{"paper_id":"d662d624-2713-4982-a543-51218d1178e7","body_text":"Nahid: AI-based Algorithm for operating fully-automatic surgery \n Sina Saadati1   \n \n \nAbstract 1   \nIn this paper, for the first time, a method is presented \nthat can provide a fully automated surgery based on \nsoftware and computer vision techniques. Then, the \nadvantages and challenges of computerization of \nmedical surgery are examined. Finally, the surgery \nrelated to isola ted ovarian endometriosis disease has \nbeen examined, and based on the presented method, a \nmore detailed algorithm is presented that is capable \nof automatically diagnosing and treating this disease \nduring surgery as proof of our proposed method \nwhere a U -net is trained to detect the endometriosis \nduring surgery.  \nKeywords: Deep Learning, AI in Medicine, Image \nProcessing, Computerizing Surgery, AI.  \n1 Introduction \nNowadays, computer technology and artificial \nintelligence have revolutionized all aspects of the \nscience. The computerization of scientific, industrial, \nclinical, and commercial processes leads to speeding \nup the process, increasing accuracy, and reducing \ncosts. In the field of medicine, artificial intelligence \nmethods have been able to perform on par  with or \neven better than humans in many clinical purposes. \nThe introduction of artificial intelligence into surgery \ncan increase the speed of surgery and thus reduce \nthe amount of bleeding and the risk of infection. This \ngoal can be achieved in two ways, fully automatic or \ncomputer-assisted way.  \n \nComputer methods based on software, high \ncomputing power, and artificial intelligence have \nsignificantly improved human ability. Today, \ncomputers can process very large amounts of data \nand information at a very h igh speed and with \nsufficient accuracy. This feature includes image \nprocessing and machine vision algorithms. In such a \nway artificial neural networks (ANN) that try to \nsimulate the brain of humans or animals can see the \n                                                                    \n1 Department of Computer Engineering, Amirkabir \nUniversity of Technology, Tehran, Iran, Sina.Saadati@aut.ac.ir   \n \n \nimages and conceptualize them. Henc e, this category of \ncalculations is also called deep learning. The \nintroduction of technology and specifically, computers to \nmedicine and surgery makes it possible for doctors and \nresearchers to carry out scientific, research or clinical \nprocesses with greater speed and less cost.[8,9,10,12,13] \n \nIn this article, focusing on the surgery for isolated \novarian endometriosis, a method is provided that \nenables fully automatic surgery by computer and \nartificial intelligence. Endometriosis is a common \ndisease among women that can lead to pain in the \nabdominal and pelvic areas and increase the risk of \ninfertility. In this disorder, tissue called endometrioma \ngrows abnormally around the uterus. This phenomenon \ncan be seen in different areas such as ovaries, uterus, \nintestines, etc. I solated ovarian endometriosis is a type \nof disease in which endometrioma tissue is observed \nonly in the ovary. This type of disorder is statistically \nless common than other conditions in which \nendometrioma tissue is found in several places. This \ndisorder can be detected using methods such as the use \nof ultrasound waves or magnetic resonance imaging \n(MRI) that are performed before surgery. Also, the \ndiagnosis and treatment of this disease can be done by \nminimally invasive surgery or laparoscopy. In some \ncases, endometrioma tissue is removed by thermal \nwaves or laser during surgery. Recent research has \nshown that using a laser for removing endometriosis has \nfewer risks compared to other methods. [11] \n \nLaparoscopy or minimally invasive surgery is a modern \nmethod in which only limited areas of the body are split \nand through which tools are inserted into the body to \nwatch, hold, observe, or cut inner organs. This surgical \nmethod has an accuracy equal to or greater than the \ntraditional surgical method, and since the con nection \nbetween inside and outside spaces of the body is \nminimized, it leads to a reduction in the risk of infection. \nIt has also advantages like less bleeding and pain. Today, \nthis method has been combined with robotic technology \nand specialists are able to operate on patients from a \nlong