Limits Of Agreement Research Articles

Proteomic analysis of serum in a population-based cohort did not reveal a biomarker for Modic changes.

Modic changes (MC) are bone marrow lesions of vertebral bones, which can be detected with magnetic resonance imaging (MRI) adjacent to degenerated intervertebral discs. Defined by their appearance on T1 and T2 weighted images, there are three interconvertible types: MC1, MC2, and MC3. The inter-observer variability of the MRI diagnosis is high, therefore a diagnostic serum biomarker complementing the MRI to facilitate diagnosis and follow-up would be of great value. We used a highly sensitive and reproducible proteomics approach: DIA/SWATH-MS to find serum biomarkers in a subset of the Northern Finland Birth Cohort 1966. Separately, we measured a panel of factors involved in inflammation and angiogenesis to confirm some potential biomarkers published before with an ELISA-based method called V-Plex. We found neither an association between the serum concentrations of the proteins detected with DIA/SWATH-MS with the presence of MC, nor a correlation with the size of the MC lesions. We did not find any association between the factors measured with the V-Plex and the presence of MC or their size. Altogether, our study suggests that a robust and generally usable biomarker to facilitate the diagnosis of MC cannot readily be found in serum.

A comparison of target volumes drawn on arterial and venous phase scans during radiation therapy planning for patients with pancreatic cancer: the PANCRINJ study

BackgroundThe planification of radiation therapy (RT) for pancreatic cancer (PC) requires a dosimetric computed tomography (CT) scan to define the gross tumor volume (GTV). The main objective of this study was to compare the inter-observer variability in RT planning between the arterial and the venous phases following intravenous contrast.MethodsPANCRINJ was a prospective monocentric study that included twenty patients with non-metastatic PC. Patients underwent a pre-therapeutic CT scan at the arterial and venous phases. The delineation of the GTV was performed by one radiologist (gold standard) and two senior radiation oncologists (operators). The primary objective was to compare the Jaccard conformity index (JCI) for the GTVs computed between the GS (gold standard) and the operators between the arterial and the venous phases with a Wilcoxon signed rank test for paired samples. The secondary endpoints were the geographical miss index (GMI), the kappa index, the intra-operator variability, and the dose-volume histograms between the arterial and venous phases.ResultsThe median JCI for the arterial and venous phases were 0.50 (range, 0.17–0.64) and 0.41 (range, 0.23–0.61) (p = 0.10) respectively. The median GS-GTV was statistically significantly smaller compared to the operators at the arterial (p < 0.0001) and venous phases (p < 0.001), respectively. The GMI were low with few tumors missed for all patients with a median GMI of 0.07 (range, 0-0.79) and 0.05 (range, 0-0.39) at the arterial and venous phases, respectively (p = 0.15). There was a moderate agreement between the radiation oncologists with a median kappa index of 0.52 (range 0.38–0.57) on the arterial phase, and 0.52 (range 0.36–0.57) on the venous phase (p = 0.08). The intra-observer variability for GTV delineation was lower at the venous phase than at the arterial phase for the two operators. There was no significant difference between the arterial and the venous phases regarding the dose-volume histogram for the operators.ConclusionsOur results showed inter- and intra-observer variability in delineating GTV for PC without significant differences between the arterial and the venous phases. The use of both phases should be encouraged. Our findings suggest the need to provide training for radiation oncologists in pancreatic imaging and to collaborate within a multidisciplinary team.

Two-tier architecture-based grading for pancreatic cancer outperforms traditional grading.

A novel architecture-based grading system for pancreatic ductal adenocarcinoma (PDAC) is tested against traditional grading. A total of 103 PDAC resections were graded by College of American Pathologists/American Joint Committee on Cancer (CAP/AJCC) guidelines and by a system using an architectural pattern (dispersed larger duct = low grade vs dense smaller duct = high grade). Survival analyses and interobserver variability were assessed. In total, 114 cases from a public data set were used for validation. Median overall survivals were 15 and 36 months for architectural high-grade and low-grade cases, respectively (P<.001). Conversely, CAP/AJCC grading showed no survival difference between well-differentiated and moderately differentiated tumors (P=.545). Architecture-based grading remained prognostically significant for recurrence-free survival (P=.004), but CAP/AJCC grading was not (P=.226). Adjusted for stage and margin status, architectural high-grade PDACs showed a hazard ratio of 2.69 relative to low grade (P<.001) for survival. The validation cohort confirmed prognostic differences in overall (P<.001) and recurrence-free survival (P=.027) for the architecture-based system, outperforming CAP/AJCC grading. Architecture-based grading exhibited a Cohen's ĸ value of 0.710 (substantial agreement), superior to traditional grading (0.488, moderate agreement). Grading PDAC based on architectural pattern results in superior prognostication and reproducibility vs CAP/AJCC grading.

