Identification of histological features of ovarian high-grade serous carcinoma with homologous recombination deficiency using artificial intelligence: A retrospective analysis

Identification of histological features of ovarian high-grade serous carcinoma with homologous recombination deficiency using artificial intelligence: A retrospective analysis | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Identification of histological features of ovarian high-grade serous carcinoma with homologous recombination deficiency using artificial intelligence: A retrospective analysis Taira Hada, Morikazu Miyamoto, Takahiro Einama, Soichiro Kakimoto, and 14 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8830192/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 7 You are reading this latest preprint version Abstract Purpose This study aimed to identify the morphological features of ovarian high-grade serous carcinoma (HGSC) based on homologous recombination (HR) using artificial intelligence (AI). Methods Seventy-seven patients with HGSC who underwent HR status testing and surgery between 2006 and 2024 were included. One hematoxylin and eosin-stained slide per case, containing a sufficient volume of tumor tissue, was digitized. Tumor areas were automatically detected and annotated using AI. Nuclei in the tumor area were detected using AI. The area of each nucleus and the total number of nuclei were calculated automatically. A trained classifier determined the ratio of the tumor area to HR deficiency (HRD). Receiver operator characteristic curve established optimum cutoff value for average nucleus size (µm 2 ), average nucleus count per area (count/mm 2 ), and HRD area ratio (%). Results The area under the curve of average nucleus size, average nucleus count per tumor area, and HRD area ratio to the tumor area for the diagnosis of HRD were 0.704, 0.668, and 0.470, respectively, with cut-offs of 52.4 µm 2 , 5,610 count/mm 2 , and 40.0%, respectively. The HRD group had a smaller average nucleus size and larger average nucleus count per area than the HR proficiency group ( p < 0.01 and p = 0.03, respectively). The sensitivity and specificity for diagnosing HRD using the combined cutoff values for the average nucleus size and average nucleus count were 43.1% and 100.0%, respectively. Conclusion AI can identify the morphological features of HGSC with HRD and detect subtle tumor differences related to different genetic backgrounds. Ovarian high-grade serous carcinoma Artificial intelligence Homologous recombination status Nucleus size Figures Figure 1 Figure 2 Figure 3 Background Epithelial ovarian carcinoma (EOC) is the most aggressive gynecological cancer and the fifth leading cause of death among women in developed countries [ 1 ]. Among the histological subtypes, the incidence of high-grade serous carcinoma (HGSC) ranges from 70% to 80%, and the treatment of EOCs has been developed mainly based on HGSC [ 2 – 4 ]. The current standard management for EOCs is debulking surgery, followed by adjuvant platinum-based chemotherapy [ 5 ]. However, 70% of patients experience recurrence within two years, and the development of more effective treatments for EOCs is desirable [ 6 ]. Homologous recombination (HR) enables cells to access and copy intact deoxyribonucleic acid (DNA) sequence information in trans, particularly when DNA damage affects both strands of the double helix [ 7 – 9 ]. DNA replication and the repair of DNA double-strand breaks (DSBs) in somatic cells and during meiosis were repaired through the HR repair pathway [ 7 – 9 ]. This HR repair pathway involves various genes, such as the breast cancer susceptibility gene (BRCA), ataxia-telangiectasia mutated, and RAD51, and mutations in any of these genes may result in HR deficiency (HRD) phenotypes [ 9 , 10 ]. HRD has been reported to be associated not only with gene mutations that contribute to cancer development but also with the density and distribution of tumor-infiltrating lymphocytes within the tumor microenvironment of HGSC [ 11 , 12 ]. Additionally, several trials have demonstrated that HRD is an effective biomarker of the efficacy of poly-ADP-ribose inhibitors such as olaparib, niraparib, and rucaparib in patients with ovarian HGSC [ 13 – 16 ]. Recently, artificial intelligence (AI) has been used to evaluate morphological features in pathology. AI has proven to be a superior tool for detecting specific morphological features, quantifying the number or size of nuclei and lymphocytes, and analyzing the relationships between morphological features assessed by AI [ 17 , 18 ]. In ovarian HGSC, only a few studies have investigated the relationship between morphological characteristics and HR status using AI-based image analysis. Recent researches had demonstrated that deep learning models could predict HR status directly from whole-slide histopathological images of EOC [ 19 , 20 ]. These findings highlighted the potential of AI to estimate the genetic status from histopathological slides. However, the application of AI for the morphological classification of ovarian HGSC in relation to HR status remained limited. Further investigations were needed to clarify how morphological patterns observed in digital pathology according to HR status. Therefore, this study aimed to compare morphological features according to the HR status of HGSC using AI. Methods Seventy-seven HGSC patients with HR status, evaluated using the Myriad MyChoiceCDx assay, who underwent primary surgery between 2006 and 2024 at the National Defense Medical College Hospital were included. Patients without clinical information or surgical tissue samples, and those with a prior history of chemotherapy, were excluded. Clinical data were collected from the medical records. Pathological reviews were conducted in accordance with the 2020 World Health Organization criteria [ 21 ]. Staging was re-evaluated using the 2014 International Federation of Gynecology and Obstetrics criteria [ 22 ]. The characteristics of includes cases were demonstrated in Table 1 . In total, 58 cases with HRD and 19 cases with homologous recombination proficiency (HRP) were included in the analysis. Among cases with HRD, 28 cases had somatic mutations in breast cancer susceptibility genes. No statistically significant differences were observed in the clinical characteristics between the two groups. Table 1 The characteristics of 77 cases with ovarian high-grade serous carcinoma according to the homologous recombination status. Group with Homologous recombination deficiency Group with Homologous recombination proficiency Variables n = 58 n = 19 p -value Age (years) Average ± standard deviation 60.93 ± 11.15 60.79 ± 12.41 0.99 ≥60 32 (55.2) 10 (52.6) 0.99 <60 26 (44.8) 9 (47.4) International Federation of Obstetrics and Gynecology stage (%) 0.79 I 2 (3.5) 0 (0.0) II 2 (3.5) 1 (5.3) III 45 (77.5) 14 (73.7) IV 9 (15.5) 4 (21.0) Residual tumor (%) 0.16 Yes 38 (65.5) 16 (84.2) No 20 (34.5) 3 (15.8) Peritoneal cytology (%) 0.54 Positive 44 (75.9) 16 (84.2) Negative 14 (24.1) 3 (15.8) Somatic breast cancer susceptibility mutation (%) < 0.01 Positive 28 (48.3) 0 (0.0) Negative 30 (51.7) 19 (100.0) All hematoxylin and eosin (HE)-stained slides were reviewed pathologically. One slide per case containing sufficient tumor tissue was selected and digitized using a NanoZoomer SQ (Hamamatsu Photonics K.K., Shizuoka, Japan). Digital whole-slide images were uploaded to HALO software (Indica Labs, Corrales, NM, USA). To annotate the tumor area, we used DenseNet V2, an AI tissue classifier standard used in HALO, without data augmentation. The accuracy of the classifier was evaluated using the F1 scores. The F1 score indicated an agreement between automatic AI detection and manual annotation in the tumor area and defined as 2×(Precision×Recall)/(Precision+Recall). Precision is the proportion of true positive predictions (TP) among all predicted positive instances, calculated as TP/(TP + FP). Recall is the proportion of true positive predictions (TP) among all actual positive instances, calculated as TP/(TP + FN). In these formulas, TP refers to true positives, FP refers to false positives, and FN refers to false negatives [ 23 ]. The classifier was run on the entire slide for all cases, and the tumor area automatically detected by the AI was then annotated and used for the analysis. The methods used to compare morphological features according to HR status using AI were as follows: Nuclei were detected within the tumor area using the Nuclei Seg module without data augmentation. The Nuclei Seg module is a standard AI segmentation model equipped with HALO. The classifier was run on the annotated tumor area, and nuclei size ≥ 20 µm 2 were identified. The area of each nucleus and the total number of nuclei were automatically calculated using this classifier. Then, average nucleus size (µm 2 ) and average nucleus count per area (count/mm 2 ) in each case were calculated. The optimum cutoff values for average nucleus size and average nucleus count per area were determined using the receiver operating characteristic (ROC) curve for the diagnosis of HRD. The sensitivity and specificity for the diagnosis of HRD were evaluated using the combined cutoff values of the average nucleus size and nucleus count per area. Sensitivity was defined as the proportion of correctly identified true-positive cases, and specificity was defined as the proportion of true-negative cases. Finally, the comparisons between the two groups were conducted. Tumor areas were evaluated according to the HR status by AI as follows. Tumor area from cases of HRD and HRP were manually selected to define the areas for training. AI learned how to distinguish the HRD and HRP areas by training the selected areas using DenseNet V2. AI could automatically assign color-coded labels to the tumor areas based on this training. Once trained, AI automatically evaluated and classified the HRD and HRP areas in the tumor areas of remaining cases that were not used for training. The percentage of HRD areas within the tumor areas was automatically calculated for each case. Then, the optimum cutoff values for the ratio of HRD area to tumor area was determined using the ROC curve for the diagnosis of HRD, and cases were divided into two groups based on this cutoff value for further comparison. Lymphocytes were detected in the tumor area in the third step using Nuclei Seg. Trained lymphocytes were manually annotated, and the classifier was trained. The accuracy of segmentation was evaluated using the aggregated Jaccard index (AJI) scores. The AJI quantifies instance-level segmentation performance by aggregating intersection-over-union values across matched objects and penalizing unmatched predictions, indicating the overall agreement between AI-based and manually annotated lymphocyte segmentations [ 24 ]. In addition, the size of assessed lymphocytes was set up ≤ 20 µm 2 using Multiplex IHC. Multiplex IHC enables the simultaneous analysis of multiple markers by adjusting the color hues and size filters to accurately differentiate and quantify cells within a tissue sample. The classifier was