Добавил:
kiopkiopkiop18@yandex.ru t.me/Prokururor I Вовсе не секретарь, но почту проверяю Опубликованный материал нарушает ваши авторские права? Сообщите нам.
Вуз: Предмет: Файл:

Ординатура / Хирургия / Библиотека им академика М.И. Перельмана / Книга_980_Библиотеки_им_академика_М_И_Перельмана

.pdf
Скачиваний:
0
Добавлен:
31.08.2026
Размер:
28 Мб
Скачать
bc
de
Brief Introduction toArticial Intelligence andMachine Learning
https://t.me/medicina_free
a
Dog
Cat
Discriminator
Training Set
279
Noise
Generator
Fake image
Fake ?
Real ?
Fig. 5 (a) Articial neural networks. By propagating structured data (green nodes—input vari- ables), such as radiomics, through hidden layers (left brown nodes), articial neural networks can model complex nonlinear relationships between input variables and outcomes (right brown nodes). (b) Convolutional neural networks. Deep learning applications rely on convolutional neural net­works as their backbone. They are composed of input and output layers that are separated by a number of hidden layers. (c) Recurrent neural network. This type of articial neural network is characterized by connections between nodes in a sequence. This model can process variable-length input sequences by utilizing their memory. (d) Generative adversarial networks. Generally, these models are used for generative modeling, which involves automatically identifying regularities or patterns in input data to produce new examples that can be used as a replacement for the original dataset. (e) Autoencoders. These structures are neural networks that learn efcient data representa­tions (encoding) by training. They can be used for denoising images, compressing images, and generating images as well
280
https://t.me/medicina_free
mimic the distribution of real-world data and may thus generate new patterns of visual data (images). Both the generating network and the discriminator network are part of the GAN.As part of their training, the discriminator and generator compete against one another to generate convincing false pictures that trick the discriminator into thinking the images are genuine. Since both models have a stake in the compe­tition’s outcome, this training method is called “adversarial training.” This kind of training may also be applied to developing a network segmentation. To differentiate between the produced and ground truth segmentation maps, we use a segmentation network instead of a generator (the target segmentation maps). This promotes the segmentation network to create segmentation maps that are more physically realis­tic [79, 80].
Autoencoders (AE)
Autoencoders (AEs) are a subclass of NNs meant to learn compressed implicit rep­resentations from the input autonomously. An AE’s usual design is composed of an encoder network and a decoder network for reconstructing the input. Numerous CNN variations have been suggested to transmit information from the encoder to the decoder and improve segmentation accuracy. The U-Net is the most well-known CNN version for biomedical image segmentation [81]. The U-Net utilizes “skip connections” between encoder and decoder for better segmentation to restore lost spatial information during the downsampling process.
S. R. Motamedian et al.
3.5 Requirements forTraining theModel
3.5.1 Software Requirements
Various programming languages can be used for deploying and implementing the models. However, the Python programming language is more favorable among sci­entic communities and industrial applications. Deep learning libraries provide a higher level of the programming interface, including many levels of mathematical concepts based on linear algebra, calculus, probability, and numerical computation to efciently use available computational resources like the GPU or CPU [82]. TensorFlow [83] and PyTorch [84] were the two most used libraries in 2021. Frequently used, these frameworks can implement NN architectures.
3.5.2 Hardware Requirements
Hardware selection, which means determining the technical specs based on a given deep learning model, is essential in running a deep learning project. Dataset volume and model complexity are two key parameters to consider in hardware selection. CPU, GPUs, or cloud computing platforms can be used for training deep learning models. GPUs and TPUs, unlike CPUs, are architectures designed for heavy paral­lel computing, with limited memory size but at very high bandwidth. Using more signicant GPU memory (currently, commercial GPUs provide memory sizes between 8 and 32GB), training deeper models with a higher number of trainable parameters will be convenient [85]. To increase computational performance, one
Brief Introduction toArticial Intelligence andMachine Learning
https://t.me/medicina_free
281
can use multi-GPU with considering additional hardware setups (e.g., power supply and cooling).