Esophagogastroscopy for predicting endoscopic ultrasonography T-stage by utilizing deep learning methods in esophageal cancer

Endoscopic ultrasonography (EUS) is commonly utilized in preoperative staging of esophageal cancer, however with additional pain and cost as well as adverse events. Meanwhile, the accuracy of EUS is highly depend on the training and practice of operators and not universally available. Different operators would lead to high inter-observer variability. Therefore, it is desirable to explore an alternative way to determine preoperative T stage in esophageal cancer. Whether conventional endoscopy possess the ability to predict EUS T stage has never been investigated yet. In current study, with the assistance of Artificial intelligence, we have developed a deep learning model to predict EUS T stage based on 9,714 images collected from 3,333 patients. ResNet-152 pre-trained on the ImageNet dataset was trained with the appropriate transfer learning and fine-tuning strategies on the conventional endoscopic images and their corresponding labels (e.g., T1, T2, T3, T4 and Normal). Meanwhile, augmentation strategies including rotation and flipping were performed to increase the number of images to improve the prediction accuracy. Finally, 4,382 T1, 243 T2, 3,985 T3, 1,102 T4, 14,302 controls images were obtained and split into training dataset, validation dataset and independent testing dataset with the ratio of 4:1:1. Our model could achieve a satisfied performance with an area under the receiver-operating curve (AUC) were 0.9767, 0.9637, 0.9597 and 0.9442 for T1, T2, T3 and T4, respectively in independent testing dataset. In conclusion, conventional gastroscopy combined with artificial intelligence have the great potential to predict EUS T stage.

Agreement Between the National Cancer Institute's diet history questionnaire II and III in a preconception cohort.

Food frequency questionnaires require updating over time, due to population changes in diet, posing analytical challenges in consistently measuring diet in prospective studies. We compared reliability and agreement between nutrients in two versions of the National Cancer Institute's web-based Diet History Questionnaire (DHQ, III vs. II) in an ongoing North American preconception study. We invited 51 consecutively-enrolled U.S. female participants aged 21-45 years to complete both DHQ versions within a 2-week period, in a randomized order. We compared 30 nutrients from both DHQ versions and calculated within-person reliability using intraclass correlation coefficients (ICCs). Bland-Altman plots and 95% limits of agreement (LOA) were generated to assess nutrient agreement between DHQ versions. We observed highest reliability in percent energy from carbohydrates and cholesterol (ICCs: 0.88; 95% CI: 0.80-0.93) and lowest for percent energy from protein and vitamin D (ICCs: 0.56; 95% CI: 0.34-0.72). At the group level, all nutrients had most observations within the LOA. Bland-Altman plots showed assessment differences between DHQs for protein, fat, monounsaturated fat, and vitamin D. The remaining nutrients showed good agreement and good-to-moderate reliability. Some nutrients may require adjustment and calibration analysis before using them interchangeably across DHQ versions.

Evaluation of the performance of digital x-ray systems in pelvis radiography

The aim of this study was to investigate the performance of eight digital radiography systems and to optimise the dose-image quality relationship for digital pelvis radiography. The study involved eight digital radiography systems used for general examinations at Vilnius University Hospital Santaros Klinikos. An anthropomorphic pelvic phantom (CIRS, US) was used to simulate a patient undergoing clinical pelvis radiography. Dose quantities entrance surface dose, dose area product (DAP) and exposure parameters (kVp, mA, mAs) were measured and the effects on the images were evaluated, considering physical contrast to noise ratio (CNR) and observer-based evaluations as image quality metrics. Increasing the tube voltage by 5 kVp from standard protocol led to a reduction in radiation dose (DAP) by 12%–20% with a slight impact on image quality (CNR decreases by 2%–10%). There was an inter-observer variability in image rating across different equipment (kappa value between 0 and 0.3); however, both observers agreed that increasing kVp up to 85–90 kV had no effect on perceived image quality. The results indicate that optimisation strategies should be tailored specifically for each x-ray system since significant performance differences and wide variations in radiation dose exist across various digital radiography systems used in clinical settings. The use of high kVp can be used for dose optimisation in digital pelvis radiography without compromising image diagnostic accuracy.

BELUGA WHALE LION OPTIMIZATION IN DEEP-NETS FOR HUMAN AGE ESTIMATION USING HAND X-RAY

Bones undergo significant changes in size and shape with the growth of the child, and bone age estimation is crucial for determining the growth, genetic and endocrine disorders in children. Hand X-ray images are extensively utilized for diagnosing disorders in children. The variation in the chronological age and bone age indicates the presence of endocrine disorders, genetic problems, and growth abnormalities. Traditionally, bone age is estimated manually by inspecting the X-ray images, which is extremely time-consuming and prone to error. Further, the accuracy of the bone estimate depends on the experience of the medical practitioner, and thus it suffers from intra- and inter-observer variability. Hence, to overcome these issues, it is essential to devise automatic methods that can estimate the bone with high accuracy and in a short duration. In this work, bone age is estimated using a Deep Residual Network (DRN), whose learnable factors are adjusted using the devised Beluga Whale Lion Optimization (BWLO) algorithm. Further, the BWLO_DRN is examined for its superiority considering metrics, like accuracy, True Positive Rate (TPR), and True Negative Rate (TNR), and the corresponding values of 89.8%, 86.8%, and 90% are found to be achieved from the experimental results.

Concurrent validity of countermovement and squat jump height assessed with a contact mat and force platform in professional soccer players.

