{"paper_id":"b0bbad1d-0f5b-4902-ad0e-d997f50f408f","body_text":"Josphin Mary J. et al: A comprehensive review ……  Archives for Technical Sciences 2025, 33(2), 500-515 \nTechnical Institute Bijeljina, Archives for Technical Sciences. Year XVII – N 0 33           500 \nISSN 1840-4855 \ne-ISSN 2233-0046 \n \nOriginal scientific article  \nhttp://dx.doi.org/10.70102/afts.2025.1833.500 \n \nA COMPREHENSIVE REVIEW OF CLASSIFICATION \nTECHNIQUES FOR ENDOMETRIOSIS DISEASE \nIDENTIFICATION \nJ. Josphin Mary1, V. Shanthi2 \n \n1Research Scholar , Assistant Professor, Department of Computer Science, Faculty of \nHumanities and Science, Meenakshi Academy of Higher Education and Research \n(Deemed to be University), Chennai, Tamil Nadu, India.  \ne-mail: josphinasstprofessor@gmail.com, orcid: https://orcid.org/0009-0009-9224-673x \n 2Professor, Department of Computer Science , Faculty of Humanities and Science , \nMeenakshi Academy of Higher Education and Research (Deemed to be University) , \nChennai, Tamil Nadu, India. e-mail: vairavanshanthi@gmail.com,  \norcid: https://orcid.org/0000-0002-6416-6291 \nReceived: June 19, 2025; Revised: September 04, 2025; Accepted: September 29, 2025; Published: October 30, 2025 \nABSTRACT \nMedical disorders in women can often be the underlying cause of various symptoms and are frequently \nassociated with anovulatory conditions, such as Endometriosis. The limitation of finding the specific  \ndiseases in image processing approach is complex structure tissue, early detection and treatment of these \nconditions are essential. To address these challenges, this review research with Multiple Machine \nLearning (ML) approaches such as Gradient Boosted De cision Tree, SE-ResNet-34 network, and CNN-\nbased deep learning, for classification purpose for diseases identification. The datasets taken an \nultrasound image related to Endometriosis, obtained from open -source platforms provided by women's \nhealthcare facilities. Prior to analysis, these input images undergo data preprocessing techniques to \nenhance their quality and relevance, facilitating accurate evaluation of Endometriosis cases. The CNN \nnetwork architecture is applied to extract intricate features from the input datasets. SE -ResNet-34 \nnetworks are particul arly effective for image classification tasks due to their ability to address the \nadvanced gradient problem, allowing for the construction of deeper network architectures with enhanced \nperformance. Similarly, CNN, a powerful ensemble learning method, impro ves predictive accuracy by \niteratively reducing classification errors. Performance metrics such as accuracy, sensitivity, and F1-score \nare used to evaluate the efficacy of these algorithms in the early diagnosis of Endometriosis and \nimproving healthcare outcomes for women. \nKey words: endometriosis, residual network (resnet), convolutional neural network, se -resnet-34, \nmachine learning (ml),  \nINTRODUCTION \nIn endometriosis, tissue resembling the endometrium (i.e., facing of the uterus) raises external on uterus, \noften causing pelvic pain. Women with endometriosis may also experience infertility, fatigue, multisite \npain, and other comorbidities. Therefore, it  is essential to recognize endometriosis as a condition that \ncan manifest differently and have varying implications at different stages of life. Endometriosis can also \n\nJosphin Mary J. et al: A comprehensive review ……  Archives for Technical Sciences 2025, 33(2), 500-515 \nTechnical Institute Bijeljina, Archives for Technical Sciences. Year XVII – N 0 33           501 \npresent as severe menstrual cramps (dysmenorrhea), ongoing pelvic pain, and pain associated with \nbladder or bowel issues. Endometriosis can affect a woman's quality of life because of pain, fatigue, and \npossible complications related to fertility [1].  \nEndometriosis is characterized by a histological review of an endometrial -like specimen (consisting of \nglands and endometrial stroma) growing out of the uterine cavity, usually implanted in the peritoneal \ncavity. Given the clinical circumstances and patient preferences for intervention, there are opportunities \nfor both medical and surgical management of deep endometriosis. Medical treatments, such as \nprogestins, gonadotropin-releasing hormone agonists, and gonadotropin-releasing hormone antagonists, \ncan help minimize the development of lesions. Surgical management may adversely affect the bladder, \nureter, and intestines, leading to difficult -to-manage short -term consequences such as bleeding and \nureteral damage, as well as long-term sequelae that include rectal or ureteral stricture [41].  \nVariations in the pathway may include cellular adhesion and proliferation, somatic mutation, \ninflammation, localized steroidogenesis, neurogenesis, and immunological dysregulation. Changes in \nthis pathway, collectively, likely contribute to the development of endometriosis. Risk factors associated \nwith endometriosis include low birth weight, Mullerian abnormalities, early menarche, short menstrual \ncycles, heavy menstrual bleeding, low body mass index, and nulliparity. The identification of an \nendometrioma b y ultrasonography should prompt further evaluation, particularly in a patient who \ncomplains of significant pain, as endometriomas necessarily co-exist with deep endometriosis [3]. \n Chronic pelvic pain (CPP) is a common condition experienced by women that can have detrimental \neffects on their wellness and quality of life. In order to assess pelvic pain, and specifically diagnose \nendometriosis with concomitant surgical excision, laparoscopy and histology have been deemed suitable \noptions. However, 30-50% of women with pain also have endometriosis, making it challenging to link \nendometriosis to pelvic pain definitively. Women complained of pain and were then referred to a public \ngynecology clinic (as per standards of care) and then randomly assigned to one of two gynecology units, \nwhere they received standard care as patients. Women were followed for 36 months, with 6 -monthly \nsurvey assessments of their demographics, medical history, qua lity of life, and Likert scale pain \nperceptions. Staging for endometriosis was carried out, and operational notes were reviewed [42]. \nKey factors in PCOS include age, Body Mass Index (BMI), hormone levels, irregular menstruation, and \nlifestyle patterns. These variables can inform predictive models that assist healthcare providers in early \nPCOS risk detection. Hyperparameter tuning, which considers factors such as age, weight, blood group, \nand Respiratory Rate (RR), is critical to enhancing model accuracy and preventing overfitting. The \ndecision tree algorithm is utilized for model training and initialization, providing a robust and clinically \nvaluable tool for PCOS risk assessment [5]. \nIn young girls, acne is a common and typically normal condition. In such cases, acne often proves \nresistant to treatment. Another concerning symptom of PCOD is hair thinning, particularly on the scalp, \nincluding areas like the temples, forehead, or crown. Hair may also appear thinner across the body. \nDiagnosing PCOD requires a comprehensive blood test panel commonly referred to as the PCOD panel \nto measure levels of prolactin, and additionally, pelvic examinations are performed to detect masses, \nabnormal growths, or other irregularities [43]. \nFeature selection involves using statistical techniques or algorithms, such as Recursive Feature \nElimination, to identify the most significant