Cloud computing, which refers to internet-based services using a third-party hardware resource, can be the rst decision if training time matters. Although cloud computing platforms provide high computational power with unlimited storage, they suffer from specic shortcomings such as technical issues caused by data frac­tioning and data security. In this context, de-identication and patient anonymiza­tion concepts are of paramount importance that should be considered [86].
3.6 Model Evaluation
It is crucial to dene precise metrics to evaluate task performance in training deep learning models. Commonly used metrics in the classication tasks are accuracy, sensitivity, specicity, precision, and recall. F1 score combines precision and sensitivity.
The metrics are mainly used in detection and segmentation tasks to assess the similarity of an automatically created bounding box or a segmentation mask to pre­viously associated ground truth. Two common metrics are intersection over union (IOU) and Dice or Jaccard coefcients. IOU is measured by dividing the area delim­ited by the intersection of two bounding boxes and the union of the same two bound­ing boxes.
4 Conclusion
AI, and more specically machine learning, is revolutionizing healthcare and medi­cal care. Compared to other real-world problems, the progress of machine learning approach’s application is relatively slow in healthcare. This chapter introduced how a machine learning algorithm works and its requirements. This will provide researchers and clinicians basic understanding of these models for using them in research and the industry.
References
1. Hamblin MR.Shining light on the head: photobiomodulation for brain disorders. BBA Clin. 2016;6:113–24.
2. Chartrand G, Cheng PM, Vorontsov E, Drozdzal M, Turcotte S, Pal CJ, etal. Deep learning: a primer for radiologists. Radiographics. 2017;37(7):2113–31.
3. Wang S, Summers RM.Machine learning and radiology. Med Image Anal. 2012;16(5):933–51.
4. Jordan MI, Mitchell TM. Machine learning: Trends, perspectives, and prospects. Science. 2015;349(6245):255–60.
5. Kohli M, Prevedello LM, Filice RW, Geis JR.Implementing machine learning in radiology practice and research. Am J Roentgenol. 2017;208(4):754–60.
6. LeCun Y, Bengio Y, Hinton G.Deep learning. Nature. 2015;521(7553):436–44.
282
https://t.me/medicina_free
7. Litjens G, Kooi T, Bejnordi BE, Setio AAA, Ciompi F, Ghafoorian M, etal. A survey on deep learning in medical image analysis. Med Image Anal. 2017;42:60–88.
8. Ding Y, Sohn JH, Kawczynski MG, Trivedi H, Harnish R, Jenkins NW, etal. A deep learn­ing model to predict a diagnosis of Alzheimer disease by using 18F-FDG PET of the brain. Radiology. 2018;290(2):456–64.
9. Parakh A, Lee H, Lee JH, Eisner BH, Sahani DV, Do S.Urinary stone detection on CT images using deep convolutional neural networks: evaluation of model performance and generaliza­tion. Radiol Artif Intell. 2019;1(4):e180066.
10. Vos BD, Wolterink JM, Leiner T, Jong PA, Lessmann N, Išgum I.Direct automatic coronary calcium scoring in cardiac and chest CT.IEEE Trans Med Imaging. 2019;38(9):2127–38.
11. Schelb P, Kohl S, Radtke JP, Wiesenfarth M, Kickingereder P, Bickelhaupt S, etal. Classication of cancer at prostate MRI: deep learning versus clinical PI-RADS assessment. Radiology. 2019;293(3):607–17.
12. Lehman CD, Yala A, Schuster T, Dontchos B, Bahl M, Swanson K, etal. Mammographic breast density assessment using deep learning: clinical implementation. Radiology. 2018;290(1):52–8.
13. Mohammad-Rahimi H, Nadimi M, Ghalyanchi-Langeroudi A, Taheri M, Ghafouri-Fard S.Application of machine learning in diagnosis of COVID-19 through X-ray and CT images: a scoping review. Front Cardiovasc Med. 2021;8(185):638011.
14. Ngo TA, Lu Z, Carneiro G.Combining deep learning and level set for the automated segmenta­tion of the left ventricle of the heart from cardiac cine magnetic resonance. Med Image Anal. 2017;35:159–71.
15. Chang AC.Chapter 2- History of articial intelligence. In: Chang AC, editor. Intelligence­based medicine. London: Academic Press; 2020. p.23–7.