The aim of this study was to assess the concurrent validity of a contact mat against force plates to measure jump height in countermovement jump (CMJ) and squat jump (SJ) in professional soccer players. 23 male professional soccer players performed the CMJ and SJ, which were concurrently recorded using a portable contact mat (SmartJump) and a portable dual force plate system (ForceDecks). Equivalence testing between both systems (contact mat vs. force plate) and the two methods (impulse-momentum vs. flight-time and flight-time vs. flight-time) was performed compared to equivalence bounds of ±1.1 cm for the CMJ and ±1.6 cm for the SJ. Additionally, 95% Limits of Agreement (LoA) and intraclass correlation coefficients (ICC) were computed. Mean differences for the impulse-momentum vs. flight-time comparison for CMJ [3.2 cm, 95% CI (2.3-4.1)] and SJ [2.7 cm, (1.8-3.6)] were non-equivalent between both systems. LoA were larger than the equivalence bunds for CMJ and SJ, while ICCs were good [CMJ, 0.89, (0.76-0.95)] and excellent [SJ, 0.91, (0.79-0.96)]. As for the flight-time vs. flight-time comparison, mean differences were non-equivalent for the CMJ [1.0 cm (0.8 to 1.2 cm)] and equivalent for the SJ [0.9 cm (0.7-1.1 cm)]. LoA were narrower than the equivalence bounds for CMJ and SJ, while ICCs were excellent [CMJ, 0.995, 95% CI (0.989-0.998); SJ, 0.997, 95% CI (0.993-0.997)]. Our findings indicate that the SmartJump contact mat cannot be used interchangeably with the ForceDecks force platform to measure jump height for the CMJ and SJ.

Defining tumor growth in vestibular schwannomas: a volumetric inter-observer variability study in contrast-enhanced T1-weighted MRI

PurposeFor patients with vestibular schwannomas (VS), a conservative observational approach is increasingly used. Therefore, the need for accurate and reliable volumetric tumor monitoring is important. Currently, a volumetric cutoff of 20% increase in tumor volume is widely used to define tumor growth in VS. The study investigates the tumor volume dependency on the limits of agreement (LoA) for volumetric measurements of VS by means of an inter-observer study.MethodsThis retrospective study included 100 VS patients who underwent contrast-enhanced T1-weighted MRI. Five observers volumetrically annotated the images. Observer agreement and reliability was measured using the LoA, estimated using the limits of agreement with the mean (LOAM) method, and the intraclass correlation coefficient (ICC).ResultsThe 100 patients had a median average tumor volume of 903 mm3 (IQR: 193–3101). Patients were divided into four volumetric size categories based on tumor volume quartile. The smallest tumor volume quartile showed a LOAM relative to the mean of 26.8% (95% CI: 23.7–33.6), whereas for the largest tumor volume quartile this figure was found to be 7.3% (95% CI: 6.5–9.7) and when excluding peritumoral cysts: 4.8% (95% CI: 4.2–6.2).ConclusionAgreement limits within volumetric annotation of VS are affected by tumor volume, since the LoA improves with increasing tumor volume. As a result, for tumors larger than 200 mm3, growth can reliably be detected at an earlier stage, compared to the currently widely used cutoff of 20%. However, for very small tumors, growth should be assessed with higher agreement limits than previously thought.

A prognostic and predictive computational pathology immune signature for ductal carcinoma in situ: retrospective results from a cohort within the UK/ANZ DCIS trial

The density of tumour-infiltrating lymphocytes (TILs) could be prognostic in ductal carcinoma in situ (DCIS). However, manual TIL quantification is time-consuming and suffers from interobserver and intraobserver variability. In this study, we developed a TIL-based computational pathology biomarker and evaluated its association with the risk of recurrence and benefit of adjuvant treatment in a clinical trial cohort. In this retrospective cohort study, a computational pathology pipeline was developed to generate a TIL-based biomarker (CPath TIL categories). Subsequently, the signature underwent a masked independent validation on H&E-stained whole-section images of 755 patients with DCIS from the UK/ANZ DCIS randomised controlled trial. Specifically, continuous biomarker CPath TIL score was calculated as the average TIL density in the DCIS microenvironment and dichotomised into binary biomarker CPath TIL categories (CPath TIL-high vs CPath TIL-low) using the median value as a cutoff. The primary outcome was ipsilateral breast event (IBE; either recurrence of DCIS [DCIS-IBE] or invasive progression [I-IBE]). The Cox proportional hazards model was used to estimate the hazard ratio (HR). CPath TIL-score was evaluable in 718 (95%) of 755 patients (151 IBEs). Patients with CPath TIL-high DCIS had a greater risk of IBE than those with CPath TIL-low DCIS (HR 2·10 [95% CI 1·39-3·18]; p=0·0004). The risk of I-IBE was greater in patients with CPath TIL-high DCIS than those with CPath TIL-low DCIS (3·09 [1·56-6·14]; p=0·0013), and the risk of DCIS-IBE was non-significantly higher in those with CPath TIL-high DCIS (1·61 [0·95-2·72]; p=0·077). A significant interaction (pinteraction=0·025) between CPath TIL categories and radiotherapy was observed with a greater magnitude of radiotherapy benefit in preventing IBE in CPath TIL-high DCIS (0·32 [0·19-0·54]) than CPath TIL-low DCIS (0·40 [0·20-0·81]). High TIL density is associated with higher recurrence risk-particularly of invasive recurrence-and greater radiotherapy benefit in patients with DCIS. Our TIL-based computational pathology signature has a prognostic and predictive role in DCIS. National Cancer Institute under award number U01CA269181, Cancer Research UK (C569/A12061; C569/A16891), and the Breast Cancer Research Foundation, New York (NY, USA).