predictors, including BMI, hormone levels, and menstrual \ncycle patterns. To optimize model performance and preven t feature overload, three distinct feature \nselection techniques were applied to extract multiple reduced feature sets. After that, the dataset was \nseparated into subsets for testing, validation, and training to guarantee a thorough assessment. Neural \nNetworks, particularly suited for capturing complex feature relationships in large datasets, were utilized \nto achieve the most accurate and descriptive representation of the data [7]. \nContribution of the Work  \nThis study focuses on the study analysis of  Endometriosis, providing a comprehensive analysis of \nEndometriosis disease classification methods and emphasizing the limitations of current approaches. It \n\nJosphin Mary J. et al: A comprehensive review ……  Archives for Technical Sciences 2025, 33(2), 500-515 \nTechnical Institute Bijeljina, Archives for Technical Sciences. Year XVII – N 0 33           502 \nconsolidates diverse datasets for disease detection and evaluates three key machine learning algorithms \nGradient Boosted Decision Tree, Residual Network (ResNet), and CNN based deep learning technique \nwith feature selection predictive capabilities and limitations [2] [4]. By synthesizing prior research, the \nstudy underscores underexplored challenges and variability in the field, offering detailed descriptions of \nrelevant datasets and classification strategies. Additionally, it examines performance metrics, such as \naccuracy, commonly used in Endometriosis research [12]. \nThe analysis explores various Endometriosis detection methods, focusing on classification techniques \nemployed across studies. The three key algorithms Gradient boosted decision tree, Residual Network \n(ResNet), and CNN based deep learning technique, are critically examined for their effectiveness in \nmachine learning applications. A detailed explanation of the different dataset types used in these studies \nis also provided, emphasizing their role in influencing algorithmic performance [8]. The study identifies \nthe performance of datasets when applied to different machine learning algorithms, enabling a \ncomparison of their efficacy. \nA significant emphasis is placed on comparing the performance of current methods for Endometriosis \ndetection. Moreover, the review presents research ideas for future work, identifying critical challenges \nassociated with using machine learning for Endometriosis detection. These include issues related to data \nvariability, feature selection, and model optimization, which must be addressed to improve the reliability \nand applicability of machine learning approaches in this domain. \nLITERATURE REVIEW  \nThis section explores the methodologies and algorithms discussed in various studies, providing a concise \nsummary and comparative review for a brief analysis. The identified research gaps highlight key \nalgorithms, their parameters, workflows, and outcomes, followed by a performance evaluation \nSumana et al. [ 44] proposed the Gynaecological Disease Diagnosis Expert System (GDDES), which \nleverages Natural Language Processing (NLP) to compute cosine similarities and retrieve the most \nrelevant voice recordings of disease diagnoses [10]. The system initiates the process by prompting users \nto report their symptoms in their native language. Subsequently, a Support Vector Classifier (SVC) \nmodel predicts the disease, storing diagnostic results in a knowledge base that also serves as the dataset \nrepository. The SVC model demonstrated robust performance, achieving an accuracy and precision of \n93% and F1 scores of 92%. \nJunfang Fan et al. [9] introduced a lightweight classification and diagnostic network incorporating a \nreverse bottleneck design to enhance feature extraction. Shuffle Net, a lightweight mobile terminal, \nemploys pointwise and depth wise convolutions. Initially, it uses a conventional convolution with a stride \nof 4 and a kernel size of 4 × 4 for feature extraction, followed by a down sampling procedure based on \noptimal pooling, The classification accuracy of the network is reported at 95.93%. \nRen et al. [ 45] reviewed the in clinical practice, determining the lymph node metastatic status of \nEndometrial Cancer (EC) is a significant difficulty.  Machine learning has been used by some researchers \nto detect lymph node metastases in EC patients early.  However, be cause of the variety of models and \nmodelling variables, the predictive usefulness of machine learning is debatable.  However, when there \nis a significant disparity in the number of lymph node metastatic and non -metastasis samples, the c -\nindex is unable to accurately represent the model's predictive accuracy for lymph node metastasis. The \nkind of machine learning models built using clinical features, radiomic features, and radiomic \ncharacteristics mixed with clinical features were used to conduct subgroup analyses. \nFazakis et al. [11] detailed a diabetes risk prediction framework using a Knowledge Discovery in \nDatabase (KDD) process. The dataset construction, feature selection, and classification tasks are \naddressed using several supervised machine learning technique s. Decision Trees create classification \nmodels by segmenting datasets into smaller subsets, while Random Forests build multiple decision trees \nto perform regression and prediction simultaneously. The proposed ensemble Weighted V oting LRRFs \nML model demonstrates enhanced diabetes prediction performance with an Area Under the ROC Curve \n(AUC) of 0.884. \n\nJosphin Mary J. et al: A comprehensive review ……  Archives for Technical Sciences 2025, 33(2), 500-515 \nTechnical Institute Bijeljina, Archives for Technical Sciences. Year XVII – N 0 33           503 \nYan Xiao et al. [ 46] explored the viability of treating female infertility with Low -Intensity Focused \nUltrasound (LIFU), especially when PCOS and Premature Ovarian Insufficiency (POI) are present. It is \ncommonly acknowledged that POI and PCOS are the main causes of infertili ty in women. Because of \nits mechanical effects, Low-intensity ultrasound focus with pulses (LIFU) has the potential to minimize \novarian tissue damage while promoting follicle formation. Potential treatments have been investigated \nthrough experimental studies carried out in controlled facilities with conditions of 22 ± 2 °C, 45 –55 % \nhumidity, and a 12-hour light/dark cycle. \nFouzia Akhter et al. [13] discussed a detection of clinical and biochemical evidence of \nhyperandrogenism (after ruling out other possible diseases) in conjunction with chronic menstrual \nabnormalities allowed for the diagnosis of PCOS in teenage females age of 10 and 19 were included in \nthe inclusion criteria, however certain medical problems and active therapies were excluded. According \nto BMI study, overweight (29.70%) and obesity (39.40%) were quite prevalent. 