16. Turing AM.On computable numbers, with an application to the Entscheidungs problem. Proc Lond Math Soc. 1937;2(1):230–65.
17. Turing AM. Computing machinery and intelligence. Parsing the Turing test. New York: Springer; 2009. p.23–65.
18. Taulli T, Oni M.Articial intelligence basics. NewYork: Springer; 2019.
19. Luger GF. Articial intelligence: structures and strategies for complex problem solving. Pearson education; 2005.
20. Agah A. Introduction to medical applications of articial intelligence. Medical Applications of Articial Intelligence. 2013;19–26.
21. McCarthy J, Minsky ML, Rochester N, Shannon CE. A proposal for the Dartmouth summer research project on articial intelligence. 1955. http://www-formal.stanford.edu/jmc/history/
dartmouth/dartmouth.html.
22. Russell S, Norvig P.Articial intelligence: a modern approach; 2002.
23. Matheny M, Israni ST, Ahmed M, Whicher D. Articial intelligence in health care: the hope, the hype, the promise, the peril. NAM Special Publication. Washington, DC: National Academy of Medicine; 2019. p.154.
24. van Melle W.MYCIN: a knowledge-based consultation program for infectious disease diagno­sis. Int J Man Mach Stud. 1978;10(3):313–22.
25. Zhou L, Sordo M.Chapter 5 - Expert systems in medicine. In: Xing L, Giger ML, Min JK, editors. Articial intelligence in medicine. London: Academic Press; 2021. p.75–100.
26. Ren R, Luo H, Su C, Yao Y, Liao W.Machine learning in dental, oral and craniofacial imaging: a review of recent progress. PeerJ. 2021;9:e11451.
27. Grosan C, Abraham A.Intelligent systems. Cham: Springer; 2011.
28. Baxt WG.Use of an articial neural network for the diagnosis of myocardial infarction. Ann Intern Med. 1991;115(11):843–8.
29. Mohammad-Rahimi H, Nadimi M, Rohban MH, Shamsoddin E, Lee VY, Motamedian SR.Machine learning and orthodontics, current trends and the future opportunities: a scoping review. Am J Orthod Dentofacial Orthop. 2021;160(2):170–92.e4.
30. Khanna SS, Dhaimade PA.Articial intelligence: transforming dentistry today. Indian J Basic Appl Med Res. 2017;6(3):161–7.
S. R. Motamedian et al.
Brief Introduction toArticial Intelligence andMachine Learning
https://t.me/medicina_free
31. Stheeman SE, van der Stelt PF, Mileman PA.Expert systems in dentistry. Past performance-
-future prospects. J Dent. 1992;20(2):68–73.
32. White SC.Computer-aided differential diagnosis of oral radiographic lesions. Dentomaxillofac Radiol. 1989;18(2):53–9.
33. Hyman JJ, Diehl MC.A dental trauma diagnostic program. Proc Annu Symp Comput Appl Med Care. 1983:133–4.
34. Hyman JJ, Doblecki W. Computerized endodontic diagnosis. J Am Dent Assoc. 1983;107(5):755–8.
35. Sims-Williams JH, Brown ID, Matthewman A, Stephens CD.A computer-controlled expert system for orthodontic advice. Br Dent J. 1987;163(5):161–6.
36. Abbey LM.An expert system for oral diagnosis. J Dent Educ. 1987;51(8):475–80.
37. Bayaraa T, Hyun CM, Jang TJ, Lee SM, Seo JK.A two-stage approach for beam hardening artifact reduction in low-dose dental CBCT.IEEE Access. 2020;8:225981–94.
38. Lee J-H, Kim D-h, Jeong S-N, Choi S-H.Diagnosis and prediction of periodontally compro­mised teeth using a deep learning-based convolutional neural network algorithm. J Periodontal Implant Sci. 2018;48(2):114–23.
39. Mohammad-Rahimi H, Motamadian SR, Nadimi M, Hassanzadeh-Samani S, Minabi MAS, Mahmoudinia E, etal. Deep learning for the classication of cervical maturation degree and pubertal growth spurts: a pilot study. Korean J Orthod. 2022;52(2):112–22.