run on the annotated tumor area, and lymphocytes were automatically detected. Then, average lymphocyte count per area (count/mm 2 ) in each case were calculated, and comparison between the two groups was performed. Statistical analyses were performed using the JMP Pro 17 software (SAS Institute Inc., Cary, NC, USA). The chi-square test, Fisher’s exact test, Mann-Whitney U test, and Wilcoxon test were used to evaluate the statistical significance of the clinical factors. The level of statistical significance was set at p < 0.05. Results The trained area was manually annotated as shown in Fig. 1 (a). Red indicated the tumor area, yellow indicated annotation boundaries, blue indicated background areas, such as the stroma or necrotic tissue, and purple indicated no tissue area. A total of 2.215 mm 2 for tumor area from six cases was selected, and the classifier was trained for 8,190 iterations. The F1 score was evaluated for the three cases that were not used for training, and the F1 score value was 0.83. The classifier was run on the entire slide for all cases (Fig. 1 (b)), and the tumor areas were automatically annotated as shown in Fig. 1 (c). The average tumor area size was 179.51 (7.46-368.59) mm 2 . Nuclei were detected and calculated as demonstrated in Fig. 1 (d). The average number of nuclei was 867,055 (33,251-1,044,690) count. The ROC curves of the average nucleus size and average nucleus count per area for the diagnosis of HRD were shown in Fig. 2 (a)(b). The areas under the curve were 0.704 and 0.668, and the cutoff values were 52.4 µm 2 and 5,610 count/mm 2 , respectively. The HRD and HRP areas were classified according to the workflow shown in Fig. 3 . A total of 3.069 mm² of tumor area from 3 cases of HRD and 2.744 mm² from 3 cases of HRP were manually selected. AI trained the selected areas for 5,015 iterations and could automatically assign color-coded labels to the tumor areas based on this training. As results, AI assigned green to HRD areas and purple to HRP areas. Once trained, AI automatically evaluated and classified the HRD and HRP areas in the tumor areas of remaining cases that were not used for training. Finally, the percentage of HRD areas within the tumor areas was automatically calculated for each case. The ROC curves of the ratio of the HRD area to the tumor area for the diagnosis of HRD was shown in Fig. 2 (c). The area under the curve was 0.470, and the cutoff value was 40.0%. To train lymphocytes, a total of 2,117 lymphocytes from six cases were manually selected. The classifier was trained for 8,005 iterations. The accuracy of segmentation was evaluated using the AJI scores of 951 lymphocytes from three cases that were not used for training, and the AJI score was 0.56. Lymphocytes were detected and calculated as demonstrated in Fig. 1 (e). The average number of lymphocytes was 59,054 (8,328 − 320,967) count. The results of the comparison between the HRD and HRP groups regarding the characteristics and morphological features assessed using AI are shown in Table 2 . The HRD group had a smaller average nucleus size and larger average nucleus count per area than the HRP group ( p < 0.01 and p = 0.03, respectively). In addition, cases of average nucleus size less than 52.4 µm 2 and of average nucleus count per area with 5,160 count/mm 2 or more were seen more frequently in the Group with HRD than the Group with HRP ( p < 0.01 and p < 0.01). Furthermore, the sensitivity and specificity for diagnosing HRD using the combined cutoff values for the average nucleus size and average nucleus count were 43.1% and 100.0%, respectively. Table 2 The morphological features assessed by artificial intelligence of 77 cases with ovarian high-grade serous carcinoma according to the homologous recombination status. Group with Homologous recombination deficiency Group with Homologous recombination proficiency Variables n = 58 n = 19 p -value Average nucleus size (µm 2 ) (%) Average ± standard deviation 50.64 ± 6.87 55.39 ± 5.94 < 0.01 <52.4 36 (62.1) 4 (21.0) < 0.01 ≥52.4 22 (37.9) 15 (79.0) Average nucleus counts per area (count/mm 2 ) Average ± standard deviation 5610.96 ± 1597.21 4751.04 ± 1282.22 0.03 ≥5160.0 34 (58.6) 4 (21.0) < 0.01 <5160.0 24 (41.4) 15 (79.0) Average nucleus size < 52.4 µm 2 and average nucleus counts per area ≥ 5160.0 count/mm 2 < 0.01 Yes 25 (43.1) 0 (0.0) No 33 (56.9) 19 (100.0) The ratio of the homologous recombination deficiency area among tumor area (%) a) 0.20 ≥40.0% 17 (30.9) 2 (12.5) <40.0% 38 (69.1) 14 (87.5) Average lymphocyte count per area (count/mm 2 ) Average ± standard deviation 401.13 ± 278.85 287.10 ± 214.33 0.10 a) Three cases of homologous recombination deficiency and three cases of homologous recombination proficiency were excluded from analysis. Discussion In our study, we demonstrated that the HRD group had a smaller average nucleus size and a larger nucleus count per unit area than the HRP group, with high F1 and AJI scores. In addition, the sensitivity and specificity for the diagnosis of HRD using the cutoff values of the average nucleus size and average density of nuclei per unit area were 43.1% and 100.0%, respectively. The F1 and AJI scores measured the agreement between the automatic detection by AI and manual annotation. Both values ranged from 0 to 1.0, with a value closer to 1.0, indicating higher precision [ 23 , 24 ]. The F1 score was calculated as the harmonic mean of precision (P) and recall (R) using the formula [2PR/(P + R)] [ 23 ]. In previous studies involving AI-based pathological evaluations, F1 scores ranged from 0.78 to 0.97 [ 25 – 27 ]. The F1 score in our study was within this range, suggesting that the accuracy of our method is comparable to that of previous reports. Hence, the AJI score considers pixel-level (segmentation) errors, in contrast to the F1 score, and can treat unified detection and segmentation [ 24 , 28 ]. A study that compared the quality of nuclear segmentation among four open-source software programs showed that the overall AJI score ranged from 0.12 to 0.51 [ 24 ]. Our study preserved the accuracy of AJI scores. First, the AI detected the nuclei using a built-in function. Additionally, among these nuclei, nuclei size ≥ 20 µm 2 were defined as the nuclei. The average nucleus size was smaller, and the average nucleus count per area, which was equal to the average cell count per area, was larger in the HRD group than in the HRP group. Cellular responses to DSBs damage include mechanisms that restore the structure of broken chromosomes, and HR is associated with the maintenance of chromosomal structure [ 29 , 30 ]. Some nuclear chromosomal fragments caused by DNA DSBs flow out of the nucleus and remain in the cytoplasm [ 31 ]. Therefore, we assumed that the nuclei of cancer cells with HRD were smaller than those with HRP, and that the density of the nuclei of cancer cells with HRD was larger than that of HRP. DNA repair mechanisms, such as the HR status, may influence the size of nuclei or cells and the appearance of morphological structures. Second, we trained the AI model using manually annotated cancer cell structures based on the built-in classifier. Subsequently, the AI utilized this learned information to classify the structures in the remaining cases. A previous study reported that tumors with HRD may exhibit distinct structures, such as solid, pseudo-endometrioid, and transitional-like morphologies, compared with HRP tumors [ 32 ]. Based on this, we trained the AI to learn the randomized tumor area rather than each tumor cell, as the structures constructed by tumor cells using 3 cases with HRD or HRP, and the AI was used to evaluate the remaining cases. However, the AI determined the area of the HRD or HRP in each case. Therefore, this training method for memorizing the tumor area was not effective in determining the HR status. The lymphocytes were identified using a built-in AI segmentation module. Of these, cells with an area less than or equal to 20 µm² were defined as lymphocytes. There were no differences in lymphocyte counts between HGSC with HRD and those with HRP, which is inconsistent with previous literature [ 11 , 12 ]. Previous studies have demonstrated a relationship between HR status and lymphocytes with a cluster of differentiation 3 or cluster of differentiation 8. The present study did not classify the lymphocyte clusters because they were assessed using hematoxylin and eosin staining. Further studies are needed to categorize clusters of lymphocytes to explore the relationship between lymphocytes and HR status. The review of literature on the morphological comparison based on AI according to the HR status including our study was summarized in Table 3 . Previous studies reported the potential of deep learning models for predicting HR status through the analysis of tumor area from histopathological images [ 19 , 20 ]. Particularly, Zhang et al. achieved a precision of 0.800 and recall of 0.727 [ 19 ]. However, in our study, we did not reach the same level of the analysis for tumor area. This discrepancy might develop from the differences of AI devices or the limited number of cases included in our study. On contrast, our study also examined nuclear characteristics and the presence of lymphocytes. We demonstrated the significant differences about the nuclear size and density between the HRD and HRP groups and 100.0% of the specificity for the diagnosis of HRD using the combined cutoff values for the average nucleus size and average nucleus density. These findings were the novel findings of our study and might suggest that nuclear morphology might be a potential histopathological biomarker for HRD detection in the future. Table 3 Review of literature on the morphological comparison based on artificial Intelligence according to the homologous recombination status including the present study. Main findings Author Sample size Input data Analysis of nuclei Analysis of tumor regions Analysis of lymphocytes Zhang K et al. [ 19 ] 205 cases HRD: 64 cases HRP: 141 cases H&E-stained whole-slide images Not conducted Precision 0.800, Recall 0.727 Not conducted Frenel J-S et al. [ 20 ] 93 cases (HRD subtypes was not specified) H&E-stained whole-slide images Not conducted Detailed metrics not available in abstract. Not conducted The present study 77 cases HRD: 58 cases HRP: 19 cases H&E-stained whole-slide images HRD group had a smaller average nucleus size and larger average nucleus density ( p < 0.01, p = 0.03). There was no statistical difference (Precision 0.895, Recall 0.309, p = 0.20). There was no statistical difference ( p = 0.10). Abbreviations HRD: Homologous recombination deficiency, HRP: Homologous recombination proficiency, HE: Hematoxylin and eosin This study has several limitations, including the small sample size from a single institution and its retrospective design. Further large-scale prospective studies are warranted to validate the clinical and pathological significance of HGSC stratified according to HR status. Conclusions Our study demonstrates that AI may identify the morphological features of HGSC with HRD and may be useful for identifying several other small changes