Agreement and repeatability of scotopic pupil size measurement with the 2WIN-S portable refractor in Chinese adults

To assess the agreement and repeatability of scotopic pupil size measurement using 2WIN-S (Adaptica, Padova, Italy) portable refractor in Chinese adults. This prospective non-randomized open-label controlled study assessed the scotopic pupil size of 100 right eyes using OPD-Scan III (Optical path difference) (Nidek Technologies, Gamagori, Japan) and 2WIN-S. OPD-Scan III and 2WIN-S measure pupil size using infrared light and detector, while 2WIN-S measures bilateral eyes simultaneously, OPD-Scan III measures unilateral eyes individually. Participants were first measured once using OPD-Scan III and two consecutive measurements were performed using 2WIN-S after 15 min of rest interval. The primary outcome was to evaluate the agreement between 2WIN-S and OPD-Scan III, and the secondary outcome was to evaluate the repeatability of 2WIN-S. Scotopic pupil size of 100 right eyes of 100 adults (28 male and 72 female) aged 18–53 years (mean 36 ± 12 years) was assessed using OPD-Scan III and 2WIN-S, respectively. The mean scotopic pupil size of OPD-Scan III and 2WIN-S was recorded to be 6.24 ± 0.88 mm and 6.27 ± 0.81 mm, respectively. For the mean scotopic pupil size of OPD-Scan III and 2WIN-S the difference was − 0.03 mm (95%CI − 0.10 to 0.04 mm), p = 0.445, the 95% limits of agreement (LOA) was − 0.71 to 0.66 mm. ICC between the two devices was 0.92 (95% CI 0.88–0.94) (ICC > 0.9 indicates excellent consistency). Coefficients of repeatability (CoR) of 2WIN-S was 0.37, which has a high repeatability. For the mean scotopic pupil size of 2WIN-S of the repeated measurements, the difference was -0.04 mm (95%CI − 0.08 to 0.01 mm), p = 0.019, the 95% limits of agreement (LOA) was − 0.41 to 0.32 mm, with a narrow LOA. However, the majority of the variations were less than ± 0.50 mm (98% of scotopic pupil size measurements were below this threshold), within the clinically acceptable range (± 0.50 mm). Our study showed excellent agreement between 2WIN-S and OPD-Scan III (ICC > 0.9) and a good repeatability of 2WIN-S (CoR = 0.37). This study suggests a novel technique for measuring pupillary responses in low light conditions, which can be considered an alternative to OPD-Scan III in clinical settings.

Diagnosis of pediatric central nervous system tumors using methylation profiling of cfDNA from cerebrospinal fluid

Pediatric central nervous system tumors remain challenging to diagnose. Imaging approaches do not provide sufficient detail to discriminate between different tumor types, while the histopathological examination of tumor tissue shows high inter-observer variability. Recent studies have demonstrated the accurate classification of central nervous system tumors based on the DNA methylation profile of a tumor biopsy. However, a brain biopsy holds significant risk of bleeding and damaging the surrounding tissues. Liquid biopsy approaches analyzing circulating tumor DNA show high potential as an alternative and less invasive tool to study the DNA methylation pattern of tumors. Here, we explore the potential of classifying pediatric brain tumors based on methylation profiling of the circulating cell-free DNA (cfDNA) in cerebrospinal fluid (CSF). For this proof-of-concept study, we collected cerebrospinal fluid samples from 19 pediatric brain cancer patients via a ventricular drain placed for reasons of increased intracranial pressure. Analyses on the cfDNA showed high variability of cfDNA quantities across patients ranging from levels below the limit of quantification to 40 ng cfDNA per milliliter of CSF. Classification based on methylation profiling of cfDNA from CSF was correct for 7 out of 20 samples in our cohort. Accurate results were mostly observed in samples of high quality, more specifically those with limited high molecular weight DNA contamination. Interestingly, we show that centrifugation of the CSF prior to processing increases the fraction of fragmented cfDNA to high molecular weight DNA. In addition, classification was mostly correct for samples with high tumoral cfDNA fraction as estimated by computational deconvolution (> 40%). In summary, analysis of cfDNA in the CSF shows potential as a tool for diagnosing pediatric nervous system tumors especially in patients with high levels of tumoral cfDNA in the CSF. Further optimization of the collection procedure, experimental workflow and bioinformatic approach is required to also allow classification for patients with low tumoral fractions in the CSF.

Volumetric Analysis: Effect on Diagnosis and Management of Indeterminate Solid Pulmonary Nodules in Routine Clinical Practice.

To evaluate the effect of volumetric analysis on the diagnosis and management of indeterminate solid pulmonary nodules in routine clinical practice. This was a retrospective study with 107 computed tomography (CT) cases of solid pulmonary nodules (range, 6-15 mm), 57 pathology-proven malignancies (lung cancer, n = 34; metastasis, n = 23), and 50 benign nodules. Nodules were evaluated on a total of 309 CT scans (average number of CTs/nodule, 2.9 [range, 2-7]). CT scans were from multiple institutions with variable technique. Nine radiologists (attendings, n = 3; fellows, n = 3; residents, n = 3) were asked their level of suspicion for malignancy (low/moderate or high) and management recommendation (no follow-up, CT follow-up, or care escalation) for baseline and follow-up studies first without and then with volumetric analysis data. Effect of volumetry on diagnosis and management was assessed by generalized linear and logistic regression models. Volumetric analysis improved sensitivity (P = 0.009) and allowed earlier recognition (P < 0.05) of malignant nodules. Attending radiologists showed higher sensitivity in recognition of malignant nodules (P = 0.03) and recommendation of care escalation (P < 0.001) compared with trainees. Volumetric analysis altered management of high suspicion nodules only in the fellow group (P = 0.008). κ Statistics for suspicion for malignancy and recommended management were fair to substantial (0.38-0.66) and fair to moderate (0.33-0.50). Volumetric analysis improved interobserver variability for identification of nodule malignancy from 0.52 to 0.66 (P = 0.004) only on the second follow-up study. Volumetric analysis of indeterminate solid pulmonary nodules in routine clinical practice can result in improved sensitivity and earlier identification of malignant nodules. The effect of volumetric analysis on management recommendations is variable and influenced by reader experience.