40. Mohammad-Rahimi H, Motamedian SR, Rohban MH, Krois J, Uribe S, Mahmoudinia E, etal. Deep learning for caries detection: a systematic review: DL for caries detection. J Dent. 2022;122:104115.
41. Saghiri MA, Asgar K, Boukani KK, Lot M, Aghili H, Delvarani A, etal. A new approach for locating the minor apical foramen using an articial neural network. Int Endod J. 2012;45(3):257–65.
42. Kim DW, Kim H, Nam W, Kim HJ, Cha IH.Machine learning to predict the occurrence of bisphosphonate-related osteonecrosis of the jaw associated with dental extraction: a prelimi­nary report. Bone. 2018;116:207–14.
43. Aliaga I, Vera V, Vera M, García E, Pedrera M, Pajares G.Automatic computation of mandibu­lar indices in dental panoramic radiographs for early osteoporosis detection. Artif Intell Med. 2020;103:101816.
44. Lee KS, Kwak HJ, Oh JM, Jha N, Kim YJ, Kim W, etal. Automated detection of TMJ osteo­arthritis based on articial intelligence. J Dent Res. 2020;99(12):1363–7.
45. Setzer FC, Shi KJ, Zhang Z, Yan H, Yoon H, Mupparapu M, etal. Articial intelligence for the computer-aided detection of periapical lesions in cone-beam computed tomographic images. J Endod. 2020;46(7):987–93.
46. Alabi RO, Elmusrati M, Sawazaki-Calone I, Kowalski LP, Haglund C, Coletta RD, et al. Machine learning application for prediction of locoregional recurrences in early oral tongue cancer: a Web-based prognostic tool. Virchows Arch. 2019;475(4):489–97.
47. Alhazmi A, Alhazmi Y, Makrami A, Masmali A, Salawi N, Masmali K, etal. Application of articial intelligence and machine learning for prediction of oral cancer risk. J Oral Pathol Med. 2021;50(5):444–50.
48. Xie X, Wang L, Wang A.Articial neural network modeling for deciding if extractions are necessary prior to orthodontic treatment. Angle Orthod. 2010;80(2):262–6.
49. Murphy KP.Machine learning: a probabilistic perspective. Cambridge: MIT Press; 2012.
50. Ho T.Random decision forests. In: International conference on document analysis and recog­nition, Montreal; 1995.
51. Prinzie A, Van den Poel D, editors. Random multiclass classication: generalizing random forests to random MNL and random NB.In: International conference on database and expert systems applications. Cham: Springer; 2007.
52. Chilamkurthy S, Ghosh R, Tanamala S, Biviji M, Campeau NG, Venugopal VK, etal. Deep learning algorithms for detection of critical ndings in head CT scans: a retrospective study. Lancet. 2018;392(10162):2388–96.
283
284
https://t.me/medicina_free
53. Lindsey R, Daluiski A, Chopra S, Lachapelle A, Mozer M, Sicular S, etal. Deep neural net­work improves fracture detection by clinicians. Proc Natl Acad Sci. 2018;115(45):11591–6.
54. Erickson BJ, Koratis P, Akkus Z, Kline TL. Machine learning for medical imaging. Radiographics. 2017;37(2):505–15.
55. Hastie T, Tibshirani R, Friedman J.Statistical learning: data mining, inference, and prediction. Heidelberg: Springer; 2009.
56. Silver D, Huang A, Maddison CJ, Guez A, Sifre L, Van Den Driessche G, etal. Mastering the game of Go with deep neural networks and tree search. Nature. 2016;529(7587):484–9.
57. Kidoh M, Shinoda K, Kitajima M, Isogawa K, Nambu M, Uetani H, etal. Deep learning based noise reduction for brain MR imaging: tests on phantoms and healthy volunteers. Magn Reson Med Sci. 2020;19(3):195.
58. Rathi VGP, Palani S.Brain tumor detection and classication using deep learning classier on MRI images. Res J Appl Sci Eng Technol. 2015;10(2):177–87.
59. Yang S, Kweon J, Roh J-H, Lee J-H, Kang H, Park L-J, etal. Deep learning segmentation of major vessels in X-ray coronary angiography. Sci Rep. 2019;9(1):16897.