among tumors with different genetic backgrounds. Abbreviations EOC: epithelial ovarian carcinoma HGSC: high-grade serous carcinoma HR: homologous recombination DNA: deoxyribonucleic acid DSBs: double-strand breaks BRCA: breast cancer susceptibility gene HRD: homologous recombination deficiency AI: artificial intelligence HRP: homologous recombination proficiency ROC: receiver operator characteristic Declarations Funding The authors declare that no funds, grants, or other support were received during the preparation of this manuscript. Competing interests The authors have no relevant financial or non-financial interests to disclose. Authors’ contributions Protocol/project development: TH, MM, TE, KS, and MT. Data collection and management: TH, MM, SK, KO, and YH. Data analysis: TH, TE, MK, TW, YO, JS, TI, NK, RT, SN, KK, and HS. Manuscript writing/editing: TH, MM, and MT. All authors read and approved the final manuscript. Ethics approval and consent to participate This study was approved by the Institutional Review Board of National Defense Medical College, Tokorozawa, Japan, and was conducted in accordance with the Declaration of Helsinki. The records and information of all women in the study were anonymized and de-identified prior to analysis. The study was exempt from obtaining informed consent from all participants. Consent for publication Not applicable Clinical trial number Not applicable Availability of data and materials The datasets used and/or analyzed in the current study are available from the corresponding author upon reasonable request. Acknowledgments The authors would like to thank Editage (www.editage.jp) for the English language editing. We also appreciate the Defense Medicine Basic Research Program B and Indica Labs Inc. References Siegel RL, Miller KD, Fuchs HE, Jemal A (2021) Cancer statistics, 2021. CA Cancer J Clin 71:7–33. https://doi.org/10.3322/caac.21654 Prat J (2012) Ovarian carcinomas: five distinct diseases with different origins, genetic alterations, and clinicopathological features. Virchows Arch 460:237–249. https://doi.org/10.1007/s00428-012-1203-5 Vergote I, Tropé CG, Amant F, Kristensen GB, Ehlen T, Johnson N, et al (2010) Neoadjuvant chemotherapy or primary surgery in stage IIIC or IV ovarian cancer. N Engl J Med 363:943–953. https://doi.org/10.1056/NEJMoa0908806 Kehoe S, Hook J, Nankivell M, Jayson GC, Kitchener H, Lopes T, et al (2015) Primary chemotherapy versus primary surgery for newly diagnosed advanced ovarian cancer (CHORUS): an open-label, randomised, controlled, non-inferiority trial. Lancet 386:249–257. https://doi.org/10.1016/S0140-6736(14)62223-6 Armstrong DK, Alvarez RD, Bakkum-Gamez JN, Barroilhet L, Behbakht K, Berchuck A, et al (2021) Ovarian Cancer, Version 2.2020, NCCN Clinical Practice Guidelines in Oncology. J Natl Compr Canc Netw 19:191–226. https://doi.org/10.6004/jnccn.2021.0007 International Collaborative Ovarian Neoplasm Group (2002) Paclitaxel plus carboplatin versus standard chemotherapy with either single-agent carboplatin or cyclophosphamide, doxorubicin, and cisplatin in women with ovarian cancer: the ICON3 randomised trial. Lancet 360:505–515. https://doi.org/10.1016/S0140-6736(02)09738-6 Wright WD, Shah SS, Heyer WD (2018) Homologous recombination and the repair of DNA double-strand breaks. J Biol Chem 293:10524–10535. https://doi.org/10.1074/jbc.TM118.000372 Haber JE (1999) DNA recombination: the replication connection. Trends Biochem Sci 24:271–275. https://doi.org/10.1016/s0968-0004(99)01413-9 Doig KD, Fellowes AP, Fox SB (2023) Homologous recombination repair deficiency: an overview for pathologists. Mod Pathol 36:100049. https://doi.org/10.1016/j.modpat.2022.100049 Heeke AL, Pishvaian MJ, Lynce F, Xiu J, Brody JR, Chen WJ, et al (2018) Prevalence of homologous recombination-related gene mutations across multiple cancer types. JCO Precis Oncol 2018:PO.17.00286. https://doi.org/10.1200/PO.17.00286 Morse CB, Toukatly MN, Kilgore MR, Agnew KJ, Bernards SS, Norquist BM, et al (2019) Tumor infiltrating lymphocytes and homologous recombination deficiency are independently associated with improved survival in ovarian carcinoma. Gynecol Oncol 153:217-222. http://doi.org/10.1016/j.ygyno.2019.02.011 Strickland KC, Howitt BE, Shukla SA, Rodig S, Ritterhouse LL, Liu JF, et al (2016) Association and prognostic significance of BRCA1/2-mutation status with neoantigen load, number of tumor-infiltrating lymphocytes and expression of PD-1/PD-L1 in high grade serous ovarian cancer. Oncotarget 7:13587-98. http://doi.org/10.18632/oncotarget.7277 Loeb LA (2011) Human cancers express mutator phenotypes: origin, consequences and targeting. Nat Rev Cancer 11:450–457. https://doi.org/10.1038/nrc3063 Ray-Coquard I, Leary A, Pignata S, Cropet C, González-Martín A, Marth C, et al (2023) Olaparib plus bevacizumab first-line maintenance in ovarian cancer: final overall survival results from the PAOLA-1/ENGOT-ov25 trial. Ann Oncol 34:681–692. https://doi.org/10.1016/j.annonc.2023.05.005 Monk BJ, Barretina-Ginesta MP, Pothuri B, Vergote I, Graybill W, Mirza MR, et al (2024) Niraparib first-line maintenance therapy in patients with newly diagnosed advanced ovarian cancer: final overall survival results from the PRIMA/ENGOT-OV26/GOG-3012 trial. Ann Oncol 35:981–992. https://doi.org/10.1016/j.annonc.2024.08.2241 Monk BJ, Parkinson C, Lim MC, O’Malley DM, Oaknin A, Wilson MK, et al (2022) A randomized, Phase III trial to evaluate Rucaparib monotherapy as maintenance treatment in patients with newly diagnosed ovarian cancer (ATHENA-MONO/GOG-3020/ENGOT-ov45). J Clin Oncol 40:3952–3964. https://doi.org/10.1200/JCO.22.01003 Kouzu K, Tsujimoto H, Imamura Y, Kajiwara Y, Nearchou IP, Einama T, et al (2024) Development and validation study of the prognostic impact of deep learning-determined myxoid stroma after neoadjuvant chemotherapy in patients with esophageal squamous cell carcinoma. Ann Surg Oncol 31:6300–6308. https://doi.org/10.1245/s10434-024-15626-w Kouzu K, Tsujimoto H, Nearchou IP, Einama T, Watanabe T, Horiguchi H, et al (2025) Prognostic impact of tumor cell nuclear size assessed by artificial intelligence in esophageal squamous cell carcinoma. Lab Invest 105:102221. https://doi.org/10.1016/j.labinv.2024.102221 Zhang K, Qiu Y, Feng S, Yin H, Liu Q, Zhu Y, et al (2025) Development of model for identifying homologous recombination deficiency (HRD) status of ovarian cancer with deep learning on whole slide images. J Transl Med 23:267. http://doi.org/10.1186/s12967-025-06234-7 Frenel J-S, Bossard C, Rynkiewicz J, Thomas F, Salhi Y, Salhi S, et al (2024) Artificial intelligence to predict homologous recombination deficiency in ovarian cancer from whole-slide histopathological images. J Clin Oncol 42(16_suppl):5578. https://doi.org/10.1200/JCO.2024.42.16_suppl.5578 Lokuhetty D, White VA, Cree IA, editors (2020) WHO classification of tumours of female reproductive organs. 5th ed. Lyon: IARC Pereira A, Pérez-Medina T, Magrina JF, Magtibay PM, Rodríguez-Tapia A, Peregrin I, et al (2015) International Federation of Gynecology and Obstetrics staging classification for cancer of the ovary, fallopian tube, and peritoneum: estimation of survival in patients with node-positive epithelial ovarian cancer. Int J Gynecol Cancer 25:49–54. https://doi.org/10.1097/IGC.0000000000000316 Ghosh AK, Thammasudjarit R, Jongkhajornpong P, Attia J, Thakkinstian A (2022) Deep learning for discrimination between fungal keratitis and bacterial keratitis: DeepKeratitis. Cornea 41:616–622. https://doi.org/10.1097/ICO.0000000000002830 Kumar N, Verma R, Sharma S, Bhargava S, Vahadane A, Sethi A (2017) A dataset and a technique for generalized nuclear segmentation for computational pathology. IEEE Trans Med Imaging 36:1550–1560. https://doi.org/10.1109/TMI.2017.2677499 Saha M, Guo X, Sharma A (2021) TilGAN : GAN for facilitating tumor-infiltrating lymphocyte pathology image synthesis with improved image classification. IEEE Access 9:79829–79840. https://doi.org/10.1109/access.2021.3084597 Barua B, Chyrmang G, Bora K, Ahmed GN, Kakoti L, Saikia MJ (2025) Classifying tumour infiltrating lymphocytes in oral squamous cell carcinoma histopathology using joint learning framework. Sci Rep 15:2938. https://doi.org/10.1038/s41598-025-86527-5 Swiderska-Chadaj Z, Pinckaers H, van Rijthoven M, Balkenhol M, Melnikova M, Geessink O, et al (2019) Learning to detect lymphocytes in immunohistochemistry with deep learning. Med Image Anal. 58:101547. https://doi.org/10.1016/j.media.2019.101547 You Z, Jiang M, Shi Z, Ning X, Shi C, Du S, et al (2021) Evaluation of automated segmentation algorithms for neurons in macaque cerebral microscopic images. Microsc Res Tech 84:2311–24. https://doi.org/10.1002/jemt.23786 [EPub] Gospodinov A, Herceg Z (2013) Chromatin structure in double strand break repair. DNA Repair (Amst) 12:800–10. https://doi.org/10.1016/j.dnarep.2013.07.006 Verma P, Greenberg RA (2021) Communication between chromatin and homologous recombination. Curr Opin Genet Dev 71:1–9. https://doi.org/10.1016/j.gde.2021.05.006 Hopfner KP, Hornung V (2020) Molecular mechanisms and cellular functions of cGAS-STING signalling. Nat Rev Mol Cell Biol 21:501–521. https://doi.org/10.1038/s41580-020-0244-x Chapagain U, Huecker JB, Sun L (2025) Morphologic correlations with homologous recombination deficiency in high-grade serous carcinomas. Int J Gynecol Pathol 44:398-406. http://doi.org/10.1097/PGP.0000000000001090 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 13 Apr, 2026 Reviews received at journal 30 Mar, 2026 Reviewers agreed at journal 20 Mar, 2026 Reviewers invited by journal 02 Mar, 2026 Editor assigned by journal 10 Feb, 2026 Submission checks completed at journal 10 Feb, 2026 First submitted to journal 09 Feb, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8830192","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":599821905,"identity":"832dd924-bbdd-44a4-8eb5-f4545ffdeee6","order_by":0,"name":"Taira Hada","email":"data:image/png;base64,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","orcid":"","institution":"National Defense Medical College Hospital","correspondingAuthor":true,"prefix":"","firstName":"Taira","middleName":"","lastName":"Hada","suffix":""},{"id":599821906,"identity":"d643bc2a-bdb0-481f-8813-7efdfd05b398","order_by":1,"name":"Morikazu Miyamoto","email":"","orcid":"","institution":"National Defense Medical College Hospital","correspondingAuthor":false,"prefix":"","firstName":"Morikazu","middleName":"","lastName":"Miyamoto","suffix":""},{"id":599821908,"identity":"36991982-7d76-4ae1-a77a-a9eeb69a742a","order_by":2,"name":"Takahiro Einama","email":"","orcid":"","institution":"National Defense Medical College Hospital","correspondingAuthor":false,"prefix":"","firstName":"Takahiro","middleName":"","lastName":"Einama","suffix":""},{"id":599821909,"identity":"5a1d1e55-4e95-4c47-bf7a-7c03cf769f00","order_by":3,"name":"Soichiro Kakimoto","email":"","orcid":"","institution":"Self-Defense Force Central Hospital","correspondingAuthor":false,"prefix":"","firstName":"Soichiro","middleName":"","lastName":"Kakimoto","suffix":""},{"id":599821910,"identity":"db49623a-a851-4721-b1d9-7fbddfea1cd1","order_by":4,"name":"Makiko