76.60% of people had \nnormal levels of abdomin al fat, whereas 20.00% had pre -hypertensive conditions and 3.40% had \nhypertension. There was variation in the glycaemic state, with 21.10% prediabetic, 2.90% diabetic, and \n76.00% normoglycemic. \nKrishna et al. [14] developed a diagnose PCOS, the tunica albuginea oculi region is separated from full \neye images using a visual segmentation technique. Using pre-trained deep learning, the tunica albuginea \noculi images were segmented and then classified as either PCOS or healthy. To guarantee that each \ncategory had an equal number of women, women who were selected at random from the university \ncampus were questioned about whether they showed any symptoms of PCOS. The \"no\" class has a 92 \n% precise predictability, while the \"yes\" class has an 85 % precision, with recall rates of 85 and 92 %, \nrespectively. \nPushkarini et al. [15] utilized a PCOS dataset is used for model testing and training. Divide the \npreviously processed dataset into train and test sets; for instance, designate 20% of the dataset as a test \nand 80% as a training set. The models are trained and adjusted using training sets, With the highest 𝑅2 \n= 0.985, R2 = 0.985, the lowest Mean Absolute Error (MAE = 1.556), and the lowest Root Mean Square \nError (RMSE = 3.079), the Random Forest model performs the best and makes reliable predictions. With \n𝑅2 = 0.978, R2 = 0.978, greater MAE (3.282), and RMSE (3.930), Linear Regression performs \nmarginally worse. \nZhang et al. [16] In this retrospective study, 122 patients with pre-operative MRI were included (78 AEH \nand 44 CEC). Radiomics features were extracted from apparent diffusion coefficient (ADC), diffusion-\nweighted imaging (DWI), and T2-weighted imaging (T2WI) maps. The best area under the curve (AUC) \nwas 0.932 (95% confidential interval [CI]: 0.880-0.984), with a bootstrap corrected AUC of 0.922 in the \ntraining set, and an AUC of 0.942 (95% CI: 0.852-1.000) in the validation set for the radiomics-clinical \nmodel. The radiomics -clinical model included multimodal radiomics features and clinical variables - \nendometrial thickness >11mm, and nulliparous status. Our output data (F1 score = 0.900 for inconsistent \ngroup; F1 score = 0.865 for consistent group). \nAgrawal et al. [17] conducted a single centre cross -sectional study enrolled 80 women diagnosed with \nPCOD to examine the impact of body image perception on depression severity and quality of life. The \nWHOQOL-BREF scale was employed to assess quality of lif e, revealing that women with PCOD and \ndepression had significantly lower physical (p < 0.001), psychological (p < 0.001), and overall quality \nof life (p = 0.025) scores. Additionally, 73.8% of the participants were found to have depression, a \nnotably high prevalence. The study also highlighted the compounded effects of other mental health \nconditions and substance use disorders (excluding caffeine and nicotine addiction) on these outcomes. \nKumar et al. [18] Depending on the population and diagnostic criteria employed, the prevalence rate of \nPCOS, the most prevalent endocrine disorder affecting women of reproductive age, can range from 8 to \n13%. two train -test ratios (70:30 and 80:20), a thor ough examination of nine machine learning \ntechniques for PCOS classification was carried out. Particle Swarm Optimization (PSO) was used in \nconjunction with these models to improve performance, and the results showed 94.44% sensitivity, \n97.22% specificity, and 94.44% precision. The accuracy of the models' positive case detection was \ngreatly enhanced by this integration. \n\nJosphin Mary J. et al: A comprehensive review ……  Archives for Technical Sciences 2025, 33(2), 500-515 \nTechnical Institute Bijeljina, Archives for Technical Sciences. Year XVII – N 0 33           504 \nElsayed et al. [19] investigated the impact of female infertility, imposing both financial and \npsychological burdens on patients. A clinical study comparing two groups revealed significant \ndifferences in pregnancy outcomes. Group 1, with 35 out of 50 cases confirmed via ultrasound to have \ngestational sacs (score > 6), showed a significantly higher clinical pregnancy rate compared to Group 2, \nwith 18 out of 50 cases (p = 0.0007). Furthermore, the miscarriage rate was markedly lower in Group 1 \n(1 out of 35) c ompared to Group 2 (5 out of 18; p = 0.006), emphasizing the importance of effective \ntreatment approaches. \nParamasivam et al. [20] proposed the particle swarm optimization (PSO) The Hybrid Attention -\nEnhanced MobileNetV2 using Particle Swarm Optimization (PSO) is a deep learning model that \naccurately classifies endometrial cancer using CT image data. MobileNetV2  acts as a lightweight \nbackbone, and is substituted with a hybrid attention mechanism that concentrates on critical tumour \nregions, while reducing extraneous background noise from medical imaging features. The specificity \n91.45 %, sensitivity 86.02 %, prec ision 86.75 %parameters and feature selection are enhanced using \nPSO to optimize speed and accuracy. Diabetic diagnoses and treatment recommendations can now take \nplace significantly quicker using our hybrid model, as our documentation demonstrates the \ndocumentation described considerably less computational expense; while still retaining diagnostics. The \nmodel design is ideally suited for early detection of cancer in clinical settings where computational \nresources may be scarce and limited. \nPREVIOUS ENDOMETRIOSIS DETECTION TECHNIQUE \nIn table 1 below presents a review and analysis of various classification methods, outcomes, from \nEndometriosis approaches. This comparison highlights the differences between conventional approaches \nand Machine Learning (ML)-based techniques, focusing on key performance metrics such as F1-score, \nrecall, precision, and accuracy. \nMany research studies have utilized ultrasound images as input datasets, often incorporating data \naugmentation techniques to analyse large datasets, identify patterns, and improve analytical accuracy. \nAdvanced algorithms, including DenseNet-121, Logistic Regression, Random Forests, and ResNet-50, \nhave been employed to detect conditions such as PCOD and PCOS. \nTable 1. Various review and parameter analysis of disorder endometriosis \nReferences Types of \ndisorder \nDataset Algorithm \nUsed \nOutput Limitation \nVisalaxia et \nal  \ntraumatic \ndisorder \nEndometriosis \nDataset \nCNN 1. Accuracy 90%  \n2. Precision 83%. \n3. Recall 82% \nLimited to specific types \nof endometritis data; \ndoes not generalize well \nto other forms of \ngynaecological \ndisorders. \nKotaro \nKitaya et al \n2021 \nChronic \nendometritis \n(CE) \nEndometriosis \nDataset \n1. VGG-19 \n2. Dense \nNet-121 \n \n1. sensitivity 93.6 % \n2. specificity 92.3 % \n3. accuracy     92.8 % \n4. precision    88.0 % \n5. F1-score     90.7 % \nModerate accuracy; \nlimited model \noptimization and dataset \ndiversity. \nPing Hu et \nal 2021 \nOvarian \nEndometriosis \ndisorder  \nEndometriosis \nDataset \n1. ResNet-\n152. \n2. DenseNe\nt-161 \nAUROC  0.986 Dependence on \nstructured EHRs; \naccuracy varies with \nreporting completeness. \nZhao et al \n2022 \nEndometriosis \ndisorder  \n \nHysteroscopic \nimages \nYOLOX ACC 95.83% Relies heavily on image \npreprocessing; not \nrobust to variations in \nimaging conditions. \nJames et al \n2023. \nmicrosatellite \nstatus \nH&E-stained \nwhole slide \nimages \n(WSIs) \nMSI \nclassificatio\nn \nSensitivity 0.857  \nF1-Score 0.826  \nAUROC 0.799 \nLimited to detecting \nhyperandrogenism; does \nnot encompass other \nPCOS markers. \n\nJosphin Mary J. et al: A comprehensive review ……  Archives for Technical Sciences 2025, 33(2), 500-515 \nTechnical Institute Bijeljina, Archives for Technical Sciences. Year XVII – N 0 33           505 \nHassan et \nal \n2020 \nPCOS In Kaggle site \npatient PCOS \nhealth report \ndata set \n1. Logistic \nRegression, \n2. CART, \n3. Naïve       \nBayes \n1. Acc-92%, \n2. Acc-90% \n3. Acc-81%. \nLimited feature \nengineering; \nperformance varies \nsignificantly across \nalgorithms. \nAkanbi et \nal \n2024 \nPCOS