Current Medical Imaging - Volume 20, Issue 1, 2024

This study aimed to predict liver injury in AP patients by establishing a radiomics model based on CECT.

A total of 1223 radiomic features were extracted from late arterial-phase pancreatic CECT images of 209 AP patients (146 in the training cohort and 63 in the test cohort), and the optimal radiomic features retained after dimensionality reduction by LASSO were used to construct a radiomic model through logistic regression analysis. In addition, clinical features were collected to develop a clinical model, and a joint model was established by combining the best radiomic features and clinical features to evaluate the practicality and application value of the radiomic models, clinical model, and combined model.

Four potential features were selected from the pancreatic parenchyma to construct the radiomic model, and the AUC of the radiomic model was significantly greater than that of the clinical model for both the training cohort (0.993 vs. 0.653, p = 0.000) and test cohort (0.910 vs. 0.574, p = 0.000). The joint model had a greater AUC than the radiomics model for both the training cohort (0.997 vs. 0.993, p = 0.357) and the test cohort (0.925 vs. 0.910, p = 0.302).

The radiomic model based on CECT has good performance in predicting liver injury in AP patients and can guide clinical decision-making and improve the prognosis of patients with AP.

The primary objective of this study was to conduct a comparative analysis of the diagnostic efficacy of [¹⁸F]-FDG PET/CT and [¹⁸F]-FDG PET/MRI in the detection of breast cancer lymph node metastasis.

We conducted a comprehensive search on PubMed, Embase, and Web of Science databases, encompassing eligible articles until March 2023. The pooled sensitivity and specificity for [¹⁸F]-FDG PET/CT and [¹⁸F]-FDG PET/MRI have been reported as estimates with 95% Confidence Intervals (CIs) using a bivariate random-effect model. Utilizing the I square (I²) statistic, heterogeneity among pooled studies was evaluated. The quality assessment of the included studies was conducted using the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) methodology.

We included 18 studies (2057 patients). The sensitivity, specificity, and AUC (Area Under the Curve) values of [¹⁸F]-FDG PET/CT for overall lymph node metastasis in breast cancer have been found to be 0.58 (0.39 - 0.75), 0.83 (0.69-0.92), and 0.79 (0.75-0.82), respectively. Correspondingly, the values for [¹⁸F]-FDG PET/MRI were found to be 0.76 (0.60-0.88), 0.85 (0.77-0.91), and 0.89 (0.86-0.91), respectively. The sensitivity, specificity, and AUC values of [¹⁸F]-FDG PET/CT for axillary lymph node metastasis in breast cancer were 0.52 (0.37-0.67), 0.84 (0.68-0.92), and 0.73 (0.69-0.76), respectively. Correspondingly, the values for [¹⁸F]-FDG PET/MRI were 0.84 (0.76-0.89), 0.87 (0.75-0.94), and 0.86 (0.83-0.89), respectively.

This study has suggested [¹⁸F]-FDG PET/MRI to have greater diagnostic power than [¹⁸F]-FDG PET/CT in detecting lymph node metastasis in breast cancer. However, the [¹⁸F]-FDG PET/MRI results have been obtained from a small sample size study, and more and larger prospective studies are needed for further confirmation on this issue.

Early detection of pulmonary nodules is critical for the clinical diagnosis and management of pulmonary nodules. Computed tomography imaging is currently the best imaging method for detecting pulmonary nodules.

This study proposes and applies a new thresholding-based method for identifying pulmonary nodules in computed tomography images.

The proposed method involves segmenting the lung volume and identifying candidate nodules based on their intensity levels, which are higher than those of the lung parenchyma. Reference points on the histogram curve are used to determine a threshold value, and filtering by geometric characteristics is applied to reduce false positives. The performance of the proposed method is evaluated on a training set consisting of 35 nodules distributed among 16 cases with ground truth using the SPIE-AAPM Lung CT Challenge Database and ELCAP Public Lung Image Database.

The proposed method shows a significant reduction in false positives, filtering from an average of 12,380 candidate nodules to 19 detected nodules. The method also demonstrates a sensitivity of 88.6% for detecting pulmonary nodules with an error of 1 nodule in cases where complete detection is not reached.

The proposed thresholding-based method improves the sensitivity of identifying pulmonary nodules in computed tomography images while reducing false positives.

Accurate identification of vascular lumen region founded the base of bubble detection and bubble grading, which played a significant role in the detection of vascular gas emboli for the diagnosis of decompression sickness.

To assist in the detection of vascular bubbles, it is crucial to develop an automatic algorithm that could identify vascular lumen areas in ultrasound videos with the interference of bubble presence.

This article proposed an automated vascular lumen region recognition (VLRR) algorithm that could sketch the accurate boundary between vessel lumen and tissues from dynamic 2D ultrasound videos. It adopts 2D ultrasound videos of the lumen area as input and outputs the frames with circled vascular lumen boundary of the videos. Normalized cross-correlation method, distance transform technique, and region growing technique were adopted in this algorithm.