Koga","email":"","orcid":"","institution":"National Defense Medical College","correspondingAuthor":false,"prefix":"","firstName":"Makiko","middleName":"","lastName":"Koga","suffix":""},{"id":599821911,"identity":"7d8f9295-26ce-4016-b53a-85bbd7ec5e27","order_by":5,"name":"Takanori Watanabe","email":"","orcid":"","institution":"National Defense Medical College","correspondingAuthor":false,"prefix":"","firstName":"Takanori","middleName":"","lastName":"Watanabe","suffix":""},{"id":599821912,"identity":"29aab75b-318a-44b1-aa0f-fa80fc66f5d9","order_by":6,"name":"Yuka Otsuka","email":"","orcid":"","institution":"National Defense Medical College Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yuka","middleName":"","lastName":"Otsuka","suffix":""},{"id":599821913,"identity":"565e2724-0b14-464d-b5a1-9847e6a5e2d1","order_by":7,"name":"Jin Suminokura","email":"","orcid":"","institution":"National Defense Medical College Hospital","correspondingAuthor":false,"prefix":"","firstName":"Jin","middleName":"","lastName":"Suminokura","suffix":""},{"id":599821914,"identity":"079e279c-6a9c-4622-a056-8ce739a54744","order_by":8,"name":"Tsubasa Ito","email":"","orcid":"","institution":"National Defense Medical College Hospital","correspondingAuthor":false,"prefix":"","firstName":"Tsubasa","middleName":"","lastName":"Ito","suffix":""},{"id":599821915,"identity":"b53fd0f9-079c-4104-8b02-2086e9ba54ad","order_by":9,"name":"Naohisa Kishimoto","email":"","orcid":"","institution":"National Defense Medical College Hospital","correspondingAuthor":false,"prefix":"","firstName":"Naohisa","middleName":"","lastName":"Kishimoto","suffix":""},{"id":599821916,"identity":"3ea8f232-8d26-4e9e-9616-32fa4458c6fe","order_by":10,"name":"Risa Tanabe","email":"","orcid":"","institution":"National Defense Medical College Hospital","correspondingAuthor":false,"prefix":"","firstName":"Risa","middleName":"","lastName":"Tanabe","suffix":""},{"id":599821917,"identity":"a64507d5-470f-4e70-aad3-cad1ce915733","order_by":11,"name":"Soko Nishimura","email":"","orcid":"","institution":"National Defense Medical College Hospital","correspondingAuthor":false,"prefix":"","firstName":"Soko","middleName":"","lastName":"Nishimura","suffix":""},{"id":599821919,"identity":"a3c888b8-9b81-49da-a045-17f10d7849f9","order_by":12,"name":"Kento Kato","email":"","orcid":"","institution":"National Defense Medical College Hospital","correspondingAuthor":false,"prefix":"","firstName":"Kento","middleName":"","lastName":"Kato","suffix":""},{"id":599821920,"identity":"08e2367b-47b1-4872-b1d2-5845f3a5ab85","order_by":13,"name":"Hiroaki Soyama","email":"","orcid":"","institution":"National Defense Medical College Hospital","correspondingAuthor":false,"prefix":"","firstName":"Hiroaki","middleName":"","lastName":"Soyama","suffix":""},{"id":599821921,"identity":"59a3d24c-726c-4b08-8564-80d69b88f8f0","order_by":14,"name":"Kohei Omatsu","email":"","orcid":"","institution":"National Defense Medical College Hospital","correspondingAuthor":false,"prefix":"","firstName":"Kohei","middleName":"","lastName":"Omatsu","suffix":""},{"id":599821922,"identity":"8d269a42-23b0-4d4c-abf6-962f0b3bbd89","order_by":15,"name":"Yoshinobu Hamada","email":"","orcid":"","institution":"National Defense Medical College Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yoshinobu","middleName":"","lastName":"Hamada","suffix":""},{"id":599821923,"identity":"707c5d99-67ac-47f7-ace3-f0201677bab3","order_by":16,"name":"Kimiya Sato","email":"","orcid":"","institution":"National Defense Medical College","correspondingAuthor":false,"prefix":"","firstName":"Kimiya","middleName":"","lastName":"Sato","suffix":""},{"id":599821927,"identity":"63d99afe-2b6b-46a4-b9a7-69341a863d04","order_by":17,"name":"Masashi Takano","email":"","orcid":"","institution":"National Defense Medical College Hospital","correspondingAuthor":false,"prefix":"","firstName":"Masashi","middleName":"","lastName":"Takano","suffix":""}],"badges":[],"createdAt":"2026-02-09 12:08:24","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8830192/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8830192/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":104181283,"identity":"283b105b-2d60-4734-8db5-4d6a8d14699e","added_by":"auto","created_at":"2026-03-08 17:27:02","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":477750,"visible":true,"origin":"","legend":"\u003cp\u003eRepresentative images of the manually annotated area for training, the area detected by the classifier, the final annotated tumor area, nuclei detected by artificial intelligence, and lymphocytes detected by artificial intelligence.\u003c/p\u003e\n\u003cp\u003eThis figure is shown here to demonstrate both the workflow and representative results. (a) Manually annotated area used for training: red indicates tumor regions, yellow indicates annotated regions, blue indicates stroma or necrotic tissue, and purple indicates non-tissue areas. (b) Area detected by the classifier: red indicates tumor regions, yellow and blue indicate background regions such as stroma or necrotic tissue, and purple indicates non-tissue areas. (c) Final annotated tumor area: the area enclosed by a red line indicates the tumor region, while the area enclosed by both red and black lines within the tumor region represents the area excluded from the analysis. (d) Blue indicates nuclei that were automatically identified by artificial intelligence. The area of each nucleus and the total number of nuclei were automatically calculated, and average nucleus size (μm\u003csup\u003e2\u003c/sup\u003e) and density (count/mm\u003csup\u003e2\u003c/sup\u003e) in each case were calculated. (e)Yellow indicates lymphocytes identified by artificial intelligence. The total number of lymphocytes were automatically calculated, and the average lymphocytes density (count/mm\u003csup\u003e2\u003c/sup\u003e) in each case were calculated.\u003c/p\u003e","description":"","filename":"Figure1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8830192/v1/68d8fcf7ad7bba16800e3df7.jpg"},{"id":104181282,"identity":"2cf8320a-12ae-4ee2-9247-b43ac0adfdf6","added_by":"auto","created_at":"2026-03-08 17:27:02","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":116941,"visible":true,"origin":"","legend":"\u003cp\u003eReceiver operating characteristic curve analysis of average nucleus size (μm²) (a), average nucleus count per area (count/mm²) (b), and the ratio of homologous recombination deficiency area within the tumor area (%) (c), in comparison with the diagnosis of homologous recombination deficiency.\u003c/p\u003e\n\u003cp\u003eThe area under the curve was 0.704, 0.668, and 0.470, respectively.\u003c/p\u003e","description":"","filename":"Figure2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8830192/v1/6b28e6005b6ad6ca6079423c.jpg"},{"id":104181281,"identity":"2719983e-268e-4989-8b92-3fa0276358c4","added_by":"auto","created_at":"2026-03-08 17:27:02","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":420229,"visible":true,"origin":"","legend":"\u003cp\u003eWorkflow of artificial intelligence (AI) based analysis to evaluate tumor areas according to homologous recombination (HR) status.\u003c/p\u003e\n\u003cp\u003e(a) Three cases with HR deficiency (HRD) and 3 cases with HR proficiency (HRP) were randomly selected. (b) Learning phase: A total of 3.069 mm² of tumor area from 3 cases of HRD and 2.744 mm² from 3 cases of HRP were manually selected. AI trained these areas for 5,015 iterations and was able to automatically assign color-coded labels to the tumor areas based on the learning. Green represented HRD areas and purple represented HRP areas. (c) Evaluating phase: AI automatically evaluated and classified tumor areas of remaining cases that were not used for training. (d) The percentage of HRD area within the tumor areas was automatically calculated for each case. The cutoff value for the diagnosis of HRD was determined by the receiver operating characteristic curves of the ratio of the HRD area to the tumor area.\u003c/p\u003e\n\u003cp\u003eAbbreviations\u003c/p\u003e\n\u003cp\u003eHRD: Homologous recombination deficiency, HRP: Homologous recombination proficiency, ROC: Receiver operating characteristic.\u003c/p\u003e","description":"","filename":"Figure3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8830192/v1/e638983ce6f303f798ea0702.jpg"},{"id":104779449,"identity":"e6dca979-12ea-441e-b9c4-c66c60bcba09","added_by":"auto","created_at":"2026-03-17 07:40:25","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1833715,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8830192/v1/f51d0e3d-158b-49d3-997f-7d0f7fb9f085.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Identification of histological features of ovarian high-grade serous carcinoma with homologous recombination deficiency using artificial intelligence: A retrospective analysis","fulltext":[{"header":"Background","content":"\u003cp\u003eEpithelial ovarian carcinoma (EOC) is the most aggressive gynecological cancer and the fifth leading cause of death among women in developed countries [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Among the histological subtypes, the incidence of high-grade serous carcinoma (HGSC) ranges from 70% to 80%, and the treatment of EOCs has been developed mainly based on HGSC [\u003cspan additionalcitationids=\"CR3\" citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. The current standard management for EOCs is debulking surgery, followed by adjuvant platinum-based chemotherapy [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. However, 70% of patients experience recurrence within two years, and the development of more effective treatments for EOCs is desirable [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eHomologous recombination (HR) enables cells to access and copy intact deoxyribonucleic acid (DNA) sequence information in trans, particularly when DNA damage affects both strands of the double helix [\u003cspan additionalcitationids=\"CR8\" citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. DNA replication and the repair of DNA double-strand breaks (DSBs) in somatic cells and during meiosis were repaired through the HR repair pathway [\u003cspan additionalcitationids=\"CR8\" citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. This HR repair pathway involves various genes, such as the breast cancer susceptibility gene (BRCA), ataxia-telangiectasia mutated, and RAD51, and mutations in any of these genes may result in HR deficiency (HRD) phenotypes [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. HRD has been reported to be associated not only with gene mutations that contribute to cancer development but also with the density and distribution of tumor-infiltrating lymphocytes within the tumor microenvironment of HGSC [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. Additionally, several trials have demonstrated that HRD is an effective biomarker of the efficacy of poly-ADP-ribose inhibitors such as olaparib, niraparib, and rucaparib in patients with ovarian HGSC [\u003cspan additionalcitationids=\"CR14 CR15\" citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eRecently, artificial intelligence (AI) has been used to evaluate morphological features in pathology. AI has proven to be a superior tool for detecting specific morphological features, quantifying the number or size of nuclei and lymphocytes, and analyzing the relationships between morphological features assessed by AI [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. In ovarian HGSC, only a few studies have investigated the relationship between morphological characteristics and HR status using AI-based image analysis. Recent researches had demonstrated that deep learning models could predict HR status directly from whole-slide histopathological images of EOC [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. These findings highlighted the potential of AI to estimate the genetic status from histopathological slides. However, the application of AI for the morphological classification of ovarian HGSC in relation to HR status remained limited. Further investigations were needed to clarify how morphological patterns observed in digital pathology according to HR status.