Patient PCOS \nhealth report \ndata set on the \nKaggle \nwebsite \n1. AdaBoo\nst \n2. Logistic \nRegressi\non \n3. Gradient \nBoost \n1. Acc-91%, \n2. Acc-90% \n3. Acc-95%. \nHigh accuracy but \nlimited generalizability \nto non-Kaggle datasets. \nMukta \nAgarwal et \nal \n2024 \ngynecological \nissues \nWomen \nHealth care \nData \n1. Diseases \ndiagnosi\ns \nalgorith\nm \n1. Abdominal \ndiscomfort 15.6% \n2. Vaginal discharge \n7.2%. \nLimited to symptomatic \nclassification; lacks \nvalidation on diverse \ndata sources. \nAl-Ghazali \net al \n2022 \nPCOS Specific \nwomen \nHealth care \nData \nclassificatio\nn technique \n1. Sensitivity 90.9%, \n2. Specificity 90.9%, \n3. Accuracy 90% \nClassification depends \non structured and high -\nquality data; results may \ndegrade with noisy \ninputs. \nNandipati \net al \n2020 \nPCOS significance \nhealth care \nreport data set \nKNN In RapidMiner \n1. Accuracy 90.38 % \n2. Precision 90.83 % \n3. Recall 90.83 % \nKNN is computationally \nintensive for large \ndatasets; prone to \noverfitting with noisy \ndata. \n \nMETHODS  \nThis review explores feature selection techniques, three key classification methods, and performance \nmetrics in the context of machine learning applications for detecting Endometriosis disorders. The input \ndataset primarily consists of ultrasound images, including both affected and unaffected cases, \ncomplemented by healthcare data specific to women's medical analyses. The study evaluates key \nperformance metrics and assesses the effective ness of various approaches. By iteratively training \nalgorithms on these images and refining their parameters based on prediction errors, the models learn to \nidentify patterns distinguishing healthy ovaries from those affected by different types of Endometrioses. \nThe integration of machine learning techniques holds great promise for significantly improving the early \ndetection and treatment of Endometriosis. \nData Set \nThe ovarian ultrasound dataset analyzed in this study was obtained from the open -source website and \nconsists of medical data, which we analyzed in collaboration with four radiologists in the Department \nof Radiology. The radiologists assessed 1,250 patients, which consisted of 1,000 patients with the normal \novaries or other pathologies and 250 patients with endometriosis, a chronic gynecological disease \ncharacterized by the growth of endometrial -like tissue outside the uterus. While some patients had \nmultiple ultrasound studies, only the most recent study for each patient was included in the review. In \nthe interest of consistency and clarity, we selected only images showing the ovary with surrounding \nstructures. The radiologists categorized the images into tw o groups for classification: images showing \nsonographic features consistent with endometriosis (such as ovarian endometriomas or deep infiltrating \nlesions) and ones with normal ovarian morphology (Figure 1). \n\nJosphin Mary J. et al: A comprehensive review ……  Archives for Technical Sciences 2025, 33(2), 500-515 \nTechnical Institute Bijeljina, Archives for Technical Sciences. Year XVII – N 0 33           506 \n \n \nNon-Affected \n(Normal) \n \n \n \nAffected \n(Endometriosis) \n \n \n \nHypoechoic \nwith \ninternal echoes \n \n \n \nFigure 1. Ultrasound image dataset of endometriosis \nEndometriosis \nEndometriosis has a complex and multifaceted etiology influenced by environmental, genetic, and \nintergenerational factors. These factors contribute to ovarian and adrenal hyperandrogenism, which, in \nturn, disrupts the signaling of the hypothalamic pituitar y ovarian axis. The syndrome is further \ncharacterized by metabolic dysfunctions, including lipid toxicity, oxidative stress, and insulin resistance, \nall of which are exacerbated by adipose tissue accumulation associated with hyperandrogenism. \nConsequently, Endometriosis manifests as a broad clinical spectrum, affecting metabolic, reproductive, \nand psychological health. While genetic predisposition plays a pivotal role in Endometriosis, \nenvironmental factors likely interact with these genetic elements to wor sen the condition. However, \nmore recent examinations have revealed a polygenic basis for the syndrome. The genetic complexity of \nthe syndrome is underscored by the identification of 19 risk loci associated with neuroendocrine, \nmetabolic, and reproductive pathways through genome-wide association studies. \nAlthough Endometriosis lacks a well-established physiopathology, it is an inflammatory condition, and \nendocrine-immunological interactions likely influence its etiology.  Endometrial cells that are lost during \nmenstruation and exit the uterus are typically  cleared by the immune system.   These cells may not be \nremoved efficiently in Endometriosis, possibly due to decreased NK cell activity or other immunological \ndysfunctions.  Instead of removing endometrial cells, macrophages may release cytokines and growt h \nfactors [31]. \nEndometriosis frequently manifests as infertility, pelvic discomfort, and dysmenorrhea, which can lead \nto a worse quality of life and a high rate of morbidity in chronic situations.   Although the precise \npathophysiology and natural history of Endometriosis are not entirely known, the most widely accepted \nexplanation states that endometrial cells are implanted and develop in the pelvic cavity during retrograde \nmenstruation due to the intricate interactions of growth, angiogenic, and immunological \nfactors.  Although little is known about the mediating processes behind the inverse association between \nobesity and endometriosis risk, several theories and pathways have been proposed in published research \ndomains. \nPCOD \nPCOD is Stein-Leventhal Syndrome, is characterized by clusters of small, pearl-like cysts in the ovaries. \nThese fluid-filled cysts contain immature eggs and often outcome from a combination of genetic and \nenvironmental factors. PCOD leads to a variety of s ymptoms, including physical changes, irregular \nmenstruation, and, if left untreated, serious health complications such as diabetes, heart disease, obesity, \nmood disorders, endometrial cancer, and sleep apnea. The condition primarily affects women between \n\n\nJosphin Mary J. et al: A comprehensive review ……  Archives for Technical Sciences 2025, 33(2), 500-515 \nTechnical Institute Bijeljina, Archives for Technical Sciences. Year XVII – N 0 33           507 \nthe ages of 14 and 44. A hallmark feature of PCOD is the overproduction of androgens, which are \nessential for follicular development. While luteinizing hormone levels are noticeably elevated, the \nabsence of hormonal balance impairs optimal progesterone and estrogen production.  \nCompensatory hyperinsulinemia, a common characteristic of the disorder, exacerbates ovarian \ndysfunction. In women health care data analysis hyperinsulinemia increases ovarian androgen \nproduction, inhibits ovulation, and contributes to hyperandrogenism. Thi s occurs as insulin stimulates \ntheca cells ovarian cells responsible for testosterone production through androgen biosynthesis. As an \noutcome of this cascade, the excess androgens generated by insulin resistance and abnormal ovarian \nfunction lead to hyperandrogenism. Hyperandrogenism disrupts normal ovulation and inhibits follicular \ngrowth, impairing the development of eggs within the ovarian sacs [32]. \nFeature Selection \nIn ovarian and ovary disease prediction, feature selection also helps to address challenges like \noverfitting, which occurs when the model learns noise instead of underlying patterns in the data. By \nremoving irrelevant features, feature selection reduces th e risk of overfitting and enhances the \ngeneralizability of the model to new patient data. For example, in detecting ovarian cancer, feature \nselection might isolate a small set of biomarkers strongly associated with malignancy, enabling early \ndiagnosis and personalized treatment planning. Additionally, this process supports the identification of \nnovel disease mechanisms, paving the way for improved diagnostic tools and therapeutic targets. \nGrey Wolf Optimizer (GWO) Algorithm \nThe Grey Wolf pack or group, which typically consists of five to twelve wolves, is the model for this \noptimizer algorithm. Every wolf can be classified as an alpha, beta, delta, or omega wolf, and it has a \nspecial connection to teamwork. \nDeviation∶  σ ∑ (1 − μ)2 h(i)n\ni=1           …       (1) \nIn equation (1) Let f(i), where Ax is the vision's surface and (i = 1, 2, n) is the number of points in the \nimage with strength i. \n𝑤𝑐𝑜𝑏𝑥 =  \n∑ 𝑖 𝐵(𝑖,𝑗)𝑒\n∑ 𝑖 𝐵(𝑖,𝑗)𝑒\n         …        (2) \nIn equation (2) Between 0 and 180°, recovered H1 and two weighted centres of spectrum features for \neach 1° Radon transform. Consequently, 724 (181 × 3) characteristics would be extracted in total. In \nsummary, the steps to determine the HOS traits from an ultrasound image are as follows: \n \nFigure 2. Flow chart of GWO Feature selection \nNo  \n \n \n \n \n \n \nStart \nCollect GWO data (e.g., genetic)  \nAnalyse possible features  \nValidation of Each Pixel \npredicts disease \noutcomes \n          Training \n Testing  \nEnd \nDeployment and Iteration \nNo  \n\nJosphin Mary J. et al: A comprehensive review ……  Archives for Technical Sciences 2025, 33(2), 500-515 \nTechnical Institute Bijeljina, Archives for Technical Sciences. Year XVII – N 0 33           508 \nIn training and testing for ovarian and ovary disease prediction, feature selection works by identifying \nand retaining only the most relevant variables from the image dataset to build predictive models shown \nin Figure 2. During the training phase, feature selection methods analyze the dataset to determine which \nfeatures (e.g., biomarkers, genetic mutations, or clinical parameters) have the strongest correlation with \nthe target outcome, such as the presence of a di sease. GWO reduces the dimensionality of the  data, \nallowing the model to learn patterns more efficiently and minimizing the risk of overfitting. During \ntesting, the selected features are used to evaluate the model's performance on unseen data, ensuring that \nthe model generalizes well and accurately predicts disease outcomes. This streamlined approach \nimproves model interpretability, computational efficiency, and diagnostic [33]. \nClassification Technique  \nClassification plays an essential role in classifying patient disorder related to Endometriosis In the \nreview approaches such as Gradient Boosted Decision Tree, SE -ResNet-34 network, and CNN based \ndeep learning technique have demonstrated exceptional struc tural efficiency and predictive accuracy. \nThese models excel in predicting group identifiers or class labels for previously hidden data, enabling \nprecise categorization and deeper insights into complex diagnostic challenges. \nGradient Boosted Decision Tree \nIn order to create a strong predictive model, gradient boosted decision tree classifiers for Endometriosis \ndisorder employ an ensemble learning technique that sequentially combines multiple weak models, \ntypically decision trees. An initial model that generates predictions, typically a basic one, is used to start \nthe process. Using the residual errors differences between the actual and predicted values from the earlier \nmodels, Gradient Boosted Decision Tree iteratively improves this model by training more decision trees. \nBy focused on areas where the earlier models underperformed, each new tree progressively raises the \noverall accuracy. Combining these trees creates a strong classifier that can handle intricate, non -linear \nrelationships in the data, which makes it ideal for identifying ovaries related patterns [34].  \n     Data                               Weight Data                     Weight Data                    Weight Data                     \n                                     \n \nFitting                                  Fitting                                 Fitting                            Fitting \n \n                                          \nDecision Tree 1                  Decision Tree 2                 Decision Tree 3                   Decision Tree 4 \n \nPrediction \n                                                                   \n                                               Ensemble Prediction (Strong Classifier) \nFigure 3. Architecture of gradient boosted decision tree classifiers [36] \n\n\nJosphin Mary J. et al: A comprehensive review ……  Archives for Technical Sciences 2025, 33(2), 500-515 \nTechnical Institute Bijeljina, Archives for Technical Sciences. Year XVII – N 0 33           509 \nFigure 3 show the prediction models in order to improve medical decision -making, an event rate of R \npercent among patients with a predicted risk of R percent is commonly used to define calibration. \nPlotting calibration curves and calculating brier scores were used to confirm the reliability of the models. \nBy averaging the squared difference between expected and observed risk, the Brier score an estimated \ncalibration index that builds upon a flexible calibration analysis is converted into a number between 0 \nand 1 \n𝑓1(𝑥) ≈ 𝑦      …       (1) \n𝑓1 (𝑋)  ≈ 𝑦 − 𝑓1 (𝑋) , 𝑓2 (𝑋),  ---    𝑓3 (𝑛)    …      (2) \nIn equation (1) classifier evaluates a 𝑓1  number of characteristics, including body measurements, \nlifestyle factors, hormonal levels, and ultrasound results, in order to determine a person's probability of \nhaving Endometriosis. Gradient Boosted Decision Tree maintains its accuracy even when there are more \nhealthy samples than diseased ones because of its capacity to manage imbalanced datasets, which are \ntypical in medical diagnosis scenarios. 𝑓𝑛(𝑛) such as Tree's feature importance analysis assist in \ndetermining which clinical factors have the greatest influence on predictions, which helps to better \nunderstand and enhance diagnostic procedures and personalized medicine, this approach works \nespecially well [35]. \nSE-ResNet-34 network \nThe vanishing gradient issue, which can arise in very deep networks, is addressed with SE -ResNet-34 \nnetwork classifiers, a kind of neural network. ResNet introduces the concept of residual learning through \nskip connections or shortcuts, which bypass one or more layers in the network. These connections allow \nthe network to learn identity mappings, ensuring that the deeper layers can focus on learning the residual \n(or incremental changes) instead of the full transformation. This architecture facilitates the t raining of \nmuch deeper networks by enabling gradients to flow directly through the skip connections, improving \nconvergence and avoiding overfitting. ResNet classifiers are especially effective for extracting high -\nlevel features in complex data, such as patterns in medical imaging or multidimensional datasets. \n \nFigure 4. Architecture of SE-ResNet-34 network [37] \nFigure 4 shows working architecture of a SE -ResNet-34 network detailed features from ultrasound \nimages, hormonal patterns, or other diagnostic data to detect subtle