representationwith variational auto-encoder for anomaly detection,” pp. 33 353–33 361, 2018.

5.      H. Uzunova, S. Schultz, H. Handels, and J. Ehrhardt, “Unsupervisedpathology detection in medical images using conditional variationalautoencoders,” International journal of computer assisted radiology andsurgery, vol. 14, no. 3, pp. 451–461, 2019.

6.      T. Schlegl, P. Seeb ¨ock, S. M. Waldstein, U. Schmidt-Erfurth, andG. Langs, “Unsupervised anomaly detection with generative adversarialnetworks to guide marker discovery,” in International conference oninformation processing in medical imaging. Springer, 2017, pp. 146–157.

7.      T. Schlegl, P. Seeb ¨ock, S. M. Waldstein, G. Langs, and U. SchmidtErfurth, “f-anogan: Fast unsupervised anomaly detection with generative adversarial networks,” Medical image analysis, vol. 54, pp. 30–44, 2019.

8.      C. Baur, B. Wiestler, S. Albarqouni, and N. Navab, “Deep autoencodingmodels for unsupervised anomaly segmentation in brain mr images,”in International MICCAI Brainlesion Workshop. Springer, 2018, pp.161–169.

9.      D. Gong, L. Liu, V. Le, B. Saha, M. R. Mansour, S. Venkatesh, andA. v. d. Hengel, “Memorizing normality to detect anomaly: Memoryaugmented deep autoencoder for unsupervised anomaly detection,” inProceedings of the IEEE/CVF International Conference on ComputerVision, 2019, pp. 1705–1714.

10.   W. Li, W. Mo, X. Zhang, J. J. Squiers, Y. Lu, E. W. Sellke, W. Fan, J. M.DiMaio, and J. E. Thatcher, “Outlier detection and removal improvesaccuracy of machine learning approach to multispectral burn diagnosticimaging,” Journal of biomedical optics, vol. 20, no. 12, p. 121305,2015.

11.   W. Li, W. Mo, X. Zhang, Y. Lu, J. J. Squiers, E. W. Sellke, W. Fan,J. M. DiMaio, and J. E. Thatcher, “Burn injury diagnostic imagingdevice’s accuracy improved by outlier detection and removal,” inAlgorithms and Technologies for Multispectral, Hyperspectral, andUltraspectral Imagery XXI, vol. 9472. International Society for Opticsand Photonics, 2015, p. 947206.

12.   K. Li, C. Ye, Z. Yang, A. Carass, S. H. Ying, and J. L. Prince,“Quality assurance using outlier detection on an automatic segmentationmethod for the cerebellar peduncles,” in Medical Imaging 2016: ImageProcessing, vol. 9784. International Society for Optics and Photonics,2016, p. 97841H.

13.   Bercovich, E., & Javitt, M. C. (2018). Medical imaging: from roentgen to the digital revolution, and beyond. Rambam Maimonides medical journal, 9(4).

14.   D. M. Tax and R. P. Duin, “Uniform object generation for optimizingone-class classifiers,” Journal of machine learning research, vol. 2, no.Dec, pp. 155–173, 2001.

15.   J. Randolph, “A guide to writing the dissertation literature review,”Practical Assessment, Research, and Evaluation, vol. 14, no. 1, p. 13,2009.

16.   D. Stepec and D. Skoˇcaj, “Image synthesis as a pretext for unsu ˇ pervisedhistopathological diagnosis,” in International Workshop on Simulationand Synthesis in Medical Imaging. Springer, 2020, pp. 174–183.

17.   D. Sato, S. Hanaoka, Y. Nomura, T. Takenaga, S. Miki, T. Yoshikawa,N. Hayashi, and O. Abe, “A primitive study on unsupervised anomalydetection with an autoencoder in emergency head ct volumes,” inMedical Imaging 2018: Computer-Aided Diagnosis, vol. 10575. International Society for Optics and Photonics, 2018, p. 105751P.

18.   T. Fujioka, K. Kubota, M. Mori, Y. Kikuchi, L. Katsuta, M. Kimura,E. Yamaga, M. Adachi, G. Oda, T. Nakagawa et al., “Efficient anomalydetection with generative adversarial network for breast ultrasoundimaging,” Diagnostics, vol. 10, no. 7, p. 456, 2020.

19.   Syed, L., Jabeen, S., & Manimala, S. (2018). Telemammography: a novel approach for early detection of breast cancer through wavelets based image processing and machine learning techniques. Advances in soft computing and machine learning in image processing, 149-183.