\u003c/p\u003e \u003cp\u003eTherefore, this study aimed to compare morphological features according to the HR status of HGSC using AI.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003eSeventy-seven HGSC patients with HR status, evaluated using the Myriad MyChoiceCDx assay, who underwent primary surgery between 2006 and 2024 at the National Defense Medical College Hospital were included. Patients without clinical information or surgical tissue samples, and those with a prior history of chemotherapy, were excluded. Clinical data were collected from the medical records. Pathological reviews were conducted in accordance with the 2020 World Health Organization criteria [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. Staging was re-evaluated using the 2014 International Federation of Gynecology and Obstetrics criteria [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. The characteristics of includes cases were demonstrated in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. In total, 58 cases with HRD and 19 cases with homologous recombination proficiency (HRP) were included in the analysis. Among cases with HRD, 28 cases had somatic mutations in breast cancer susceptibility genes. No statistically significant differences were observed in the clinical characteristics between the two groups.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe characteristics of 77 cases with ovarian high-grade serous carcinoma according to the homologous recombination status.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eGroup with Homologous recombination deficiency\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003eGroup with Homologous recombination proficiency\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariables\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;19\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cem\u003ep\u003c/em\u003e-value\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge (years)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAverage\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e60.93\u0026thinsp;\u0026plusmn;\u0026thinsp;11.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003e60.79\u0026thinsp;\u0026plusmn;\u0026thinsp;12.41\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.99\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026ge;60\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(55.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(52.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.99\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026lt;60\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e26\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(44.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(47.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInternational Federation of Obstetrics and Gynecology stage (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.79\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(3.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(0.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eII\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(3.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(5.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIII\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(77.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(73.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIV\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(15.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(21.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eResidual tumor (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.16\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(65.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(84.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e20\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(34.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(15.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePeritoneal cytology (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.54\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePositive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e44\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(75.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(84.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNegative\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(24.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(15.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSomatic breast cancer susceptibility mutation (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePositive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(48.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(0.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNegative\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(51.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e19\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(100.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eAll hematoxylin and eosin (HE)-stained slides were reviewed pathologically. One slide per case containing sufficient tumor tissue was selected and digitized using a NanoZoomer SQ (Hamamatsu Photonics K.K., Shizuoka, Japan). Digital whole-slide images were uploaded to HALO software (Indica Labs, Corrales, NM, USA).\u003c/p\u003e \u003cp\u003eTo annotate the tumor area, we used DenseNet V2, an AI tissue classifier standard used in HALO, without data augmentation. The accuracy of the classifier was evaluated using the F1 scores. The F1 score indicated an agreement between automatic AI detection and manual annotation in the tumor area and defined as 2\u0026times;(Precision\u0026times;Recall)/(Precision+Recall). Precision is the proportion of true positive predictions (TP) among all predicted positive instances, calculated as TP/(TP\u0026thinsp;+\u0026thinsp;FP). Recall is the proportion of true positive predictions (TP) among all actual positive instances, calculated as TP/(TP\u0026thinsp;+\u0026thinsp;FN). In these formulas, TP refers to true positives, FP refers to false positives, and FN refers to false negatives [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. The classifier was run on the entire slide for all cases, and the tumor area automatically detected by the AI was then annotated and used for the analysis.\u003c/p\u003e \u003cp\u003eThe methods used to compare morphological features according to HR status using AI were as follows:\u003c/p\u003e \u003cp\u003eNuclei were detected within the tumor area using the Nuclei Seg module without data augmentation. The Nuclei Seg module is a standard AI segmentation model equipped with HALO. The classifier was run on the annotated tumor area, and nuclei size\u0026thinsp;\u0026ge;\u0026thinsp;20 \u0026micro;m\u003csup\u003e2\u003c/sup\u003e were identified. The area of each nucleus and the total number of nuclei were automatically calculated using this classifier. Then, average nucleus size (\u0026micro;m\u003csup\u003e2\u003c/sup\u003e) and average nucleus count per area (count/mm\u003csup\u003e2\u003c/sup\u003e) in each case were calculated. The optimum cutoff values for average nucleus size and average nucleus count per area were determined using the receiver operating characteristic (ROC) curve for the diagnosis of HRD. The sensitivity and specificity for the diagnosis of HRD were evaluated using the combined cutoff values of the average nucleus size and nucleus count per area. Sensitivity was defined as the proportion of correctly identified true-positive cases, and specificity was defined as the proportion of true-negative cases. Finally, the comparisons between the two groups were conducted.\u003c/p\u003e \u003cp\u003eTumor areas were evaluated according to the HR status by AI as follows. Tumor area from cases of HRD and HRP were manually selected to define the areas for training. AI learned how to distinguish the HRD and HRP areas by training the selected areas using DenseNet V2. AI could automatically assign color-coded labels to the tumor areas based on this training. Once trained, AI automatically evaluated and classified the HRD and HRP areas in the tumor areas of remaining cases that were not used for training. The percentage of HRD areas within the tumor areas was automatically calculated for each case. Then, the optimum cutoff values for the ratio of HRD area to tumor area was determined using the ROC curve for the diagnosis of HRD, and cases were divided into two groups based on this cutoff value for further comparison.\u003c/p\u003e \u003cp\u003eLymphocytes were detected in the tumor area in the third step using Nuclei Seg. Trained lymphocytes were manually annotated, and the classifier was trained. The accuracy of segmentation was evaluated using the aggregated Jaccard index (AJI) scores. The AJI quantifies instance-level segmentation performance by aggregating intersection-over-union values across matched objects and penalizing unmatched predictions, indicating the overall agreement between AI-based and manually annotated lymphocyte segmentations [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. In addition, the size of assessed lymphocytes was set up \u0026le;\u0026thinsp;20 \u0026micro;m\u003csup\u003e2\u003c/sup\u003e using Multiplex IHC. Multiplex IHC enables the simultaneous analysis of multiple markers by adjusting the color hues and size filters to accurately differentiate and quantify cells within a tissue sample. The classifier was run on the annotated tumor area, and lymphocytes were automatically detected. Then, average lymphocyte count per area (count/mm\u003csup\u003e2\u003c/sup\u003e) in each case were calculated, and comparison between the two groups was performed.