markers of the condition. The skip \nconnections in ResNet allow the model to capture both lo w-level features (e.g., pixel intensities in an \nultrasound) and high -level features (e.g., ovarian cyst patterns or hormonal trends). This hierarchical \nfeature extraction is particularly advantageous in medical diagnostics, where small yet critical differences \nin the data can indicate the presence of  endometriosis. Additionally, ResNet’s robustness to overfitting \nmakes it suitable for medical datasets, which are often limited in size. By leveraging its deep architecture \nSE-ResNet Module \n \n \nSE-ResNet block \n  Conv \n   Conv \n   GAP \n     FC \n  ReLU \n     FC \nSigmoi\nd \nOriginal Images \n \n\nJosphin Mary J. et al: A comprehensive review ……  Archives for Technical Sciences 2025, 33(2), 500-515 \nTechnical Institute Bijeljina, Archives for Technical Sciences. Year XVII – N 0 33           510 \nand residual learning, ResNet provides accurate and reliable classification, assisting in early detection and \npersonalized treatment planning [38]. \nCNN Based Deep Learning Technique \nCNNs are distinguished by their convolutional layers, which are made to automatically and adaptively \nconstruct spatial hierarchies of information from input images. The foundation of a CNN is its \nconvolutional layer. which filters the input image in a vari ety of ways. The subsequent layer of the \nnetwork receives the resultant collection of feature -rich maps. Identifying and learning features from \nimages is the responsibility of CNNs' convolutional layers, which use filters that extract local patterns \nand hierarchies. In addition, pooling layers provide a better representation of the features, reduce the \nspatial dimensions of the feature maps, increase computational efficiency, and prevent overfitting. When \ncombined, these layers allow in Figure 5 CNNs architecture to effectively and efficiently carry out tasks \nlike object detection, image classification, and medical image analysis [6] [39]. \n \nFigure 5. CNN Architecture [40] \nCNNs can now recognize intricate patterns linked to endometriosis, like ovarian morphology or follicle \ndistribution, without the need for manually created features. characterized by a combination of \nsymptoms related to hormonal imbalance, Endometrial Disorders , and reproductive health. Image \nclassification, particularly through medical imaging techniques enhanced by machine learning approach \nfor understanding endometriosis as well as identifying associated diseases and disorders.  \nPERFOAMCNE EVALUATION  \nThe performance comparison between the suggested endometriosis classification approach and previous \nresearch approach is calculated below. By contribution better classification techniques that enable the \ndata to be arranged as women's medical health care data with a specified age, this review work has \nhelped to gain recognition. Th e suggested networks are able to more effectively classify the available \ndata. \nTable 2. Dataset comparison of feature selection response for endometriosis \nDifferent Parameters Different classification Value \nGradient Boosted Decision Tree Affected Ultrasound Image Matthews Corr. Coeff.: 0.87 \nNon-Affected Ultrasound Image Log Loss: 0.12 \nSE-ResNet-34 network Affected Ultrasound Image Matthews Corr. Coeff.: 0.89 \nNon-Affected Ultrasound Image Log Loss: 0.09 \nCNN-based Deep Learning Technique Affected Ultrasound Image Matthews Corr. Coeff.: 0.92 \nNon-Affected Ultrasound Image Log Loss: 0.07 \n \nIn Table 2 performance of different classification models on ultrasound image data can be compared \nthrough various parameters. For the Gradient Boosted Decision Tree, the Matthews Correlation \nCoefficient (MCC) for affected ultrasound images is 0.87, and for non-affected images, the Log Loss is \n0.12. The SE-ResNet-34 network achieves a slightly higher MCC of 0.89 for affected images and a Log \nLoss of 0.09 for non -affected images. The CNN -based deep learning technique shows the highest \nHormonal \nImbalance \nEndometrial \nDisorders \n \nInput Image \n Convolution Layer \n Flattened Layer \nPooling Layer \n\nJosphin Mary J. et al: A comprehensive review ……  Archives for Technical Sciences 2025, 33(2), 500-515 \nTechnical Institute Bijeljina, Archives for Technical Sciences. Year XVII – N 0 33           511 \nperformance with an MCC of 0.92 for affected ultrasound images and a Log Loss of 0.07 for non -\naffected images, indicating better accuracy and lower error compared to the other models. \n \nFigure 6. Comparing different performance metrics of accuracy  \nFigure 6 show the classification accuracy of various machine learning and deep learning models. \nPrevious machine learning models such as Naïve Bayes achieve a lower accuracy of 80.46%, while \nCNN show a 90.06%. Dense Net -121 further enhance accuracy to 92.1 3%. The proposed Gradient \nBoosted Decision Tree achieves 95.16%, SE -ResNet-34 network reaching 96.83% and a CNN -based \ndeep learning technique achieving the highest accuracy of 98.23%. This progression highlights the \nsuperior performance of deep learning techniques, particularly in complex classification tasks. \n \n Figure 7. Comparing different performance metrics of sensitivity and specificity  \nFigure 7 Shows the Sensitivity and Specificity ratios for various models, showcasing their ability to \nidentify positive and negative cases. The K-Nearest Neighbors (KNN) achieves 90.13% sensitivity and \n91.13% specificity, reflecting solid but relatively lo wer performance. The Support Vector Classifier \n(SVC) improves these values to 93.13% and 93.21%. Particle Swarm Optimization (PSO) further \nenhances sensitivity to 94.29% and specificity to 94.41%. The proposed Gradient Boosted Decision Tree \nachieves 95.16% sensitivity and 95.08% specificity, showing a balanced improvement. The SE-ResNet-\n34 network, a deep learning approach, sensitivity to 96.83% and specificity to 96.39%. Finally, the \nreview that CNN-based deep learning technique outperforms all models with  the highest sensitivity of \n98.23% and specificity of 98.73%, underscoring its improved performance in accurately identifying both \npositive and negative cases. \n60\n65\n70\n75\n80\n85\n90\n95\n100\nNaïve Bayes\n[26]\nCNN [21] Dense Net-121\n[22]\nSupport Vector\nClassifier [8]\nGradient\nBoosted\nDecision Tree\nSE-ResNet-34\nnetwork\nCNN based\ndeep learning\ntechnique\nClassification Ratio (%)\nDifferent classification Methods\nAccurnacy  Performance Analysis\n60\n65\n70\n75\n80\n85\n90\n95\n100\n105\nKNN [30] Support Vector\nClassifier (SVC) [8]\nParticle Swarm\nOptimization (PSO)\n[18]\nGradient Boosted\nDecision Tree\nSE-ResNet-34\nnetwork\nCNN based deep\nlearning technique\nCLASSIFICATION RATIO (%)\nDIFFERENT METHODS\nSensitivity and Specificity Performance Analysis\nClassification Sensitivity (%) Classification Specificity (%)\n\nJosphin Mary J. et al: A comprehensive review ……  Archives for Technical Sciences 2025, 33(2), 500-515 \nTechnical Institute Bijeljina, Archives for Technical Sciences. Year XVII – N 0 33           512 \n \nFigure 8. Comparing different performance metrics of F1 score \nFigure 8 shows the Classification F1 Score (%) for various models, reflecting their balance between \nprecision and recall. CatBoost achieves a moderate F1 score of 85.86%, while SVM further enhances \nthe F1 score to 92.06%, indicating better performance. The  Gradient Boosted Decision Tree continues \nthis evaluate with an F1 score of 94.16%, showing it improve capability for