20.   S. Venkataramanan, K.-C. Peng, R. V. Singh, and A. Mahalanobis,“Attention guided anomaly localization in images,” in European Conference on Computer Vision. Springer, 2020, pp. 485–503.

21.   P. Seeb ¨ock, J. I. Orlando, T. Schlegl, S. M. Waldstein, H. Bogunovi´c,S. Klimscha, G. Langs, and U. Schmidt-Erfurth, “Exploiting epistemicuncertainty of anatomy segmentation for anomaly detection in retinaloct,” IEEE transactions on medical imaging, vol. 39, no. 1, pp. 87–98,2019.

22.   Bercovich, E., & Javitt, M. C. (2018). Medical imaging: from roentgen to the digital revolution, and beyond. Rambam Maimonides medical journal, 9(4).

23.   J. Wolleb, R. Sandk ¨uhler, and P. C. Cattin, “Descargan: Disease-specificanomaly detection with weak supervision,” in International Conferenceon Medical Image Computing and Computer-Assisted Intervention.Springer, 2020, pp. 14–24.

24.   T. Tlusty, G. Amit, and R. Ben-Ari, “Unsupervised clustering ofmammograms for outlier detection and breast density estimation,” in2018 24th International Conference on Pattern Recognition (ICPR).IEEE, 2018, pp. 3808–3813.

25.   Q. Wei, Y. Ren, R. Hou, B. Shi, J. Y. Lo, and L. Carin, “Anomalydetection for medical images based on a one-class classification,”in Medical Imaging 2018: Computer-Aided Diagnosis, vol. 10575.International Society for Optics and Photonics, 2018, p. 105751M.

26.   N. Pawlowski, M. C. Lee, M. Rajchl, S. McDonagh, E. Ferrante,K. Kamnitsas, S. Cooke, S. Stevenson, A. Khetani, T. Newman et al.,“Unsupervised lesion detection in brain ct using bayesian convolutionalautoencoders,” 2018.

27.   K. Armanious, C. Jiang, S. Abdulatif, T. K ¨ustner, S. Gatidis, andB. Yang, “Unsupervised medical image translation using cyclemedgan,” in 2019 27th European Signal Processing Conference (EUSIPCO). IEEE, 2019, pp. 1–5.

28.   M. Khosla, K. Jamison, A. Kuceyeski, and M. R. Sabuncu, “Detectingabnormalities in resting-state dynamics: An unsupervised learningapproach,” in International Workshop on Machine Learning in MedicalImaging. Springer, 2019, pp. 301–309.

29.   Z. Alaverdyan, J. Chai, and C. Lartizien, “Unsupervised feature learningfor outlier detection with stacked convolutional autoencoders, siamesenetworks and wasserstein autoencoders: application to epilepsy detection,” in Deep Learning in Medical Image Analysis and MultimodalLearning for Clinical Decision Support. Springer, 2018, pp. 210–217.

30.   X. Chen, N. Pawlowski, B. Glocker, and E. Konukoglu, “Unsupervisedlesion detection with locally gaussian approximation,” in InternationalWorkshop on Machine Learning in Medical Imaging. Springer, 2019, pp. 355–363.

31.   Z. Alaverdyan, J. Jung, R. Bouet, and C. Lartizien, “Regularizedsiamese neural network for unsupervised outlier detection on brainmultiparametric magnetic resonance imaging: application to epilepsylesion screening,” Medical image analysis, vol. 60, p. 101618, 2020.

32.   Fujita, H. (2020). AI-based computer-aided diagnosis (AI-CAD): the latest review to read first. Radiological physics and technology, 13(1), 6-19.

33.   Tandel, G. S., Biswas, M., Kakde, O. G., Tiwari, A., Suri, H. S., Turk, M., ... & Suri, J. S. (2019). A review on a deep learning perspective in brain cancer classification. Cancers, 11(1), 111.

34.   C. Baur, R. Graf, B. Wiestler, S. Albarqouni, and N. Navab,“Steganomaly: Inhibiting cyclegan steganography for unsupervisedanomaly detection in brain mri,” in International Conference on Medical Image Computing and Computer-Assisted Intervention. Springer,2020, pp. 718–727.

35.   D. Zimmerer, F. Isensee, J. Petersen, S. Kohl, and K. Maier-Hein, “Unsupervised anomaly localization using variational auto-encoders,” inInternational Conference on Medical Image Computing and ComputerAssisted Intervention. Springer, 2019, pp. 289–297.

 

 

 

Copyright holders:

Moazzam Siddiq (2020)

First publication right:

Devotion - Journal of Research and Community Service

This article is licensed under a Creative Commons Attribution-ShareAlike 4.0 International