\u003c/p\u003e \u003cp\u003eStatistical analyses were performed using the JMP Pro 17 software (SAS Institute Inc., Cary, NC, USA). The chi-square test, Fisher\u0026rsquo;s exact test, Mann-Whitney U test, and Wilcoxon test were used to evaluate the statistical significance of the clinical factors. The level of statistical significance was set at \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eThe trained area was manually annotated as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e(a). Red indicated the tumor area, yellow indicated annotation boundaries, blue indicated background areas, such as the stroma or necrotic tissue, and purple indicated no tissue area. A total of 2.215 mm\u003csup\u003e2\u003c/sup\u003e for tumor area from six cases was selected, and the classifier was trained for 8,190 iterations. The F1 score was evaluated for the three cases that were not used for training, and the F1 score value was 0.83. The classifier was run on the entire slide for all cases (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e(b)), and the tumor areas were automatically annotated as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e(c). The average tumor area size was 179.51 (7.46-368.59) mm\u003csup\u003e2\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eNuclei were detected and calculated as demonstrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e(d). The average number of nuclei was 867,055 (33,251-1,044,690) count. The ROC curves of the average nucleus size and average nucleus count per area for the diagnosis of HRD were shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e(a)(b). The areas under the curve were 0.704 and 0.668, and the cutoff values were 52.4 \u0026micro;m\u003csup\u003e2\u003c/sup\u003e and 5,610 count/mm\u003csup\u003e2\u003c/sup\u003e, respectively.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe HRD and HRP areas were classified according to the workflow shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. A total of 3.069 mm\u0026sup2; of tumor area from 3 cases of HRD and 2.744 mm\u0026sup2; from 3 cases of HRP were manually selected. AI trained the selected areas for 5,015 iterations and could automatically assign color-coded labels to the tumor areas based on this training. As results, AI assigned green to HRD areas and purple to HRP areas. Once trained, AI automatically evaluated and classified the HRD and HRP areas in the tumor areas of remaining cases that were not used for training. Finally, the percentage of HRD areas within the tumor areas was automatically calculated for each case. The ROC curves of the ratio of the HRD area to the tumor area for the diagnosis of HRD was shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e(c). The area under the curve was 0.470, and the cutoff value was 40.0%.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eTo train lymphocytes, a total of 2,117 lymphocytes from six cases were manually selected. The classifier was trained for 8,005 iterations. The accuracy of segmentation was evaluated using the AJI scores of 951 lymphocytes from three cases that were not used for training, and the AJI score was 0.56. Lymphocytes were detected and calculated as demonstrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e(e). The average number of lymphocytes was 59,054 (8,328\u0026thinsp;\u0026minus;\u0026thinsp;320,967) count.\u003c/p\u003e \u003cp\u003eThe results of the comparison between the HRD and HRP groups regarding the characteristics and morphological features assessed using AI are shown in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. The HRD group had a smaller average nucleus size and larger average nucleus count per area than the HRP group (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.01 and \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.03, respectively). In addition, cases of average nucleus size less than 52.4 \u0026micro;m\u003csup\u003e2\u003c/sup\u003e and of average nucleus count per area with 5,160 count/mm\u003csup\u003e2\u003c/sup\u003e or more were seen more frequently in the Group with HRD than the Group with HRP (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.01 and \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.01). Furthermore, the sensitivity and specificity for diagnosing HRD using the combined cutoff values for the average nucleus size and average nucleus count were 43.1% and 100.0%, respectively.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe morphological features assessed by artificial intelligence of 77 cases with ovarian high-grade serous carcinoma according to the homologous recombination status.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eGroup with Homologous recombination deficiency\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003eGroup with Homologous recombination proficiency\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariables\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;19\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cem\u003ep\u003c/em\u003e-value\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAverage nucleus size (\u0026micro;m\u003csup\u003e2\u003c/sup\u003e) (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAverage\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e50.64\u0026thinsp;\u0026plusmn;\u0026thinsp;6.87\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003e55.39\u0026thinsp;\u0026plusmn;\u0026thinsp;5.94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026lt;52.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e36\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(62.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(21.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026ge;52.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e22\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(37.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(79.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAverage nucleus counts per area (count/mm\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAverage\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e5610.96\u0026thinsp;\u0026plusmn;\u0026thinsp;1597.21\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003e4751.04\u0026thinsp;\u0026plusmn;\u0026thinsp;1282.22\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.03\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026ge;5160.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e34\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(58.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(21.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026lt;5160.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e24\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(41.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(79.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAverage nucleus size\u0026thinsp;\u0026lt;\u0026thinsp;52.4 \u0026micro;m\u003csup\u003e2\u003c/sup\u003e and average nucleus counts per area\u0026thinsp;\u0026ge;\u0026thinsp;5160.0 count/mm\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(43.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(0.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e33\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(56.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e19\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(100.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eThe ratio of the homologous recombination deficiency area among tumor area (%) \u003csup\u003ea)\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.20\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026ge;40.0%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e17\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(30.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(12.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026lt;40.0%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(69.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(87.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAverage lymphocyte count per area (count/mm\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAverage\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e401.13\u0026thinsp;\u0026plusmn;\u0026thinsp;278.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003e287.10\u0026thinsp;\u0026plusmn;\u0026thinsp;214.33\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.10\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"6\"\u003ea) Three cases of homologous recombination deficiency and three cases of homologous recombination proficiency were excluded from analysis.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn our study, we demonstrated that the HRD group had a smaller average nucleus size and a larger nucleus count per unit area than the HRP group, with high F1 and AJI scores. In addition, the sensitivity and specificity for the diagnosis of HRD using the cutoff values of the average nucleus size and average density of nuclei per unit area were 43.1% and 100.0%, respectively.\u003c/p\u003e \u003cp\u003eThe F1 and AJI scores measured the agreement between the automatic detection by AI and manual annotation. Both values ranged from 0 to 1.0, with a value closer to 1.0, indicating higher precision [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. The F1 score was calculated as the harmonic mean of precision (P) and recall (R) using the formula [2PR/(P\u0026thinsp;+\u0026thinsp;R)] [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. In previous studies involving AI-based pathological evaluations, F1 scores ranged from 0.78 to 0.97 [\u003cspan additionalcitationids=\"CR26\" citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. The F1 score in our study was within this range, suggesting that the accuracy of our method is comparable to that of previous reports. Hence, the AJI score considers pixel-level (segmentation) errors, in contrast to the F1 score, and can treat unified detection and segmentation [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. A study that compared the quality of nuclear segmentation among four open-source software programs showed that the overall AJI score ranged from 0.12 to 0.51 [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. Our study preserved the accuracy of AJI scores.\u003c/p\u003e \u003cp\u003eFirst, the AI detected the nuclei using a built-in function. Additionally, among these nuclei, nuclei size\u0026thinsp;\u0026ge;\u0026thinsp;20 \u0026micro;m\u003csup\u003e2\u003c/sup\u003e were defined as the nuclei. The average nucleus size was smaller, and the average nucleus count per area, which was equal to the average cell count per area, was larger in the HRD group than in the HRP group. Cellular responses to DSBs damage include mechanisms that restore the structure of broken chromosomes, and HR is associated with the maintenance of chromosomal structure [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. Some nuclear chromosomal fragments caused by DNA DSBs flow out of the nucleus and remain in the cytoplasm [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. Therefore, we assumed that the nuclei of cancer cells with HRD were smaller than those with HRP, and that the density of the nuclei of cancer cells with HRD was larger than that of HRP. DNA repair mechanisms, such as the HR status, may influence the size of nuclei or cells and the appearance of morphological structures.