classification tasks. Deep \nlearning models outperform these approaches, with the SE -ResNet-34 network achieving 95.83% and \nthe CNN-based deep learning technique delivering the highest F1 score of 96.23% which indicates the \nperformance Metrix \n \nFigure 9. Comparing different performance metrics of error rate analysis \nFigure 9 shows the Error Rate (%) of different models, indicating their misfeatures ratio. The Gradient \nBoosted Decision Tree has the highest error rate at 12.6%, followed by the SE-ResNet-34 network with \n8.41%, while the CNN -based deep learning technique achieves the lowest error rate of 1.55%, \ndemonstrating its superior accuracy and minimal errors. \nCONCLUSION  \nThis study reviewed and theoretically evaluated various methods for detecting Endometriosis disorder, \nwith a particular focus on neural network algorithms. It provided a detailed description of previous \nresearch algorithms, highlighting their features, women health care data, analysis procedures, and \noutcomes. Additionally, the ultrasound dataset s used in these algorithms were briefly discussed. The \nlimitations identified in this review include a small number of datasets, imbalanced datasets, low \ndetection rates, and the absence of additional feature selection techniques. Furthermore, performance \nmetrics were used to evaluate three key approaches: CNN-based deep learning, SE-ResNet-34 network, \nand gradient-boosted decision trees. In the initial study, when ovarian cyst types were classified using a \nCNN-based deep learning technique, the accuracy of the classification models improved 98% with low \nerror rate 1.55%. This demonstrates that the enhanced performance directly contributes to improved \nclassification outcomes. \n75\n80\n85\n90\n95\n100\nCatBoost [21] support vector\nmachine (SVM)\n[8]\nGradient Boosted\nDecision Tree\nSE-ResNet-34\nnetwork\nCNN based deep\nlearning technique\nMAE (dB)\nDifferent Classification Methods\nPerformance Analysis of F1 Score\n0\n2\n4\n6\n8\n10\n12\n14\nGradient\nBoosted\nDecision Tree\nSE-ResNet-34\nnetwork\nCNN based\ndeep learning\ntechnique\nError Rate (%)\nError Rate (%)\n\nJosphin Mary J. et al: A comprehensive review ……  Archives for Technical Sciences 2025, 33(2), 500-515 \nTechnical Institute Bijeljina, Archives for Technical Sciences. Year XVII – N 0 33           513 \nFuture Scope \nFuture research must focus on collecting extensive datasets to refine the augmentation techniques and \nensure the production of more accurate and reliable findings. In the context of polycystic ovarian \nsyndrome detection, advanced deep learning methods offe r significant potential. Bridging the gap \nbetween informatics and medical experts is crucial this can be achieved by aligning the model's \ndevelopment with the specific needs of healthcare challenges, rather than solely focusing on the machine \nlearning aspect. Additionally, establishing clinical practice guidelines that leverage AI/ML models to \npredict histological types of ovarian lesions could revolutionize patient care. Furthermore, ongoing \nresearch into AI/ML techniques has already demonstrated improved predictive accuracy for ovarian \nborderline disease, laying the groundwork for more precise and effective diagnostic tools. \nREFERENCES \n[1] Horne AW, Missmer SA. Pathophysiology, diagnosis, and management of endometriosis. bmj. 2022 Nov \n14;379. https://doi.org/10.1136/bmj-2022-070750 \n[2] Bhandage, V ., Asuti, M. G., Siddappa, N. G., Challagidad, P. S., Benni, N. S., Cenitta, D., & Vijaya Arjunan, \nR. HRAESN-IoT: A Hybrid Residual Attention and Echo State Network Approach for IoT -Enabled Heart \nDisease Prediction. Journal of Wireless Mobile Ne tworks, Ubiquitous Computing, and Dependable \nApplications, 2025;16(1), 375-389. https://doi.org/10.58346/JOWUA.2025.I1.023 \n[3] Allaire C, Bedaiwy MA, Yong PJ. Diagnosis and management of endometriosis. Cmaj. 2023 Mar 14;195(10): \nE363-71. https://doi.org/10.1503/cmaj.220637 \n[4] Pai P, Amutha S, Patil S, Sridharan S, Shobha T, Arjunan RV . Slice Residual U-Net Based Rice Plant Disease \nClassification Using Convolutional Attentional BiGRU. Journal of Internet Services and Information \nSecurity. 2025;15(2):387-408. https://doi.org/10.58346/JISIS.2025.I2.028 \n[5] Priyadharshini M, Srimathi A, Sanjay C, Ramprakash K. Pcos disease prediction using machine learning \nalgorithms. Int Res J Adv Eng Hub (IRJAEH). 2024;2(03):651-5. \n[6] Umamaheswari TS. Mathematical and Visual Comprehension of Convolutional Neural Network Model for \nIdentifying Crop Diseases. International Academic Journal of Innovative Research. 2025;12(1):1 -7. \nhttps://doi.org/10.71086/IAJIR/V12I1/IAJIR1201 \n[7] Suha SA, Islam MN. Exploring the dominant features and data -driven detection of polycystic ovary \nsyndrome through modified stacking ensemble machine learning technique. Heliyon. 2023 Mar 1;9(3).  \n[8] Nayak A, Raghatate KS. Image segmentation and classification of aquatic plants using convolutional neural \nnetwork. International Journal of Aquatic Research and Environmental Studies. 2024;4:14 -9.  \nhttps://doi.org/10.70102/IJARES/V4S1/3 \n[9] Fan J, Liu J, Chen Q, Wang W, Wu Y . Accurate ovarian cyst classification with a lightweight deep learning \nmodel for ultrasound images. IEEE Access. 2023 Oct 2;11:110681 -91. \nhttps://doi.org/10.1109/ACCESS.2023.3321408 \n[10] Krishnan, R., & Dandekar, G. Medical Image Analysis for Disease Detection and Categorisation Using Deep \nLearning Algorithm. International Academic Journal of Science and Engineering, 2022;9(3), 16–20.  \n[11] Fazakis N, Kocsis O, Dritsas E, Alexiou S, Fakotakis N, Moustakas K. Machine learning tools for long-term \ntype 2 diabetes risk prediction. ieee Access. 2021 Jul 20;9:103737 -57. \nhttps://doi.org/10.1109/ACCESS.2021.3098691 \n[12] Shi S, Caluyo F, Hernandez R, Sarmiento J, Rosales CA. Automatic Classification and Identification of Plant \nDisease Identification by Using a Convolutional Neural Network. Natural and Engineering Sciences. \n2024;9(2):184-97.  https://doi.org/10.28978/nesciences.1569560 \n[13] Akhter F, Siddiqua A. Management of Adolescent PCOD: A Real Challenge. Sch Int J Obstet Gynec. \n2024;7(2):64-9. https://doi.org/10.36348/sijog.2024.v07i02.006 \n[14] Krishna AS, Anandaraj S, Shaik MS, Surarapu NJ, Donapati P. A novel approach to diagnosing PCOS \nthrough tunica albuginea oculi. InChallenges in Information, Communication and Computing Technology \n2024 (pp. 224-228). CRC Press. \n[15] Pushkarini H, Anusuya MA. A prediction model for evaluating the risk of developing PCOS. Int Res J Eng \nTechnol (IRJET). 2020 Sep;7(9):1150-6. \n[16] Zhang J, Zhang Q, Wang T, Song Y , Yu X, Xie L, Chen Y , Ouyang H. Multimodal MRI-based radiomics-\nclinical model for preoperatively differentiating concurrent endometrial carcinoma from atypical endometrial \nhyperplasia. Frontiers in oncology. 2022 May 27;12:887546. https://doi.org/10.3389/fonc.2022.887546 \n[17] Agrawal P, Tandon Sk, Kanhere A, Gupta P, Borasi M. Depression In Polycystic Ovarian Syndrome Patients \nand Effect of Body Image Perception And Quality Of Life. Int J Acad Med Pharm. 2023;5(3):392-8. \n\nJosphin Mary J. et al: A comprehensive review ……  Archives for Technical Sciences 2025, 33(2), 500-515 \nTechnical Institute Bijeljina, Archives for Technical Sciences. Year XVII – N 0 33           514 \n[18] Kumar A, Singh J, Khan AA. A comprehensive machine learning framework with particle swarm \noptimization for improved polycystic ovary syndrome (PCOS) diagnosis. Engineering Research Express. \n2024 Sep 11;6(3):035233. https://doi.org/10.1088/2631-8695/ad76f9 \n[19] Elsayed MA. Ultrasonic predictors of good ovarian response in infertile poly cystic ovarian disease patients. \nInternational Journal of Reproduction, Contraception, Obstetrics and Gynecology. 