60. Rostami B, Anisuzzaman DM, Wang C, Gopalakrishnan S, Niezgoda J, Yu Z. Multiclass wound image classication using an ensemble deep CNN-based classier. Comput Biol Med. 2021;134:104536.
61. Sakellaropoulos T, Vougas K, Narang S, Koinis F, Kotsinas A, Polyzos A, etal. A deep learning framework for predicting response to therapy in cancer. Cell Rep. 2019;29(11):3367–73.e4.
62. Harvey H, Glocker B.A standardised approach for preparing imaging data for machine learn­ing tasks in radiology. In: Ranschaert ER, Morozov S, Algra PR, editors. Articial intelligence in medical imaging: opportunities, applications and risks. Cham: Springer; 2019. p.61–72.
63. Gebru T, Morgenstern J, Vecchione B, Vaughan JW, Wallach H, Iii HD, Crawford K. Datasheets for datasets. Communications of the ACM. 2021;64(12):86–92.
64. Nelson GS. Practical implications of sharing data: a primer on data privacy, anonymization, and de-identication. InSAS global forum proceedings 2015;1–23.
65. Neubauer T, Heurix J.A methodology for the pseudonymization of medical data. Int J Med Inform. 2011;80(3):190–204.
66. Tang A, Tam R, Cadrin-Chênevert A, Guest W, Chong J, Barfett J, etal. Canadian Association of Radiologists white paper on articial intelligence in radiology. Can Assoc Radiol J. 2018;69:120–35.
67. Soffer S, Ben-Cohen A, Shimon O, Amitai MM, Greenspan H, Klang E.Convolutional neural networks for radiologic images: a radiologist’s guide. Radiology. 2019;290(3):590–606.
68. Papandreou G, Chen L-C, Murphy KP, Yuille AL.Weakly-and semi-supervised learning of a deep convolutional network for semantic image segmentation. In: 2015 IEEE international conference on computer vision (ICCV); 2015. p.1742–50.
69. Wang Y, Yao Q, Kwok JT, Ni LM. Generalizing from a few examples: A survey on few-shot learning. ACM computing surveys (csur). 2020;53(3):1–34.
70. Paul R, Hawkins SH, Balagurunathan Y, Schabath M, Gillies RJ, Hall LO, etal. Deep feature transfer learning in combination with traditional features predicts survival among patients with lung adenocarcinoma. Tomography. 2016;2(4):388–95.
71. Deng J, Dong W, Socher R, Li L-J, Li K, Fei-Fei L, editors. ImageNet: a large-scale hierarchi­cal image database. In: 2009 IEEE conference on computer vision and pattern recognition. IEEE; 2009.
72. Abdulazeez AM, Salim BW, Zeebaree DQ, Doghramachi D. Comparison of VPN Protocols at Network Layer Focusing on Wire Guard Protocol. iJIM. 2020;14(18):157.
73. Acharya MS, Armaan A, Antony AS, editors. A comparison of regression models for predic­tion of graduate admissions. In: 2019 International conference on computational intelligence in data science (ICCIDS). IEEE; 2019.
74. Zhang Z, Li Y, Li L, Li Z, Liu S, editors. Multiple linear regression for high efciency video intra coding. In: ICASSP 2019-2019 IEEE international conference on acoustics, speech and signal processing (ICASSP). IEEE; 2019.
75. Schmidhuber J.Deep learning in neural networks: an overview. Neural Netw. 2015;61:85–117.
S. R. Motamedian et al.
Brief Introduction toArticial Intelligence andMachine Learning
https://t.me/medicina_free
76. Hochreiter S, Schmidhuber J.Long short-term memory. Neural Comput. 1997;9(8):1735–80.
77. Cho K, van Merrienboer B, Gulcehre C, Bahdanau D, Bougares F, Schwenk H, Bengio Y. Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation. arXiv e-prints. 2014 Jun:arXiv-1406.
78. Goodfellow I, Pouget-Abadie J, Mirza M, Xu B, Warde-Farley D, Ozair S, Courville A, Bengio Y. Generative adversarial networks. Communications of the ACM. 2020;63(11):139–44.