\u003c/p\u003e \u003cp\u003eSecond, we trained the AI model using manually annotated cancer cell structures based on the built-in classifier. Subsequently, the AI utilized this learned information to classify the structures in the remaining cases. A previous study reported that tumors with HRD may exhibit distinct structures, such as solid, pseudo-endometrioid, and transitional-like morphologies, compared with HRP tumors [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. Based on this, we trained the AI to learn the randomized tumor area rather than each tumor cell, as the structures constructed by tumor cells using 3 cases with HRD or HRP, and the AI was used to evaluate the remaining cases. However, the AI determined the area of the HRD or HRP in each case. Therefore, this training method for memorizing the tumor area was not effective in determining the HR status.\u003c/p\u003e \u003cp\u003eThe lymphocytes were identified using a built-in AI segmentation module. Of these, cells with an area less than or equal to 20 \u0026micro;m\u0026sup2; were defined as lymphocytes. There were no differences in lymphocyte counts between HGSC with HRD and those with HRP, which is inconsistent with previous literature [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. Previous studies have demonstrated a relationship between HR status and lymphocytes with a cluster of differentiation 3 or cluster of differentiation 8. The present study did not classify the lymphocyte clusters because they were assessed using hematoxylin and eosin staining. Further studies are needed to categorize clusters of lymphocytes to explore the relationship between lymphocytes and HR status.\u003c/p\u003e \u003cp\u003eThe review of literature on the morphological comparison based on AI according to the HR status including our study was summarized in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. Previous studies reported the potential of deep learning models for predicting HR status through the analysis of tumor area from histopathological images [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Particularly, Zhang et al. achieved a precision of 0.800 and recall of 0.727 [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. However, in our study, we did not reach the same level of the analysis for tumor area. This discrepancy might develop from the differences of AI devices or the limited number of cases included in our study. On contrast, our study also examined nuclear characteristics and the presence of lymphocytes. We demonstrated the significant differences about the nuclear size and density between the HRD and HRP groups and 100.0% of the specificity for the diagnosis of HRD using the combined cutoff values for the average nucleus size and average nucleus density. These findings were the novel findings of our study and might suggest that nuclear morphology might be a potential histopathological biomarker for HRD detection in the future.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eReview of literature on the morphological comparison based on artificial Intelligence according to the homologous recombination status including the present study.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMain findings\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAuthor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSample size\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eInput data\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAnalysis of nuclei\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAnalysis of tumor regions\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAnalysis of lymphocytes\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eZhang K et al. [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e205 cases\u003c/p\u003e \u003cp\u003eHRD: 64 cases\u003c/p\u003e \u003cp\u003eHRP: 141 cases\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eH\u0026amp;E-stained whole-slide images\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNot conducted\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePrecision 0.800, Recall 0.727\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNot conducted\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFrenel J-S et al. [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e93 cases\u003c/p\u003e \u003cp\u003e(HRD subtypes was not specified)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eH\u0026amp;E-stained whole-slide images\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNot conducted\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eDetailed metrics not available in abstract.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNot conducted\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eThe present study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e77 cases\u003c/p\u003e \u003cp\u003eHRD: 58 cases\u003c/p\u003e \u003cp\u003eHRP: 19 cases\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eH\u0026amp;E-stained whole-slide images\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eHRD group had a smaller average nucleus size and larger average nucleus density\u003c/p\u003e \u003cp\u003e(\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.01, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.03).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eThere was no statistical difference (Precision 0.895, Recall 0.309, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.20).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eThere was no statistical difference (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.10).\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"6\"\u003eAbbreviations\u003c/td\u003e\u003c/tr\u003e \u003ctr\u003e\u003ctd colspan=\"6\"\u003eHRD: Homologous recombination deficiency, HRP: Homologous recombination proficiency, HE: Hematoxylin and eosin\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThis study has several limitations, including the small sample size from a single institution and its retrospective design. Further large-scale prospective studies are warranted to validate the clinical and pathological significance of HGSC stratified according to HR status.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eOur study demonstrates that AI may identify the morphological features of HGSC with HRD and may be useful for identifying several other small changes among tumors with different genetic backgrounds.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eEOC: epithelial ovarian carcinoma\u003c/p\u003e\n\u003cp\u003eHGSC: high-grade serous carcinoma\u003c/p\u003e\n\u003cp\u003eHR: homologous recombination\u003c/p\u003e\n\u003cp\u003eDNA: deoxyribonucleic acid\u003c/p\u003e\n\u003cp\u003eDSBs: double-strand breaks\u003c/p\u003e\n\u003cp\u003eBRCA: breast cancer susceptibility gene\u003c/p\u003e\n\u003cp\u003eHRD: homologous recombination deficiency\u003c/p\u003e\n\u003cp\u003eAI: artificial intelligence\u003c/p\u003e\n\u003cp\u003eHRP: homologous recombination proficiency\u003c/p\u003e\n\u003cp\u003eROC: receiver operator characteristic\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that no funds, grants, or other support were received during the preparation of this manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors have no relevant financial or non-financial interests to disclose.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026rsquo; contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eProtocol/project development: TH, MM, TE, KS, and MT.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eData collection and management: TH, MM, SK, KO, and YH.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eData analysis: TH, TE, MK, TW, YO, JS, TI, NK, RT, SN, KK, and HS.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eManuscript writing/editing: TH, MM, and MT.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAll authors read and approved the final manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was approved by the Institutional Review Board of National Defense Medical College, Tokorozawa, Japan, and was conducted in accordance with the Declaration of Helsinki.\u0026nbsp;The records and information of all women in the study were anonymized and de-identified prior to analysis. The study was\u0026nbsp;exempt from obtaining informed consent from all participants.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eClinical trial number\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets used and/or analyzed in the current study are available from the corresponding author upon reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors would like to thank Editage (www.editage.jp) for the English language editing. We also appreciate the Defense Medicine Basic Research Program B and Indica Labs Inc.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eSiegel RL, Miller KD, Fuchs HE, Jemal A (2021) Cancer statistics, 2021. CA Cancer J Clin 71:7\u0026ndash;33. https://doi.org/10.3322/caac.21654\u003c/li\u003e\n\u003cli\u003ePrat J (2012) Ovarian carcinomas: five distinct diseases with different origins, genetic alterations, and clinicopathological features. Virchows Arch 460:237\u0026ndash;249. https://doi.org/10.1007/s00428-012-1203-5\u003c/li\u003e\n\u003cli\u003eVergote I, Trop\u0026eacute; CG, Amant F, Kristensen GB, Ehlen T, Johnson N, et al (2010) Neoadjuvant chemotherapy or primary surgery in stage IIIC or IV ovarian cancer. N Engl J Med 363:943\u0026ndash;953. https://doi.org/10.1056/NEJMoa0908806\u003c/li\u003e\n\u003cli\u003eKehoe S, Hook J, Nankivell M, Jayson GC, Kitchener H, Lopes T, et al (2015) Primary chemotherapy versus primary surgery for newly diagnosed advanced ovarian cancer (CHORUS): an open-label, randomised, controlled, non-inferiority trial. Lancet 386:249\u0026ndash;257. https://doi.org/10.1016/S0140-6736(14)62223-6\u003c/li\u003e\n\u003cli\u003eArmstrong DK, Alvarez RD, Bakkum-Gamez JN, Barroilhet L, Behbakht K, Berchuck A, et al (2021) Ovarian Cancer, Version 2.2020, NCCN Clinical Practice Guidelines in Oncology. J Natl Compr Canc Netw 19:191\u0026ndash;226. https://doi.org/10.6004/jnccn.2021.0007\u003c/li\u003e\n\u003cli\u003eInternational Collaborative Ovarian Neoplasm Group (2002) Paclitaxel plus carboplatin versus standard chemotherapy with either single-agent carboplatin or cyclophosphamide, doxorubicin, and cisplatin in women with ovarian cancer: the ICON3 randomised trial. Lancet 360:505\u0026ndash;515. https://doi.org/10.1016/S0140-6736(02)09738-6\u003c/li\u003e\n\u003cli\u003eWright WD, Shah SS, Heyer WD (2018) Homologous recombination and the repair of DNA double-strand breaks. J Biol Chem 293:10524\u0026ndash;10535. https://doi.org/10.1074/jbc.TM118.000372\u003c/li\u003e\n\u003cli\u003eHaber JE (1999) DNA recombination: the replication connection. Trends Biochem Sci 24:271\u0026ndash;275. https://doi.org/10.1016/s0968-0004(99)01413-9\u003c/li\u003e\n\u003cli\u003eDoig KD, Fellowes AP, Fox SB (2023) Homologous recombination repair deficiency: an overview for pathologists. Mod Pathol 