geographical distance. This article will take another \nstep in the field of modern surgery. In the second \nchapter, the works related to this research are briefly \nintroduced. Then, in the third chapter, a comprehensive \nmethod for fully automatic surgery will be presented. \nSoftware architecture and requirements engineering are \n\n  \n \ndescribed in this section. In the fourth chapter, the \nopportunities and challenges of this method will be \ndiscussed. In the fifth chapter, isolated ova rian \nendometriosis is focused, and based on the proposed \nmethod, a more detailed algorithm will be described \nto perform the surgery in a fully automatic way. Since \nthe algorithm presented in this section is based on \nmachine learning and computer vision met hods, a \nneural network model is introduced that is capable of \ndiagnosing and treating isolated ovarian \nendometriosis. Finally, in the sixth chapter, the \nconclusion will be discussed.    \n  \n    \n2 Preliminaries \nIn recent years, modern technologies have enabled \nsignificant growth in the fields of biology, medicine, \nand surgery. Nezhat et al. have examined the recent \nadvances in engineering sciences, especially artificial \nintelligence and energy production. Then, they \npredicted that in 2050, a large fraction of su rgical \nprocedures will be performed by computers. In this \nresearch, obstacles such as the possibility of bias of \npatients and even the community of doctors and \nsurgeons with the introduction of technology into \nmedicine have been pointed out. [5] In a resea rch \nfocusing on minimally invasive surgery and robotic \nsurgery, Mascagni et al. pointed out the importance of \nmachine vision issues and the availability of image \ndata to advance medical research. He has classified \nthe methods of computerization of medical services; \nHowever, surgery is not mentioned in it as  a \ncompletely automatic method. [3] Maddzadeh et al. \nhave presented a dataset that can be used in training \nmachine learning models and especially artificial \nneural networks. The purpose of this data colle ction is \nonly the production of intelligent assi stants in \nmedicine and surgery. [2] In a research, Scheikl et al. \ninvestigated different architectures for the semantic \nsegmentation of images related to endometriosis. In \nthis research, in order to meet the minimum speed of \ncalculations, the processing of each image in a period \nof less than forty milliseconds is recommended. This \ncharacteristic leads to the image processing in real -\ntime mode with the use of the surgeon's assistant. [6] \nVisalaxi et al. reporte d a research, using ResNet50 \narchitecture to produce an artificial intelligence \nmodel that can detect the presence or absence of \nendometriosis disorder in images; This model only \nresponds in a binary way and does not provide the \npossibility of localizing t he endometrioma tissue. [7] One \nof the challenges of this research is that endometriosis \ndisorder is considered an unclassifiable disease. This \ndisease has different types that are treated in different \nways. Referring to this point, in addition to facilita ting the \nprocess of training and evaluating artificial intelligence, it \ncan lead to an increase in the accuracy of the results. In a \nresearch, Hong et al. have labeled several sets of data \nrelated to laparoscopic images. In this research, the \nclassification and categorization of data has received less \nattention. [1] In a study, Naqvi et al. suggested that \nresearchers use the Glenda dataset, which includes \nlaparoscopic images related to endometriosis. [4] In \naddition to the fact that this dataset presents a very small \nand limited number of images (less than 400 samples), no \ndifference has been made between the location of \nendometriosis. So far, it can be observed that in only \ntwenty-six samples there is a possibility of watching \novary and in this subgroup, on ly sixteen cases include \nendometrioma located on the ovary. In the present study, \nin addition to Glenda's data, other data including ovarian \nendometriosis have been used.   \n \n3 Nahid: Artificial Intelligence \ncapable  of managing surgery \nIn this section, an AI -based algorithm named Nahid is \npresented for the complete automatic operation and \nmanagement of surgery. This algorithm is implemented \nbased on Sina algorithms and Sina tree data structure. To \nprovide a suitable and efficient algorithm that enables a \ncomputer to operate a surgery, we should first consider \nhow humans are capable of doing surgery. What leads a \nperson to perform surgery is related to his knowledge \nand vision. In this claim, knowledge means the human \nability to diagnose and distinguish organs and disorders. \nVision is also an essential factor in the surgery. Using the \ncombination of vision and knowledge, experts are able to \nidentify the abnormalities in the body and remove them. \nAs a result, to produce artificial intelligence that can \nappear as a surgeon or surgeon assistant, it is necessary \nto produce these two features in the form of software. In \nthis section, different methods will be presented to \nexplain how to inject the knowledge into the computer. \n \nThe first method, which seems perfect and  challenging to \ndevelop, is to produce software for modeling and \nsimulating the human body. It means software that \nsimulates all the patient's organs by receiving the \ncharacteristics of the patient, such as age, gender, \nweight, disorders, etc. The tool sho uld be able to provide \n\n  \n \nmobility and movement of organs virtually to the \nuser. Also, the developed software should be able to \nperform routing safely and optimally. For example, \ncalculate the shortest path to reach from the belly \nbutton to the uterus in such  a way that the least \ndamage is done to the internal organs. In this \nmethod, features such as the color of the internal \norgans, the elasticity of the organs, and physical \nactions like resistance to cutting and heat should be \nconsidered. The difficulty of p hysical equations \nrelated to elastic organs and their computational \ncomplexity makes this method challenging and hard \nto develop. Another method is to probabilistically \ndefine points of the body as a three -dimensional \ngraph; So that each point expresses th e probability of \npresence of a certain part of the body. For example, \nin the coordinates like (x,y,z) the heart or uterus can \nbe found with high probability. This is useful where \nthe camera (or robotic arms) are located inside the \nbody and computer needs t o find out what is \nprobable to be seen(or grasped).  In any case, we call \nthe software that enables the positioning of the \norgans and their related disorders by considering the \nconditions \"human geometric model\". In addition to \nthe human geometric model, i t is necessary for the \ncomputer to be aware of the necessary steps to \nmanage surgery. This can be defined by an algorithm \nand implemented by a computer. These capabilities \nmake it possible to inject the knowledge needed for \nsurgery into the computer. \n \nIn addition to knowledge, it is necessary to develop \nthe necessary AI models to conceptualize and \nrecognize different organs in different conditions. In \nthis research, unlike other deep learning models, it is \nsuggested that researchers separate the artific ial \nintelligence models based on the surgical states. \nGlenda's data collection includes a small collection of \nimages that are prepared during minimally invasive \nsurgery for endometriosis patients. Endometriosis is \na disorder that can occur in different are as that have \na different appearance and geometry. Figure 1 \nseparates these areas. By classifying the disorder of \nendometriosis based on its appearance location, we \nwill be able to collect images that are less different in \nterms of geometry and appearance. In this way, this \nopportunity is provided to train a model with much \nhigher accuracy even with limited data sets. With this \napproach, the time and energy required for training \nand test processes, as well as the application of \nartificial intelligence models  for image processing, \nwill be optimized. The separation of data and artificial \nintelligence models based on categories of disorders not \nonly leads to the optimization of technical criteria but \nalso increases the transparency of the behavior of neural \nnetwork models. This point is important from the point \nof view that in medical applications, having black box \nsoftware components due to the importance of the \ntreatment processes can lead to questioning the whole \ntreatment method. In this research, we call thi s \nsuggestion the \"principle of situation separation\".   \n \n \nFigure 1, Endometriosis probable locations. \n \nComputer vision refers to the use of artificial neural \nnetwork models that are capable of performing \nprocessing to understand and separate image content  \nbased on boundaries, meaning, etc. For this purpose, we \nneed a dataset including normal images and labeled \nmasks. For example, in Figure 2, a pair of images can be \nseen including a real laparoscopic ima ge and its \nassociated labels. [1] This image, provided by Hong et al., \nprovides the ability to train a neural network and use it \nto understand and segment the newer images. In this \nresearch, based on the principle of situation separation, \nit is recommended that each stage of surgery should \nhave an independen t model that is trained based on \nseparate datasets that correspond to the same state and \nlocation of surgery.     \n \na)   b)  \nFigure 2, An image segmentation example[1] \n \nObserving the situation separation principle has \nanother advantage, which is related to t he risk of data \noverfitting. In their research, Nazhat et al. pointed out \nthe difference between treatment and surgery \n\n\n  \n \nmethods between two patients suffering from one \ndisease with similar features and, as a result , \nrecommended standardization.[5] In this r esearch, \ndue to the fact that robotic surgery has provided \nthe ability to manage surgery for a wide range of \ndisorders, we suggest that certain and safe paths \nfor the movement of the camera or robotic arms \nshould be considered based on the human \ngeometric model. Considering this standardization, \nwe will be able to observe the inner organs from \nthe same angle at each state of the surgery. It is \nimportant to consider other characteristics such as \nage, weight, place of disorder, etc. One of the \nadvantages of f ollowing this principle is the \nminimization of the difference between training \nand test data. With these conditions, overfitting in \nthe process of machine learning will not be \ndangerous.      \n \nIn the following, with the aim of combining the \nhuman geometric  model with machine vision, a \ncomputational architecture is presented that can \nmanage surgery. First, an algorithm is presented \nthat uses several machine vision models to detect \nand understand images. Then, the surgical \nmanagement will be explained. \n \nSina Algorithm: \n     Input: an imae of minimally invasive surgery    \n     (1) along with the operation situation that       \n      includes the exact location of the camera in  \n      the body (2). \n      Output: detection of all the elements in the  \n       image. \n1- Based on the computer  vision model, which \nis related to the surgery situation, proceed to the \nsemantic segmentation of the image and save \nthe result as SegmentedImg. \n2- Using a suitable algorithm, the borders in the \nimage are detected and saved in th e form of an \nimage called EdgeDetectedImg. \n3- Due to the high accuracy of EdgeDetectedImg \ncompared to the borders found in \nSegmentedImg, the borders in EdgeDetectedImg \nshould be considered as real borders. \n4- For each area detected in EdgeDetectedImg, \nthe pixels in SegmentedImg should be \nseparated. Then, based on the highest number of \npixels in each area, type of  the area s must be \nlabeled.   \n5- Save the result as the output of the algorithm. \nIn order to increase the processing speed, the Sina \nalgorithm can be executed in a parallel way. Figures 3, \na, and 3, b show the second step of Sina's algorithm. As \nyou can see, it is possible to demarcate organs using \nedge detection. However, edge detecti on algorithms \nare only able to recognize the edges in an image. To \nrecognize and understand the type of each organ, it is \nnecessary to use the appropriate neural network \nmodel. For this purpose, in the third step of Sina's \nalgorithm, based on the physical position of the \ncamera(situation of the surgery), the appropriate \nmachine vision model is selected and labeling is done \naccording. In this algorithm, edge detection is used to \nremove the predictable noises. Finally, the computer \ncan understand and separate  the organs and their \nrelated disorders well at every stage of the surgery. \nThe proposed software architecture for implementing \nthe Sina algorithm is shown in Figure 4.    \n \na)  b)  \nFigure 3, Edge Detection for finding borders. \n \n \nFigure 4, Proposed architecture for executing Sina \nAlgorithm \n \nBecause in each stage of surgery according to the \nprinciple of state separation, it is essential to use an \nindependent computer vision model, it is vital to store \nthese models in a suitable data structure. In this \nresearch, a tree data structure is presented which tries \nto implement the human geometric model and respect \nthe separation principle at the same time. We call this \ndata structure the Sina tree. Each node of this building \nhas a three -dimensional coordinate that  is related to \nthe position and angle of the camera in the body. In \nother words, all the images obtained from minimally \ninvasive surgery can be divided based on the location \nand angle of the camera, and then the training and \n\n\n  \n \nproduction of independent compu ter vision \nmodels can be done. Finally, each model is stored \nin a corresponding node in the graph. Also, the \nedges between the nodes are considered the path \nof the camera movement. By implementing the \nSina tree, we will have much simpler, clearer, more \naccurate, and optimal models, and at the same \ntime, it is possible to move from one position (or \nsurgery situation) to another position of the body \n(or surgery situation). It can be easily done by \nfollowing the graph edges. The existence of the \nSina tree can also require compliance with the \napplied standards. Figure 5 shows an example of a \nSina tree. Based on this data structure, the camera \nis inserted into the abdomen from the navel area \nand can reach the uterus or ovaries by navigating a \ndirect path. It is p ossible to increase the accuracy \nand security of the operation by creating \nintermediate nodes on each edge. In this image, \nnodes are marked with red color, and edges are \nshown as green lines.   \n \n \nFigure 5, An instance of Sina Tree data structure \n \n \n4 Challenges and Advantages \nIn this section, the advantages and challenges of \nsurgery with the help of artificial intelligence from \na social, research, education, and clinical services \nperspective will be examined. Fortunately, the \nrapid growth of computer technolo gy and artificial \nintelligence has been widely welcomed by society. \nHowever, based on the sensitivities in treatment \nissues, a wide range of patients may resist \naccepting treatment with the help of artificial \nintelligence. To overcome this challenge, it is  \nrecommended that artificial intelligence \napplications have an expert supervising the \ntreatment and surgery process. This approach can \nalso resolve legal issues related to treatment. It is \nsuggested that international institutions should also \ntake necessary measures to promote public culture to \ntrust technology even in medical matters so that the \nuntrue fear of this technology can be removed. \nComputer error is much lower than humans, and as a \nresult, trusting artificial intelligence seems logical. \nHowever, in this context, the correctness of artificial \nintelligence software and models must be evaluated \nand proven. This issue requires more research in this \nfield. The results of this research will be effective not \nonly in medical services and clinics but also in \neducation. As claimed by Mascagni et al., many images \nand datasets related to the training of neural network \nmodels can be used for training by institutions and \nuniversities. Also, models and algorithms related to \nautomatic surgery can be used to produc e suitable \nsimulators for training or research. Finally, by using \nthe Nahid algorithm, we can increase the speed of \nsurgery and reduce the error. With this measure, \ntreatment costs will be also minimized.   \n \nOne of the important challenges can be seen in the \ntraining of neural networks. In this regard, there is a \nstrong need for data and images related to minimally \ninvasive surgery. Unfortunately, the presented \ndatasets do not provide enough information and ha ve \nnot followed the principle of situation separation. In \nthis context, it should be noted that the imaging \nsettings and configuration during minimally invasive \nsurgery must be accurately recorded. For example, the \nintensity of light or the type of light u sed during \nimaging must be recorded.     \n \n \n5 Isolated ovarian endometriosis \nsurgery using the presented \nalgorithm  \nIn this section, a special form of Nahid's  algorithm for \nsurgery of isolated ovarian endometriosis disorder is \npresented. This algorithm is described using the \nflowchart in Figure 6. In this process, the way to use \nthe entities related to Nahid algorithm, such as Sina \nalgorithm or Sina tree, are d epicted with appropriate \ndetails. Also, the Sina tree for this type of surgery is \ndrawn in Figure 7. In this diagram, each state is \ncomprehensively expressed by a three -row \nrectangular. The first row of each column indicates the \nstate or stage of the surge ry, the second row includes \nthe location of the camera, and the third row indicates \nthe task that artificial intelligence must perform at \n\n\n  \n \nthat state. This chart will contain much more detail \nin a real problem. For example, it is necessary to \nsave the position of the robot's arms in addition to \nthe camera position.  \nAs can be seen, in this process, three nodes are \nconsidered to move from the navel area to the \npatient's left ovary. Each node includes a powerful \ncomputer vision model that is able to specifical ly \ndetect all organs or disorders in that state. Based \non the knowledge gained from the computer vision \nmodel and the Sina algorithm, as well as the Sina \ntree data structure, it is possible to move forward \nor backward in the body during the operation \neasily. This algorithm can be designed in other \nways for other kinds of diseases, But the principles \nof its definition and design are the same as what \nwas explained in this article. \nOne of the most important steps in this process is \nrelated to the diagnosis of endometrioma tissue \nlocated in the ovary. The recognition of this tissue \nby computer vision makes the robot behave \ncorrectly in targeting the radiation of thermal \nwaves or lasers. In this research, to prove the \nfeasibility and correctness of implementing \nNahid's algorithm, it has been done to produce and \nevaluate a neural network model to diagnose \nendometrioma on the ovary. In this context, a U -\nNET neural network architecture with a sigmoid \nactivation function has been used. The input of this \nnetwork is def ined as a 128 x 128 matrix that \nrepresents the images of the ovary. Its purpose is \nto provide an image focused on the ovary and to \ncalculate the exact location of the endometrioma \ndisorder by artificial intelligence. In this context, \ndue to the insufficien t data of the Glenda dataset, \nnew data has been collected and added to the \ndataset. Also, by using the image rotation \ntechnique, the number of data elements has \nincreased to four times. It should be noted that due \nto the existence of standards in surgery, there is no \npossibility of the camera being uneven during \nsurgery. In this research, this method was used \nonly to overcome the problem of data limitations. \nNot using this technique and, on the contrary, \nincreasing the number of data will lead to an \nincrease in the accuracy of the model. Finally, a \nmodel with reasonable accuracy was produced. \nTable 1 reports the output of evaluation criteria on \nthis model. The results of the evaluation of this model \nstate that if the number of data increases and the \nprinciples defined in this article are followed, it is \npossible to fully automate and computerize surgery.   \n \nFigure 6, Flowchart of the proposed method of surgery \n \nFigure 7, An example of Sina Tree for isolated ovarian \nendometriosis.  \n \n\n\n  \n \nTable 1, Evaluation of ovarian endometriosis \nlocation detection model. \nVal IoU \nscore \nVal Loss Train IoU \nscore \nTrain Loss Metric \n0.9759 0.0122 0.9995 2.85e-04 Value \n \nFigure 8 shows some examples of images used as \ndatasets. In these images, the ovary organ is \nconsidered focused. It is possible that by other \nneural network models, in the states before \nreaching the ovary, the ovary was separated from \nother organs, and bas ed on the result, \nendometrioma tissue was diagnosed at the current \nstate. Also, it is possible to prepare images from \ndifferent angles of the ovary based on the human \ngeometric model and use them for artificial \nintelligence processing to improve the accura cy of \ncalculations. In the following, the calculated output \nof artificial intelligence (a) and the expected output \n(b) for training data (Figure 9) and test data \n(Figure 10) have been compared.      \na)   b)   \nFigure 8, Two instance of data used to train the \nendometriosis location detector model. \na)  b)  \nFigure 9, Comparison of predicted and true mask \nof the model.  \na)  b)   \nFigure 10, Comparison of predicted and true mask of \nthe model.  \n6 Conclusion   \nIn this research, for the first time, how to perform \nsurgery by artificial intelligence was described. By \ndescribing the Nahid algorithm, which consists of the \nSina algorithm and the Sina tree data structure, an \nattempt was made to provide a comprehensive,  \nefficient, and accurate method for surgical \nmanagement and operation by artificial intelligence \nbased on a geometric model of humans and observing \nthe principles of situation separation. Then, focusing \non surgery related to isolated ovarian endometriosis \ndisorder, how to use the Nahid algorithm for this type \nof surgery was explained. Also, by producing and \nevaluating a model based on an artificial neural \nnetwork that detects and localizes endometrioma \ntissue in the ovary, it was proved that by following th e \nNahid algorithm, it is possible to produce artificial \nintelligence to navigate the operation and this process \nis reliable. The obtained accuracy, the lack of \nvulnerability of the model against the phenomenon of \noverfitting based on the principle of separ ation of the \nsituation, and attention to the limitation of the \ndatasets in this model prove the reliability of the \nproposed method.     \n7 Declarations  \n7.1 Appreciation  \nWe would like to express our gr atitude to dear  Dr. \nCamran Nezhat, the father of modern surge ry and \nminimally invasive surgery in the world, for presenting \nthe idea of implemen ting artificial intelligence for doing \na surgery and providing the necessary guidance.  \n \n7.2 Funding \nNot applicable.  \n \n\n\n  \n \n7.3 Conflicts of interest/Competing \ninterests: \nNot applicable. \n7.4 Availability of data and material: \nNot applicable. \n7.5 Code Availability \nYou can access our library in Github: \nhttps://github.com/sinasaadati95/Model \n \n \n \n8 References  \n[1] Hong, W. Y., Kao, C. L., Kuo, Y. H., Wang, J. R., \n Chang, W. L., & Shih, C. S. (2020). \n Cholecseg8k: a semantic segmentation \n dataset for laparoscopic cholecystectomy \n based on cholec80. arXiv preprint \n arXiv:2012.12453.   \n \n[2] Madad Zadeh, S., Francois, T., Calvet, L., \n Chauvet, P., Canis, M., Bartoli, A., & Bourdel, \n N. (2020). SurgAI: deep lea rning for \n computerized laparoscopic image \n understanding in gynaecology. Surgical \n endoscopy, 34, 5377-5383.    \n \n[3] Mascagni, P., Alapatt, D., Sestini, L., Altieri, M. \n S., Madani, A., Watanabe, Y., ... & Hashimoto, \n D. A. (2022). Computer vision in surgery: \n from potential to clinical value. npj Digital \n Medicine, 5(1), 163.    \n \n[4] Naqvi, N. Z., Kaur, K., Khanna, S., & Singh, S.  \n (2023). An Overview of Machine Learning \n Techniques Focusing on the Diagnosis of \n Endometriosis. Machine Vision and \n Augmented Intelligence: Select Proceedings \n of MAI 2022, 61-84.     \n \n[5] Nezhat, F. R., Kavic, M., Nezhat, C. H., & \n Nezhat, C. (2023). Forward We Go!. JSLS: \n Journal of the Society of Laparoscopic & \n Robotic Surgeons, 27(1).    \n \n[6] Scheikl, P. M., Laschewski, S., Kisilenko, A., \n Davitashvili, T., Müller, B., Capek, M., ... & \n Mathis-Ullrich, F. (2020, September). Deep \n learning for semantic segmentation of organs \n and tissues in laparoscopic surgery. In \n Current Directions in Biomedical  Engineering \n (Vol. 6, No. 1, p. 20200016). De  Gruyter.    \n \n[7] Visalaxi, S., & Muthu, T. S. (2021). Automated \n prediction of endometriosis using deep \n learning. International Journal of Nonlinear \n Analysis and Applications, 12(2), 2403-2416.   \n \n[8] Goodfellow, I., Bengio, Y., & Courville, A. \n (2016). Deep learning. MIT press.  \n \n[9] Gulli, A., & Pal, S. (2017). Deep learning with \n Keras. Packt Publishing Ltd.  \n \n[10] Finlay, J. (2020). An introduction to artificial \n intelligence. Crc Press.  \n \n[11] Parker, L. (2020).  Vagina Problems: \n Endometriosis, Painful Sex, and Other Taboo \n Topics. St. Martin's Griffin.  \n \n[12] Kaul, V., Enslin, S., & Gross, S. A. (2020). History \n of artificial intelligence in medicine. \n Gastrointestinal  endoscopy, 92(4), 807-812.   \n \n[13] Ranschaert, E. R. , Morozov, S., & Algra, P. R. \n (Eds.). (2019). Artificial intelligence in medical \n imaging: opportunities, applications and risks. \n Springer.","source_license":"CC0","license_restricted":false}