79. Luc P, Couprie C, Chintala S, Verbeek J. Semantic Segmentation using Adversarial Networks. arXiv e-prints. 2016 Nov:arXiv-1611.
80. Savioli N, Silva Vieira M, Lamata P, Montana G. A Generative Adversarial Model for Right Ventricle Segmentation. arXiv eprints. 2018 Sep:arXiv-1810.
81. Ronneberger O, Fischer P, Brox T, editors. U-Net: convolutional networks for biomedical image segmentation. In: International conference on medical image computing and computer­assisted intervention. Cham: Springer; 2015.
82. Goodfellow I, Bengio Y, Courville A.Deep learning. London: MIT Press; 2016.
83. Abadi M, Agarwal A, Barham P, Brevdo E, Chen Z, Citro C, Corrado GS, Davis A, Dean J, Devin M, Ghemawat S. TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems. arXiv e-prints. 2016 Mar:arXiv-1603.
84. Paszke A, Gross S, Massa F, Lerer A, Bradbury J, Chanan G, etal. PyTorch: An imperative style, high-performance deep learning library. Adv Neural Inf Process Syst. 2019;32:8026–37.
85. Raina R, Madhavan A, Ng AY. Large-scale deep unsupervised learning using graphics pro­cessors. InProceedings of the 26th annual international conference on machine learning 2009;873–80.
86. Ardagna CA, Asal R, Damiani E, Vu QH. From security to assurance in the cloud: A survey. ACM Computing Surveys (CSUR). 2015;48(1):1–50
285
Application ofArtificial Intelligence
https://t.me/medicina_free
inDiagnosing Oral andMaxillofacial Lesions, Facial Corrective Surgeries, andMaxillofacial Reconstructive Procedures
ParisaMotie, GhazalHemmati, ParhamHazrati, MasihLazar, FatemehAghajaniVarzaneh, HosseinMohammad-Rahimi, MohsenGolkar, andSaeedRezaMotamedian
P. Motie DDS, Medical Image and Signal Processing Research Center, Isfahan University of Medical Sciences, Isfahan, Iran
G. Hemmati Dental Research Center, Research Institute of Dental Science, Shahid Beheshti University of Medical Sciences, Tehran, Iran
P. Hazrati Student Research Committee, School of Dentistry, Shahid Beheshti University of Medical Sciences, Tehran, Iran
M. Lazar Shahid Beheshti University of Medical Sciences, Tehran, Iran
F. A. Varzaneh Student Research Committee, School of Dentistry, Shahid Beheshti University of Medical Sciences, Tehran, Iran
H. Mohammad-Rahimi Topic Group Dental Diagnostics and Digital Dentistry, ITU/WHO Focus Group AI on Health, Berlin, Germany
Department of Computer Engineering, Sharif University of Technology, Tehran, Iran
M. Golkar Oral and Maxillofacial Surgery Resident, Dental School, Shahid Beheshti University of Medical Sciences, Tehran, Iran
S. R. Motamedian ( Topic Group Dental Diagnostics and Digital Dentistry, ITU/WHO Focus Group AI on Health, Berlin, Germany
Dentofacial Deformities Research Center, Research Institute of Dental Sciences, Shahid Beheshti University of Medical Science, Tehran, Iran e-mail: drmotamedian@gmail.com
*)
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 A. Khojasteh et al. (eds.), Emerging Technologies in Oral and Maxillofacial Surgery, https://doi.org/10.1007/978-981-19-8602-4_15
287
288
https://t.me/medicina_free
P. Motie et al.
1 Introduction
The term articial intelligence (AI) refers to the use of computers to simulate intel­ligent behavior with minimal human input [1]. General healthcare delivery can be achieved with two types of AI: physical and virtual. Sophisticated robots or auto­mated robotic arms represent physical applications and software-type algorithms are virtual components that can be used to support clinical decision-making [2]. Rapid developments of technologies in the medical imaging eld prompt a large amount of visual data. The collection, analysis, and application of such vast amounts of data have posed challenges for modern medicine in solving clinical problems. The leading strategy to overcome this complexity is employing medical articial intelligence, nonhuman intelligence systems which are designed to assist clinicians with the formulation of diagnosis, therapeutic decision-making, and predicting treatment outcomes [3].