Transformers for colorectal cancer segmentation in CT imaging.

Most recently transformer models became the state of the art in various medical image segmentation tasks and challenges, outperforming most of the conventional deep learning approaches. Picking up on that trend, this study aims at applying various transformer models to the highly challenging task of colorectal cancer (CRC) segmentation in CT imaging and assessing how they hold up to the current state-of-the-art convolutional neural network (CNN), the nnUnet. Furthermore, we wanted to investigate the impact of the network size on the resulting accuracies, since transformer models tend to be significantly larger than conventional network architectures. For this purpose, six different transformer models, with specific architectural advancements and network sizes were implemented alongside the aforementioned nnUnet and were applied to the CRC segmentation task of the medical segmentation decathlon. The best results were achieved with the Swin-UNETR, D-Former, and VT-Unet, each transformer models, with a Dice similarity coefficient (DSC) of 0.60, 0.59 and 0.59, respectively. Therefore, the current state-of-the-art CNN, the nnUnet could be outperformed by transformer architectures regarding this task. Furthermore, a comparison with the inter-observer variability (IOV) of approx. 0.64 DSC indicates almost expert-level accuracy. The comparatively low IOV emphasizes the complexity and challenge of CRC segmentation, as well as indicating limitations regarding the achievable segmentation accuracy. As a result of this study, transformer models underline their current upward trend in producing state-of-the-art results also for the challenging task of CRC segmentation. However, with ever smaller advances in total accuracies, as demonstrated in this study by the on par performances of multiple network variants, other advantages like efficiency, low computation demands, or ease of adaption to new tasks become more and more relevant.

P-434 Predicting endometrial thickness in hormone replacement therapy - frozen embryo transfer (HRT-FET): An AI algorithm for optimizing treatment management

Abstract Study question Can an AI-based algorithm assist in determining the optimal duration and dosage of estradiol supplementation to attain ideal endometrial width (EW) for non-responder HRT-FET patients? Summary answer An AI based algorithm can predict the probability of non-responder patients to achieve EW &amp;gt; 8mm given the current EWs’ estradiol dosage and duration. What is known already A thin endometrium is considered sub-optimal for transfer and is associated with reduced pregnancy rates. Various optional treatments have been proposed, including extended estrogen administration if time allows. Predicting whether patients will respond to an extended treatment may assist in patient selection, thus preventing prolonged treatment for non-responsive patients. Furthermore, as estradiol treatment may cause side effects and discomfort, the lowest effective dose is desired. Study design, size, duration A retrospective study consisting 683 FET HRT cycles (split to 80% train and 20% test), performed between 2018 and 2022. Each cycle had an initial treatment protocol of 6-10 days of estradiol resulting in an endometrial thickness of less than 8mm with at least one more additional ultrasound scan. Participants/materials, setting, methods A logistic regression classifier was trained to predict if the next ultrasound’s EW would exceed 8mm following any combination of oral/vaginal estradiol doses administered for 2, 3 or 4 days. The model was trained and calibrated using patients’ demographics, estradiol dose, endometrial thickness, and estradiol and progesterone levels. The model was evaluated using sensitivity and positive predictive values. Main results and the role of chance The most significant predicting factors were the first endometrial thickness, estradiol dosage - before and after the first ultrasound, the number of days until the next ultrasound scan, and the patient’s weight. The model’s performance was influenced by the number of days until the next ultrasound scan; therefore, the model was evaluated separately for 2, 3, and 4 days of following treatment for all estradiol doses. The model’s performance in predicting whether the EW will be higher than 8 is 2 day treatment prediction: Positive predictive value: 0.95, Sensitivity: 0.64 3 day treatment prediction: Positive predictive value: 0.88, Sensitivity: 0.68 4 day treatment prediction: Positive predictive value: 0.84, Sensitivity: 0.5 Average prediction: Positive predictive value: 0.9, Sensitivity: 0.64 Limitations, reasons for caution This is a retrospective study analyzing existing sonographic data with an inter-observer variability. The accuracy of the model decreased with longer intervals between ultrasound scans. A prospective study with designated sonographers is warranted to validate the results. Wider implications of the findings This decision-support algorithm can enhance the accuracy of estradiol dosing and may reduce the number of treatment days and ultrasound scans required to reach the optimal endometrial thickness during FET-HRT treatments. Trial registration number Not applicable

P-034 Revolutionizing sperm morphology assessment: A novel artificial intelligence optic microscope (AIOM)-powered platform for consistent and reliable analysis in male infertility diagnosis