2023 Sep 1;12(9):2605 -\n11. https://doi.org/10.18203/2320-1770.ijrcog20232712 \n[20] Altal OF, Sindiani AM, Amin M, Mhanna HY , Hamad R, Gharaibeh H, Akhdar HF, Alhatamleh S, \nAlmahmoud RE, Abu-azzam OH, Balaw M. Hybrid Attention-Enhanced MobileNetV2 with Particle Swarm \nOptimization for Endometrial Cancer Classification in CT Images. Informatics in Medicine Unlocked. 2025 \nJun 10:101662. https://doi.org/10.1016/j.imu.2025.101662 \n[21] Visalaxi S, Muthu TS. Automated prediction of endometriosis using deep learning. International Journal of \nNonlinear Analysis and Applications. 2021 Nov 1;12(2):2403-16. https://doi.org/10.22075/ijnaa.2021.5383 \n[22] Kitaya K, Yasuo T, Yamaguchi T, Morita Y , Hamazaki A, Murayama S, Mihara T, Mihara M. Construction \nof deep learning -based convolutional neural network model for automatic detection of fluid hysteroscopic \nendometrial micropolyps in infertile women with chro nic endometritis. European Journal of Obstetrics & \nGynecology and Reproductive Biology. 2024 Jun 1;297:249 -53. \nhttps://doi.org/10.1016/j.ejogrb.2024.04.026 \n[23] Hu P, Gao Y , Zhang Y , Sun K. Ultrasound image-based deep learning to differentiate tubal -ovarian abscess \nfrom ovarian endometriosis cyst. Frontiers in Physiology. 2023 Feb 7;14:1101810.  \nhttps://doi.org/10.3389/fphys.2023.1101810 \n[24] Zhao A, Du X, Yuan S, Shen W, Zhu X, Wang W. Automated Detection of Endometrial Polyps from \nHysteroscopic Videos Using Deep Learning. Diagnostics 2023, 13, 1409 [Internet]. 2023 . \nhttps://doi.org/10.3390/diagnostics13081409 \n[25] James J, Govind S, Francis J. Ultrasound image classification and follicle segmentation for the diagnosis of \npolycystic ovary syndrome. InInternational conference on power engineering and intelligent systems (PEIS) \n2023 Jun 24 (pp. 145 -154). Singapore: Spr inger Nature Singapore.  https://doi.org/10.1007/978-981-99-\n7216-6_12 \n[26] Hassan MM, Mirza T. Comparative analysis of machine learning algorithms in diagnosis of polycystic \novarian syndrome. Int. J. Comput. Appl. 2020 Sep;975(8887). \n[27] Akanbi K, Adepoju OG, Nti KI. Developing A System for Automatic Prediction of Polycystic Ovary \nSyndrome Using Machine Learning. InProceedings of the 2024 7th International Conference on Machine \nLearning and Machine Intelligence (MLMI) 2024 Aug 2 (pp. 20 -26). \nhttps://doi.org/10.1145/3696271.3696275 \n[28] Agarwal M, Singh S, Jyoti C, Sinha S, Simran S, Simran Sr S. Understanding adolescent gynecological \nissues: a cross-sectional study at a tertiary care center. Cureus. 2024 Apr 6;16(4). \n[29] Al Ghazali BS, Monther S. Serum Mannose Level and Its Association in Women with Polycystic Ovarian \nSyndrome: a Case Control Study of Al -Najaf City/2020. Kufa Medical Journal. 2022 Jun 15;18(1):1 -9. \nhttps://doi.org/10.36330/kmj.v18i1.3273 \n[30] Nandipati SC, Ying CX. Polycystic Ovarian Syndrome (PCOS) classification and feature selection by \nmachine learning techniques. Applied Mathematics and Computational Intelligence (AMCI). 2020 Dec \n31;9:65-74. Retrieved from https://ejournal.unimap.edu.my/index.php/amci/article/view/151  \n[31] Crispim PC, Jammal MP, Murta EF, Nomelini RS. Endometriosis: what is the influence of immune cells?. \nImmunological investigations. 2021 May 19;50(4):372-88. https://doi.org/10.1080/08820139.2020.1764577 \n[32] Havaldar VD, Jadhav NY , Shinde SS, Mali SS, Mali KK, Shinde AA, Rajput RA. A review on PCOS: Its \ncauses, symptoms, pathogenesis and management. World. 2024;7(01):014 -21. \nhttps://doi.org/10.53346/wjapmr.2024.7.1.0041 \n[33] Premalatha M, Jayasudha M, Čep R, Priyadarshini J, Kalita K, Chatterjee P. A comparative evaluation of \nnature-inspired algorithms for feature selection problems. Heliyon. 2024 Jan 15;10(1). \n[34] Nouri B, Hashemi SH, Ghadimi DJ, Roshandel S, Akhlaghdoust M. Machine learning -based detection of \nendometriosis: a retrospective study in a population of iranian female patients. International Journal of \nFertility & Sterility. 2024 Oct 30;18(4):362. https://doi.org/10.22074/IJFS.2024.2009338.1519 \n[35] Goldstein A, Cohen S. Self -report symptom -based endometriosis prediction using machine learning. \nScientific Reports. 2023 Apr 4;13(1):5499. https://doi.org/10.1038/s41598-023-32761-8 \n[36] Chan C. Polycystic ovarian syndrome (pcos ) detection using gradient boosted decision tree. Authorea \nPreprints. 2023 Nov 2. \n[37] Wang Y , Zhang H, Wang T, Yao L, Zhang G, Liu X, Yang G, Yuan L. Deep learning for the ovarian lesion \nlocalization and discrimination between borderline and malignant ovarian tumors based on routine MR \nimaging. Scientific reports. 2023 Feb 16;13(1):2770. \n[38] Guha S, Kodipalli A, Fernandes SL, Dasar S. Explainable AI for interpretation of ovarian tumor classification \nusing enhanced ResNet50. Diagnostics. 2024 Jul 19;14(14):1567.  \nhttps://doi.org/10.3390/diagnostics14141567 \n\nJosphin Mary J. et al: A comprehensive review ……  Archives for Technical Sciences 2025, 33(2), 500-515 \nTechnical Institute Bijeljina, Archives for Technical Sciences. Year XVII – N 0 33           515 \n[39] Zhang Y , Gong C, Zheng L, Li X, Yang X. Deep learning for intelligent recognition and prediction of \nendometrial cancer. Journal of healthcare engineering. 2021;2021(1):1148309.  \nhttps://doi.org/10.1155/2021/1148309 \n[40] Ziyambe B, Yahya A, Mushiri T, Tariq MU, Abbas Q, Babar M, Albathan M, Asim M, Hussain A, Jabbar S. \nA deep learning framework for the prediction and diagnosis of ovarian cancer in pre -and post-menopausal \nwomen. Diagnostics. 2023 May 11;13(10):1703. https://doi.org/10.3390/diagnostics13101703 \n[41] Daniilidis A, Angioni S, Di Michele S, Dinas K, Gkrozou F, D’Alterio MN. Deep endometriosis and \ninfertility: what is the impact of surgery?. Journal of Clinical Medicine. 2022 Nov 14;11(22):6727.  \nhttps://doi.org/10.3390/jcm11226727 \n[42] Conroy I, Mooney SS, Kavanagh S, Duff M, Jakab I, Robertson K, Fitzgerald AL, Mccutchan A, Madden S, \nMaxwell S, Nair S. Pelvic pain: what are the symptoms and predictors for surgery, endometriosis and \nendometriosis severity. Australian and New Zealand Jour nal of Obstetrics and Gynaecology. 2021 \nOct;61(5):765-72. https://doi.org/10.1111/ajo.13379 \n[43] Shailaja K. Polycystic Ovary Syndrome. Asian J Pharm Res Health Care. 2023;15(4):307 –311. \nhttps://doi.org/10.4103/ajprhc.ajprhc_68_23 \n[44] Goswami P, Faujdar N, Singh G. Gynaecological Disease Diagnosis Expert System (GDDES) Based on \nMachine Learning Algorithm and Natural Language Processing. IEEE Access. 2024 May 27;12:84204 -15. \nhttps://doi.org/10.1109/ACCESS.2024.3406162 \n[45] Ren Z, Chen B, Hong C, Y uan J, Deng J, Chen Y , Ye J, Li Y . The value of machine learning in preoperative \nidentification of lymph node metastasis status in endometrial cancer: a systematic review and meta-analysis. \nFrontiers in Oncology. 2023 Dec 20;13:1289050. https://doi.org/10.3389/fonc.2023.1289050 \n[46] Xiao Y , Yang L, Wang Y , Wang Y , Chen Y , Lu W, Pei Z, Zhang R, Ye Y , Ji X, Liu S. Pulsed Low-Intensity \nfocused ultrasound (LIFU) activation of ovarian follicles. IEEE Open Journal of Engineering in Medicine \nand Biology. 2024 Apr 26;5:316-29. https://doi.org/10.1109/OJEMB.2024.3391939","source_license":"CC0","license_restricted":false}