CNN is a subcategory of the deep learning (DL) models. The special construc­tion of CNN has made it a superior model for image processing; the neurons’ links signicantly reduce the computational overload. CNN can instantly learn and iden­tify the images’ patterns. It can also detect, classify, and segment objects [4, 5]. “Classication” ability of CNN models can be used in categorizing target lesions in radiographs. “Segmentation” of the anatomic landmarks is one of the essential image processing steps. In the former AI models, segmentation had to be done man­ually by a radiologist. In contrast, CNN can automatically segment organs or lesions and dramatically reduce the burden of clinician work. The self-regulated “detec­tion” task of CNN makes large-scale health screenings more enforceable. Kooi etal. conducted a study to evaluate the performance of CNN models in large-scale (45,000) mammographic screening. There was no signicant difference between CNN and specialized radiologist readers [6, 7].
2 AI Applications inRecent Medicine
2.1 AI inRadiology
The base of medical radiology is extracting the images’ essential features. In the recent years, the DL pattern extraction potential has pitched in, saving up a vast amount of time by automating the image analysis [8, 9]. According to a recent study on 874 chest radiographies (CXR), the DL algorithm might be a helpful tool in interpreting CXR ndings such as pulmonary opacities, hilar prominence, cardio­megaly, and pleural effusion. In addition, changes or stability of these ndings can be evaluated on follow-up CXR.It was also concluded that clinical staff might be replaced with AI if they are unavailable [10]. In detecting COVID-19, DL is one of the benecial tools for triaging patients and differentiating them from other types of pneumonia based on CXR [1113]. Moreover, DL can help the clinicians in the detection of pulmonary nodules [1417], lung cancer [16, 18, 19], pneumonia and
Application of Articial Intelligence in Diagnosing Oral and Maxillofacial Lesions…
https://t.me/medicina_free
distinguishing bacterial from viral one [2022], referable thoracic abnormalities [23], pulmonary tuberculosis (PTB) [24, 25], and pneumoconiosis [26].
Besides the respiratory system, DL can also assist clinicians in other elds like bone age assessment out of wrist and hand radiographs [27], predicting the presence of coronary artery calcium (CAC) correlated with the risk of cardiovascular disease [28], and indicating vertebral fracture [29].
289
2.2 AI inOncology
ExPecto, a DL-based framework, is organized to accurately predict tissue-specic transcriptional effects of mutations from DNA sequences, even those that are rare or not seen yet; therefore, this algorithm makes it possible to predict the risk of expres­sion and mutation disease effects [30]. The Watson for Oncology (WFO) AI system is designed to support clinicians in planning treatment for breast cancer [31]. In detecting lymph node metastasis in women with breast cancer, another DL algo­rithm proved better diagnostic performance than 11 pathologists [32]. DL can also help the cancer team supply clinical suggestions for colorectal cancer [33]. Nasal basal cell carcinoma is a complex disorder that needs a multidisciplinary team (MDT) for treatment and patients should be triaged for Mohs micrographic surgery (MMS). Machine learning approach is a helpful tool guiding the MDT with patient selection and predicting the need of MMS, which reduces the waste of time and nancial burden and improves patient care [34].
2.3 AI inOphthalmology
Diabetic retinopathy (DR) is one of the most common preventable causes of blind­ness in the world. The AI algorithm can recognize the cases that should be referred to the ophthalmologists with high reliability [35, 36]. Patients suffering from age­related macular degeneration (AMD) experience a loss of vision in the center of the visual eld because of damage to the retina [37]. It’s been reported that AI can detect the referable AMD patients and can perform as well as human experts in its management and risk assessment [3842]. AI also has a comparable performance to trained specialists in the detection of retinopathy of prematurity [4345] and glau­coma [4648].
2.4 AI inCardiology
In 1985, Willems etal. took the study in which they compared the ECG interpret­ability of 9 electrocardiographic computer programs with 8 cardiologists in 1220 cases of various cardiac disorders. They gured out that some of the programs, the best once for sure, have the same interpretation capacity as the cardiologist in seven major cardiac diseases, including left ventricular hypertrophy, right ventricular