Abstract Study question Does AIOM-based live sperm morphology annotation provide consistent and accurate results for assessing sperm morphology? Summary answer AIOM offers precise and reproducible sperm morphology evaluations which can enhance diagnostic efficiency with a single and reliable test. What is known already Semen analysis remains the primary diagnostic tool for male infertility. Unlike concentration and motility, sperm morphology shows a much higher correlation with paternal genetic integrity. However, it poses challenges due to inter-observer and inter-laboratory variability. The 2023 semen proficiency test highlighted these inconsistencies, reporting deviations of up to 13% among laboratories and acceptability ranging from 0% to 43%. This inconsistency is linked to complex staining procedures and interpretive subjectivity. To address these challenges, our study evaluates a novel AIOM-assisted platform, employing a live sperm morphology deep learning model. We aim to explore the platform’s reproducibility and reliability. Study design, size, duration In the first stage, liquefied semen specimens were split into three aliquots, each assigned to one of three operators (operators 1–3). Each operator assessed sperm morphology using Diff-Quik and live sperm morphology using the LensHooke® X12 semen analysis system (Bonraybio, Taichung). In the second stage, a single well-trained operator conducted sperm morphology assessments for three replicates (replicates 1-3) using the two methods. Interclass-correlation coefficients (ICCs) were calculated to assess consistency between replicates and operators. Participants/materials, setting, methods Semen samples (n = 30) were collected post 2-5 day abstinence and liquefied at room temperature for 30 minutes. Specimens were diluted to 100 million/mL with HTF, and underwent Diff-Quik staining according to WHO’s instruction. Aliquot of 3 μL semen specimen was loaded to 10-μm LensHooke® CS3 slide and analyzed by X12 analysis using live sperm morphology algorithm. Main results and the role of chance X12 utilizes a deep learning algorithm to automatically annotate sperm morphology, focusing on parameters such as sperm head, midpiece, tail, and excess residual cytoplasm, ultimately generating outputs for normal morphology and the teratozoospermic index (TZI) based on WHO strict criteria. Compared to manual Diff-Quik analysis, X12 showed higher consistency across all parameters, with ICCs above 0.85, unlike the manual method’s lower ICCs (below 0.70). The determination of abnormal midpiece had the lowest agreement between operators [ICC (95% CI): 0.15 (-0.61–0.58)]. In contrast, X12 demonstrated high agreement in all parameters, with ICC values consistently exceeding 0.85. When comparing the agreement of normal morphology rates between the two methods, a notable difference was observed [(Manual method: 0.59 (0.22–0.8) versus X12: 0.8 (0.61–0.9)]. To minimize inter-operator variability, a highly trained operator conducted three replicates following the manual method, thereby reducing deviations between measures. Remarkably, a single assessment of live sperm morphology produced results that were comparable to the mean average of repeated measures obtained through manual methods. In conclusion, the new platform for sperm morphology assessment, contingent on AIOM technology, is a reproducible and reliable tool in male infertility diagnosis. Limitations, reasons for caution This study’s limitations include its laboratory-based nature and a small sample size. Further research should include a more diverse and larger patient cohort to validate the efficacy and accuracy of AIOM in live sperm morphology assessment. Wider implications of the findings AIOM’s integration of automated interpretation simplifies sperm morphology assessment, reducing sample requirements and is smear-free. This advancement not only enhances diagnostic testing consistency but also eases andrologists’ workloads. This platform can contribute to standardized sperm analysis and improved laboratory management. Trial registration number Not applicable

P-619 AMH/AFC ratio as a predictive factor for ovarian responsiveness to gonadotropins during ovarian stimulation

Abstract Study question Does AMH/AFC ratio correlate to the number of follicles on the day of trigger and to the number of oocytes retrieved after ovarian stimulation? Summary answer The AMH/AFC ratio correlates significantly to the number of pre-ovulatory follicles and the number of oocytes collected after ovarian stimulation (OS). What is known already Although AFC is a robust marker of oocyte yield after OS, its combination with Anti-Mullerian hormone (AMH) seems to improve the predictability. Besides, Follicular Output Rate [FORT: (number of pre-ovulatory follicles on day of trigger/AFC at baseline) x 100] and Follicle-to-Oocyte Index [FOI: (number of retrieved oocytes/AFC) × 100] evaluate the ovarian responsiveness to gonadotropins during OS, representing its sensitivity to exogenous FSH. Both parameters take antral follicle count (AFC) only into consideration. AMH and AFC as a ratio might optimize the accuracy of FORT and FOI metrics, helping to predict which patients are at higher risk of suboptimal results. Study design, size, duration This retrospective study analyzed 2,163 ovarian stimulation cycles for IVF/ICSI, performed between May 2015 and October 2023 in a tertiary referral IVF centre. For all the cycles, measurement of AMH using Elecsys (Roche®) no longer than 6 months before starting OS, AFC at baseline, number of pre-ovulatory follicles on the day of trigger, and the treatment outcomes were available. Participants/materials, setting, methods Couples with primary/secondary infertility who underwent OS with antagonist protocol, starting on day 2/3 for IVF/ICSI were included, independent of their age or ovarian reserve parameters. Exclusion criteria were patients with body mass index (BMI) &amp;gt;40 kg/m2, non-vaginal ultrasound measurement for AFC and/or follicles during OS, cycles with previous estradiol priming or oral contraceptives. Ethical approval was obtained from the Research Ethics Committee. Main results and the role of chance 2,163 ovarian stimulation cycles were included in the analysis. The mean age, BMI, AMH, AFC, FOI and FORT of the included women were: 35.6±6.1 years, 27.5±4.7 kg/m2, 2.4±2.1 ng/mL, 12±7 follicles, 1.02±0.66 and 1.38±0.79. The correlation of AMH/AFC ratio (coefficient: 0.62, 95% CI: 0.59 to 0.65, p&amp;lt;0.001) to FOI was stronger than the correlation of AMH (coefficient: 0.31, 95% CI: 0.27 to 0.34, p&amp;lt;0.001) or AFC alone (coefficient: -0.09, 95% CI: -0.14 to -0.05, p&amp;lt;0.001). The correlation of AMH/AFC ratio (coefficient: 0.62, 95% CI: 0.59 to 0.64, p&amp;lt;0.001) to FORT was also stronger than the correlation of AMH (coefficient: 0.21, 95% CI: 0.16 to 0.25, p&amp;lt;0.001) or AFC alone (coefficient: -0.23, 95% CI: -0.27 to -0.19, p&amp;lt;0.001). AMH/AFC ratio was better at predicting stimulation efficiency parameters, FOI and FORT, compared to joint use of AMH and AFC (p&amp;gt;0.001 for both). However, regression analysis showed stand-alone use of the AMH/AFC ratio did not provide a better prediction for collected oocyte count compared to joint use of AMH and AFC. In fact, prediction errors were significantly larger for the AMH/AFC ratio (median: -0.4, range: -11 to 11) compared to the model using AMH and AFC separately (median: -0.2, range: -6 to 7). Limitations, reasons for caution Although clinical assessment was performed in the same centre following the same methodology, inter-observer variability for AFC is a limitation. Additionally, AMH levels used in the study were not obtained in the same cycles, which may adversely affect the performance of the ratio. Wider implications of the findings While AMH/AFC ratio was better at predicting stimulation efficiency compared to FOI and FORT parameters, it was not significantly better at predicting absolute oocyte yield. This may imply AMH/AFC ratio can represent the ovarian sensitivity to exogenous FSH. Trial registration number Not applicable

