نرم افزاری با r یا stat

---

شرح پروژه

سلام وقتتون بخیر  توضیحات رو در ویس ارسال میکنم 

قسمت نتایج نیاز به گزارش SROC, Forest plot, Bivariate box, Fagan’s nomogram and Likelihood matrix می باشد. تمامی مقالات متاانالیز تشخیصی این نوع گزارشات را حتما شامل خواهند شد برای انتشار. 

پیشنهادات و کامنت های اختیاری: 

برای انالیز پیشنهاد می کنم از دستور metadta استفاده کنید. این دستور محدودیت انالیز با 3 مطالعه اولیه را برای متاانالیز تشخیصی حل کرده است و میتواند محاسبات لازم را برای انالیزهای زیر گروهی که تعداد مطالعات کمتر از 3 است، گزارش دهید.

Method 

A diagnostic test accuracy (DTA) meta-analysis was conducted to evaluate the performance of models for detecting lymph node metastasis, with analyses performed in a stepwise, hierarchical manner to investigate the effects of input data types, CT phases, segmentation techniques, and DL model architectures. Only subgroups with at least three included studies were eligible for statistical pooling. All analyses were conducted separately for internal validation (IVC) and external validation (EVC) datasets to assess model performance across different validation settings. For the subgroup with more than three studies, statistical analysis was performed using Stata with the MIDAS command.A bivariate random-effects model was employed to estimatepooled sensitivity and specificity, and hierarchical summary receiver operating characteristic (HSROC) curves were generated to assess overall diagnostic performance. Diagnostic odds ratios (DOR), positive likelihood ratios (LR+), and negative likelihood ratios (LR-) were calculated. The area under the HSROC curve (AUC) and its 95% confidence intervals (CIs) were calculated for each subgroup. Heterogeneity was assessed using the Chi-square test and I² statistic, with I² values interpreted to indicate the degree of variability across studies. Publication bias was evaluated using Egger’s tests, with p-values reported to determine statistical significance. For subgroups with fewer than three included studies, implementation of the bivariate model was inappropriate due to the instability of variance-covariance estimates. Instead, a random-effects meta-analysis for sensitivity and specificity was performed separately using the meta package in the R environment (version 4.5.1), applying logit transformation (PLOGIT) to stabilize variances. In these subgroups, pooled sensitivity and specificity with 95% CIs were reported. Due to limited sample size and unstable estimates, AUC was not calculated. Assessment of publication bias was not performed as statistical tests for bias, such as Egger's test, have insufficient statistical power and yield unreliable results with such a small number of studies. The statistical analyses were performed by A.A. and S.T..

Subgroup analyses

The meta-analysis was conducted in a stepwise, hierarchical manner to systematically evaluate the different factors on diagnostic performance. The analyses were structured as follows: 1. Input Data Types: Pooled sensitivity and specificity were calculated for internal and external validation datasets, stratified by input data types (DLF, DLF+ HCRF, or DLF+ HCRF+ clinical variables). This analysis aimed to assess the baseline performance of models based on the type of input data used. 2. CT Phases by Input Data Type: Models were further stratified by CT phase within the DLF-based model and analyzed separately in the internal validation set. While studies utilized arterial, venous, portal venous, parenchymal, and unenhanced CT imaging, only arterial and combined portal venous/venous phase models had sufficient data for analysis.Due to limited data for the venous phase, the portal phase was combined with the venous phase for analysis, as both are functionally similar. 3. Segmentation Techniques by Input Data Type: Models were categorized by segmentation technique (manual, semi-automatic, or automatic) within DLF-based models, and analyzed for internal datasets. Due to insufficient data on semi-automatic and automatic segmentation techniques, only manual segmentation had enough studies to be included in the meta-analysis. Results for semi-automatic and automatic segmentation were reported descriptively. 4. DL model Architectures by Input Data Type: DL models were categorized by algorithm type within each input data type and analyzed separately in internal and external validation datasets. OnlyCNN-based models had a sufficient number of studies for meta-analysis, as other architectures (e.g., Graph-based models and Vision Transformers) lacked sufficient data. 5.

 CNN models by Segmentation Technique, CT Phase, and Input Data Type: CNN models were analyzed by segmentation technique and CT phase within the DLF-based models for internal validation datasets. All models in this step utilized venous-phase CT (including portal phase) and manual segmentation.

اسکریپت r رو بفرستم نگاه کنید

این پروژه شامل 3 فایل مهم است، لطفا قبل از ارسال پیشنهاد حتما نسبت به بررسی این فایل اقدام فرمایید.

مهارت ها و تخصص های مورد نیاز

---

---