36:100049. https://doi.org/10.1016/j.modpat.2022.100049\u003c/li\u003e\n\u003cli\u003eHeeke AL, Pishvaian MJ, Lynce F, Xiu J, Brody JR, Chen WJ, et al (2018) Prevalence of homologous recombination-related gene mutations across multiple cancer types. JCO Precis Oncol 2018:PO.17.00286. https://doi.org/10.1200/PO.17.00286\u003c/li\u003e\n\u003cli\u003eMorse CB, Toukatly MN, Kilgore MR, Agnew KJ, Bernards SS, Norquist BM, et al (2019) Tumor infiltrating lymphocytes and homologous recombination deficiency are independently associated with improved survival in ovarian carcinoma. Gynecol Oncol 153:217-222. http://doi.org/10.1016/j.ygyno.2019.02.011\u003c/li\u003e\n\u003cli\u003eStrickland KC, Howitt BE, Shukla SA, Rodig S, Ritterhouse LL, Liu JF, et al (2016) Association and prognostic significance of BRCA1/2-mutation status with neoantigen load, number of tumor-infiltrating lymphocytes and expression of PD-1/PD-L1 in high grade serous ovarian cancer. Oncotarget 7:13587-98. http://doi.org/10.18632/oncotarget.7277\u003c/li\u003e\n\u003cli\u003eLoeb LA (2011) Human cancers express mutator phenotypes: origin, consequences and targeting. Nat Rev Cancer 11:450\u0026ndash;457. https://doi.org/10.1038/nrc3063\u003c/li\u003e\n\u003cli\u003eRay-Coquard I, Leary A, Pignata S, Cropet C, Gonz\u0026aacute;lez-Mart\u0026iacute;n A, Marth C, et al (2023) Olaparib plus bevacizumab first-line maintenance in ovarian cancer: final overall survival results from the PAOLA-1/ENGOT-ov25 trial. Ann Oncol 34:681\u0026ndash;692. https://doi.org/10.1016/j.annonc.2023.05.005\u003c/li\u003e\n\u003cli\u003eMonk BJ, Barretina-Ginesta MP, Pothuri B, Vergote I, Graybill W, Mirza MR, et al (2024) Niraparib first-line maintenance therapy in patients with newly diagnosed advanced ovarian cancer: final overall survival results from the PRIMA/ENGOT-OV26/GOG-3012 trial. Ann Oncol 35:981\u0026ndash;992. https://doi.org/10.1016/j.annonc.2024.08.2241\u003c/li\u003e\n\u003cli\u003eMonk BJ, Parkinson C, Lim MC, O\u0026rsquo;Malley DM, Oaknin A, Wilson MK, et al (2022) A randomized, Phase III trial to evaluate Rucaparib monotherapy as maintenance treatment in patients with newly diagnosed ovarian cancer (ATHENA-MONO/GOG-3020/ENGOT-ov45). J Clin Oncol 40:3952\u0026ndash;3964. https://doi.org/10.1200/JCO.22.01003\u003c/li\u003e\n\u003cli\u003eKouzu K, Tsujimoto H, Imamura Y, Kajiwara Y, Nearchou IP, Einama T, et al (2024) Development and validation study of the prognostic impact of deep learning-determined myxoid stroma after neoadjuvant chemotherapy in patients with esophageal squamous cell carcinoma. Ann Surg Oncol 31:6300\u0026ndash;6308. https://doi.org/10.1245/s10434-024-15626-w\u003c/li\u003e\n\u003cli\u003eKouzu K, Tsujimoto H, Nearchou IP, Einama T, Watanabe T, Horiguchi H, et al (2025) Prognostic impact of tumor cell nuclear size assessed by artificial intelligence in esophageal squamous cell carcinoma. Lab Invest 105:102221. https://doi.org/10.1016/j.labinv.2024.102221\u003c/li\u003e\n\u003cli\u003eZhang K, Qiu Y, Feng S, Yin H, Liu Q, Zhu Y, et al (2025) Development of model for identifying homologous recombination deficiency (HRD) status of ovarian cancer with deep learning on whole slide images. J Transl Med 23:267. http://doi.org/10.1186/s12967-025-06234-7\u003c/li\u003e\n\u003cli\u003eFrenel J-S, Bossard C, Rynkiewicz J, Thomas F, Salhi Y, Salhi S, et al (2024) Artificial intelligence to predict homologous recombination deficiency in ovarian cancer from whole-slide histopathological images. J Clin Oncol 42(16_suppl):5578. https://doi.org/10.1200/JCO.2024.42.16_suppl.5578\u003c/li\u003e\n\u003cli\u003eLokuhetty D, White VA, Cree IA, editors (2020) WHO classification of tumours of female reproductive organs. 5th ed. Lyon: IARC\u003c/li\u003e\n\u003cli\u003ePereira A, P\u0026eacute;rez-Medina T, Magrina JF, Magtibay PM, Rodr\u0026iacute;guez-Tapia A, Peregrin I, et al (2015) International Federation of Gynecology and Obstetrics staging classification for cancer of the ovary, fallopian tube, and peritoneum: estimation of survival in patients with node-positive epithelial ovarian cancer. Int J Gynecol Cancer 25:49\u0026ndash;54. https://doi.org/10.1097/IGC.0000000000000316\u003c/li\u003e\n\u003cli\u003eGhosh AK, Thammasudjarit R, Jongkhajornpong P, Attia J, Thakkinstian A (2022) Deep learning for discrimination between fungal keratitis and bacterial keratitis: DeepKeratitis. Cornea 41:616\u0026ndash;622. https://doi.org/10.1097/ICO.0000000000002830\u003c/li\u003e\n\u003cli\u003eKumar N, Verma R, Sharma S, Bhargava S, Vahadane A, Sethi A (2017) A dataset and a technique for generalized nuclear segmentation for computational pathology. IEEE Trans Med Imaging 36:1550\u0026ndash;1560. https://doi.org/10.1109/TMI.2017.2677499\u003c/li\u003e\n\u003cli\u003eSaha M, Guo X, Sharma A (2021) \u003cem\u003eTilGAN\u003c/em\u003e: GAN for facilitating tumor-infiltrating lymphocyte pathology image synthesis with improved image classification. IEEE Access 9:79829\u0026ndash;79840. https://doi.org/10.1109/access.2021.3084597\u003c/li\u003e\n\u003cli\u003eBarua B, Chyrmang G, Bora K, Ahmed GN, Kakoti L, Saikia MJ (2025) Classifying tumour infiltrating lymphocytes in oral squamous cell carcinoma histopathology using joint learning framework. Sci Rep 15:2938. https://doi.org/10.1038/s41598-025-86527-5\u003c/li\u003e\n\u003cli\u003eSwiderska-Chadaj Z, Pinckaers H, van Rijthoven M, Balkenhol M, Melnikova M, Geessink O, et al (2019) Learning to detect lymphocytes in immunohistochemistry with deep learning. Med Image Anal. 58:101547. https://doi.org/10.1016/j.media.2019.101547\u003c/li\u003e\n\u003cli\u003eYou Z, Jiang M, Shi Z, Ning X, Shi C, Du S, et al (2021) Evaluation of automated segmentation algorithms for neurons in macaque cerebral microscopic images. Microsc Res Tech 84:2311\u0026ndash;24. https://doi.org/10.1002/jemt.23786 [EPub]\u003c/li\u003e\n\u003cli\u003eGospodinov A, Herceg Z (2013) Chromatin structure in double strand break repair. DNA Repair (Amst) 12:800\u0026ndash;10. https://doi.org/10.1016/j.dnarep.2013.07.006\u003c/li\u003e\n\u003cli\u003eVerma P, Greenberg RA (2021) Communication between chromatin and homologous recombination. Curr Opin Genet Dev 71:1\u0026ndash;9. https://doi.org/10.1016/j.gde.2021.05.006\u003c/li\u003e\n\u003cli\u003eHopfner KP, Hornung V (2020) Molecular mechanisms and cellular functions of cGAS-STING signalling. Nat Rev Mol Cell Biol 21:501\u0026ndash;521. https://doi.org/10.1038/s41580-020-0244-x\u003c/li\u003e\n\u003cli\u003eChapagain U, Huecker JB, Sun L (2025) Morphologic correlations with homologous recombination deficiency in high-grade serous carcinomas. Int J Gynecol Pathol 44:398-406. http://doi.org/10.1097/PGP.0000000000001090\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"archives-of-gynecology-and-obstetrics","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"arch","sideBox":"Learn more about [Archives of Gynecology and Obstetrics](https://www.springer.com/journal/404)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/arch/default.aspx","title":"Archives of Gynecology and Obstetrics","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Ovarian high-grade serous carcinoma, Artificial intelligence, Homologous recombination status, Nucleus size","lastPublishedDoi":"10.21203/rs.3.rs-8830192/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8830192/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003ePurpose\u003c/h2\u003e \u003cp\u003eThis study aimed to identify the morphological features of ovarian high-grade serous carcinoma (HGSC) based on homologous recombination (HR) using artificial intelligence (AI).\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eSeventy-seven patients with HGSC who underwent HR status testing and surgery between 2006 and 2024 were included. One hematoxylin and eosin-stained slide per case, containing a sufficient volume of tumor tissue, was digitized. Tumor areas were automatically detected and annotated using AI. Nuclei in the tumor area were detected using AI. The area of each nucleus and the total number of nuclei were calculated automatically. A trained classifier determined the ratio of the tumor area to HR deficiency (HRD). Receiver operator characteristic curve established optimum cutoff value for average nucleus size (\u0026micro;m\u003csup\u003e2\u003c/sup\u003e), average nucleus count per area (count/mm\u003csup\u003e2\u003c/sup\u003e), and HRD area ratio (%).\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eThe area under the curve of average nucleus size, average nucleus count per tumor area, and HRD area ratio to the tumor area for the diagnosis of HRD were 0.704, 0.668, and 0.470, respectively, with cut-offs of 52.4 \u0026micro;m\u003csup\u003e2\u003c/sup\u003e, 5,610 count/mm\u003csup\u003e2\u003c/sup\u003e, and 40.0%, respectively. The HRD group had a smaller average nucleus size and larger average nucleus count per area than the HR proficiency group (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.01 and \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.03, respectively). The sensitivity and specificity for diagnosing HRD using the combined cutoff values for the average nucleus size and average nucleus count were 43.1% and 100.0%, respectively.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eAI can identify the morphological features of HGSC with HRD and detect subtle tumor differences related to different genetic backgrounds.\u003c/p\u003e","manuscriptTitle":"Identification of histological features of ovarian high-grade serous carcinoma with homologous recombination deficiency using artificial intelligence: A retrospective analysis","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-08 17:26:57","doi":"10.21203/rs.3.rs-8830192/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-04-13T21:29:34+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-30T12:52:36+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"297697346617739523880812345754293089574","date":"2026-03-20T07:48:38+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-03-02T08:55:06+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-02-10T13:45:25+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-02-10T12:07:41+00:00","index":"","fulltext":""},{"type":"submitted","content":"Archives of Gynecology and Obstetrics","date":"2026-02-09T11:42:34+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"archives-of-gynecology-and-obstetrics","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"arch","sideBox":"Learn more about [Archives of Gynecology and Obstetrics](https://www.springer.com/journal/404)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/arch/default.aspx","title":"Archives of Gynecology and Obstetrics","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"23f183f0-c09c-4075-829d-f5c7a9c75d3a","owner":[],"postedDate":"March 8th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-04-27T08:23:26+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-08 17:26:57","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8830192","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8830192","identity":"rs-8830192","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}