P-310 Impact of adenomyosis on the live birth rate during management by in vitro fertilization : a French study

Abstract Study question Evaluate the impact of adenomyosis on the likelihood of live birth after a fresh embryo transfer. Summary answer Adenomyosis reduces the live-birth rate. The location of adenomyosis and the type of protocol appears to have no impact on the live birth rates. What is known already The prevalence of adenomyosis varies between 7 and 27%, and affects 25% of infertile women. Its impact on fertility remains debated. According to some studies, it leads to a lower pregnancy rate, a doubling of the spontaneous miscarriage rate and a lower live birth rate as compared to non-adenomyosis women. However according to some authors adenomyosis does not reduce the live birth rate. Similarly, the impact of the location of adenomyosis remains debated. On the impact of the stimulation protocol, several studies have shown that the use of GnRH agonist improves the live birth rate in patients with adenomyosis. Study design, size, duration This is a retrospective, descriptive and observational study, managed at the fertility department of Dijon University Hospital between January 2015 and March 2020. Systematically, the women infertility cause was assessed. We compared the group of patients with adenomyosis (i.e. symptomatic adenomyosis completed with a 3D ultrasound within a maximum of 1 year prior to management) with controls (i.e. patients treated for another cause of infertility). Ultrasound was performed by the same two experienced operators. Participants/materials, setting, methods We included 340 patients aged 18-41 years. Data concerning live births and type of stimulation protocol were collected from the patient’s electronic medical and therapeutic records. MUSA classification was used to determine whether the adenomyosis was localized or diffuse. The primary endpoint of this study was the live-birth rate (LBR) after fresh embryo transfer. The implantation and miscarriage rates were also compared. Main results and the role of chance We included 92 patients in adenomyosis group, and 248 patients in control group. Baseline characteristics of patients including women age, BMI and smoking habits were comparable in both groups. Moreover, no difference was found on the mean number of mature oocytes and embryos between the two groups. LBR after fresh embryo transfer was significantly lower in the adenomyosis group than in the control (34.8 % versus 41.5%, respectively, p = 0,02). There was no significant difference in implantation rate (35.7% versus 37,4%, p = 0.76). Early miscarriage rate (&amp;lt;12 weeks of amenorrhea) was higher in the control group than the adenomyosis group (15.3 versus 7.5, p = 0.012). Concerning late miscarriage (&amp;gt;12 weeks of amenorrhea), we observed 3 medical termination of pregnancy (all for fetal malformation) and 1 miscarriage (10%) in the adenomyosis group and 1 miscarriage (0.81%) in the control group (p = 0.02). Subgroup analyses showed no decrease in LBR according to adenomyosis location (32.7% versus 37.2% in the diffuse and localized groups, respectively, p = 0.46). Limitations, reasons for caution Even though our cohort is a good reflection of the all-French population of Burgundy since all patients are treated at the fertility center of Dijon. Despite its retrospective nature and the small number of included adenomyosis-patients, the ultrasound scans were performed by two experienced operators that reduced inter-observer variability. Wider implications of the findings It would be useful to carry out a prospective study including more patients to better assess the impact of location on implantation. It would also be interesting to evaluate neonatal and obstetrical complications, especially the late miscarriage issue requiring further investigation. Trial registration number non-clinical trials

P-694 Fully automated follicle counting matches human accuracy levels in predicting stimulation outcomes

Abstract Study question Can automatic measurement of follicles be used to predict the number of oocytes retrieved as accurately as manual physicians’ reports? Summary answer Automated follicular annotations generated by an artificial intelligence platform have similar predictive ability of the number of retrieved and mature oocytes as physician reports. What is known already Ovarian follicle counting is a time-consuming, frequently performed assessment, which has significant inter-observer variability. It’s a perfect candidate for automation using artificial intelligence. The main purpose of this monitoring is to determine the optimal day for trigger administration which maximises the number of retrieved oocytes and increases the chance of pregnancy. This decision typically depends on follicle sizes and hormone levels, while also considering the unique conditions of each patient. Research in this area is ongoing, but it primarily focuses on follicle sizes as measured by physicians, rather than by automated systems. Study design, size, duration This study included 589 in vitro cycles of 545 patients performed in 5 centers in Poland between 2019 and 2021. The 719 examinations that were done at most 3 days before the administration of the trigger were analyzed. Manual reports from follicle measurements have been collected, together with 2D cine loop scans (videos) of the ovaries, and final counts of retrieved oocytes and mature (stage MII) oocytes. The Folliscan artificial intelligence platform provided automatic measurements. Participants/materials, setting, methods Follicles in each ovary were grouped by size, starting from 8mm, then every 2mm up to 24mm, and one group for follicles larger than 24mm. The number of follicles in each group, with the number of days until the trigger shot, was used as features in a linear regression model to predict the number of retrieved oocytes, or MIIs. Two datasets were created: one based on physician reports, and the other on automatic measurements. Main results and the role of chance Ten-fold cross-validation was performed. The model based on manual measurements achieved an R² of 0.62 (Confidence Interval: 0.56–0.68) for predicting the number of retrieved oocytes and 0.51 (CI: 0.45–0.57) for MIIs. In comparison, the model based on automatic annotations achieved an R² of 0.63 (CI: 0.57–0.69) for oocytes and 0.50 (0.43–0.56) for MIIs. The mean average errors (MAE) in the predicted number of oocytes and MIIs were 3.12 (CI: 2.88–3.36) and 2.54 (CI: 2.36–2.72), respectively, for manual measurements. They amounted to 3.09 (CI: 2.85–3.33) and 2.56 (CI: 2.38–2.73) for automatic annotations. Two one-sided tests for equivalence were conducted on the mean average error, with a margin set at 0.2. The results yielded a p-value of 0.026 for oocytes errors and 0.0023 for MIIs errors. Limitations, reasons for caution The study had a limited scope, focusing on research from 5 in vitro clinics. A linear regression model was used, comparing only the potential of automatic measurements with manual measurements. Further development of the model is needed to achieve better accuracy. Wider implications of the findings Manual follicular measurements and automated measurements provide similar predictive values in predicting the number of retrieved and mature oocytes. Automated measurements can be used during stimulation monitoring, offering a reliable and efficient alternative to manual methods. However, clinical studies focusing on interpretable metrics are needed to further validate these findings. Trial registration number Polish National Center for Research and Development no. POIR.01.01.01-00-1634/20-00 and ERC Consolidator Grant TUgbOAT no. 772346.

O-218 Prospective validation of an artificial intelligence platform for automated follicle counting

Abstract Study question How often are fully automated follicle counts and measurements modified in expert review? Summary answer Automatic follicle annotations provided by an artificial intelligence platform in regular operation of a clinic were edited 3.27% of the time. What is known already Ovarian follicle counting is frequently performed, time-consuming, and subject to noticeable inter-observer variability; as such, it is well-suited to automatizing with artificial intelligence. FOLLISCAN (MIM Fertility) is a software platform that automatically annotates follicles on 2D or 3D ultrasound cine videos, with exact outlines and measurements, without any manual pre-processing. It has undergone several retrospective tests of its precision and recall in detecting ovarian follicles, accuracy of measurements, and time-savings compared to manual counting Study design, size, duration The platform was integrated with a clinic’s existing picture archiving and medical record systems. Ultrasound scans of ovaries were performed as part of regular infertility treatment in two IVF centers. Scans were made by 3 experts in a period from November to December 2023, automatically sent to the platform, analyzed, and then immediately visualized for review by the person performing the ultrasound, in place of manual annotation. Participants/materials, setting, methods The study included 294 cine videos from 147 examinations of 101 patients, with 4347 follicles in total (14.7 per ovary on average, from 2mm). The platform allows users to modify proposed measurements and introduce new follicle annotations. After confirming the review, results were automatically sent back to the clinic’s system to be used in medical decisions just as manual annotations would be. Main results and the role of chance Among all videos the average number of editions (follicle additions, modifications, or deletions) was 0.48 (CI: 0.37-0.61). In total, 142 follicles out of 4347 (3.27%, CI: 2.74-3.80) were edited: 66 follicles were added (1.52%, CI: 1.15-1.88), 26 were modified, (0.60%, CI: 0.37-0.83 ), and 50 were deleted (1.15%, CI: 0.83–1.47). Out of the 151 edited follicles, 13 (9.2%) had 2–5 mm, 29 (20.4%) had 5-10 mm, 51 (35.9%) had 10-15 mm, 42 (29.6%) had 15-20 mm, and 7 (4.9%) had 20 mm or more in diameter. Among follicles of 15 mm or more in size, 49 were edited. Out of 294 videos, 87 (29.6%, 95% Confidence Interval: 24.4-34.6) were edited in review in any way. Limitations, reasons for caution The study was limited in scope and did not track patients until success rates in IVF treatments could be observed. A multi-center randomized control trial could compare pregnancy rates for treatments performed with and without the platform. Wider implications of the findings An integrated platform allows for easy review, while significantly reducing the time spent on follicle counting in a real-world setting. Automated annotations result in consistent, reliable, and quick measurements, with the number of expert modifications smaller than the inter-observer reliability reported in previous studies. Trial registration number Project support was provided by the Polish National Center for Research and Development no. POIR.01.01.01-00-1634/20-00 and ERC Consolidator Grant TUgbOAT no. 772346. This study was conducted following the approval of the research protocol by the review board of the Regional Medical Chamber in Gdańsk (approval no. KB – 51/22).