A double-blind test was carried out to test the recognition accuracy of the algorithm on 180 samples in the images of 6 different grades of bubble videos, during which, intersection over union and pixel accuracy were adopted as evaluation metrics. The average IOU on the images of different bubble grades reached 0.76. The mean PA on 6 of the images of bubble grades reached 0.82.

It is concluded that the proposed method could identify the vascular lumen with high accuracy, potentially applicable to assist clinicians in the measurement of the severity of vascular gas emboli in clinics.

The second highest cause of death among males is Prostate Cancer (PCa) in America. Over the globe, it’s the usual case in men, and the annual PCa ratio is very surprising. Identical to other prognosis and diagnostic medical systems, deep learning-based automated recognition and detection systems (i.e., Computer Aided Detection (CAD) systems) have gained enormous attention in PCA.

These paradigms have attained promising results with a high segmentation, detection, and classification accuracy ratio. Numerous researchers claimed efficient results from deep learning-based approaches compared to other ordinary systems that utilized pathological samples.

This research is intended to perform prostate segmentation using transfer learning-based Mask R-CNN, which is consequently helpful in prostate cancer detection.

Lastly, limitations in current work, research findings, and prospects have been discussed.

This research investigates the impact of pruning on reducing the computational complexity of a five-layered Convolutional Neural Network (CNN) designed for classifying MRI brain tumors. The study focuses on enhancing the efficiency of the model by removing less important weights and neurons through pruning.

This research aims to analyze the impact of pruning on the computational complexity of a CNN for MRI brain tumor classification, identifying optimal pruning percentages to balance reduced complexity with acceptable classification performance.

The proposed CNN model is implemented for the classification of MRI brain tumors. To reduce time complexity, weights and neurons of the trained model are pruned systematically, ranging from 0 to 99 percent. The corresponding accuracies for each pruning percentage are recorded to assess the trade-off between model complexity and classification performance.

The analysis reveals that the model's weights can be pruned up to 70 percent while maintaining acceptable accuracy. Similarly, neurons in the model can be pruned up to 10 percent without significantly compromising accuracy.

This research highlights the successful application of pruning techniques to reduce the computational complexity of a CNN model for MRI brain tumor classification. The findings suggest that judicious pruning of weights and neurons can lead to a significant improvement in inference time without compromising accuracy.

Deep learning models have recently been preferred to perform certain image-processing tasks. Recently, with the increasing radiation, heat, and poor lighting conditions, the raw image samples may contain noisy and ambiguous information.

To process these images, the deep learning model requires a large number of data samples to learn the missing information from other clear data samples. This necessitates training the neural network with a huge dataset.

The researchers are now attempting to filter and improve such noisy images via preprocessing in order to provide valid and accurate feature information to the neural network layers. However, certain research studies claim that some useful information may be lost when the image is not preprocessed with an appropriate filter or enhancement technique. The MSA (meta-synthesis and analysis) approach is utilized in this work to present the impact of the image processing applications done with and without preprocessing steps. Also, this work summarizes the existing deep learning-based image processing models utilizing or not preprocessing steps in their implementation.

This work has also found that 85% of the existing techniques involve a preprocessing step while developing a deep learning model. However, a maximum accuracy of 96.89% is observed on Sine-Net when it is implemented without a preprocessing and the same model gave 96.85% when implemented with preprocessing.

This research provides various research insights on the requirement and non-requirement of preprocessing steps in a deep learning-based implementation.

Hyperostosis frontalis interna (HFI) is a common and often incidental finding seen on imaging. There is a significant paucity of radiology literature, particularly regarding the MRI imaging appearance of HFI.

We reported two cases of HFI on MRI, which showed focal enhancement. These were stable on long-term follow-up studies and thought to be most consistent with benign enhancement.

Further studies are needed to elucidate the underlying pathogenesis; however, it is important to be aware that regions of HFI may demonstrate variable enhancement and are sometimes mistaken for osseous metastatic disease.

To investigate the feasibility of constructing new geometric parameters that correlate well with dosimetric parameters.

100 rectal cancer patients were enrolled. The targets were identified manually, while the organs at risk (bladder, small bowel, left and right femoral heads) were segmented both manually and automatically. The radiotherapy plans were optimized according to the automatically contoured organs at risk. Forty cases were randomly selected to establish the relationship between dose and distance for each organ at risk, termed “dose-distance curves,” which were then applied to the new geometric parameters. The correlation between these new geometric parameters and dosimetric parameters was analyzed in the remaining 60 test cases.

The “dose-distance curves” were similar across the four organs at risk, exhibiting an inverse function shape with a rapid decrease initially and a slower rate at a later stage. The Pearson correlation coefficients of new geometric parameters and dosimetric parameters in the bladder, small intestine, and left and right femur heads were 0.96, 0.97, 0.88, and 0.70, respectively.

The new geometric parameters predicated on “distance from the target” showed a high correlation with corresponding dosimetric parameters in rectal cancer cases. It is feasible to utilize the new geometric parameters to evaluate the dose deviation attributable to automatic segmentation.

Patient safety is paramount in ultrasound procedures, particularly in obstetric ultrasounds involving both the mother and fetus. The thermal and mechanical indices (TI and MI) serve as crucial indicators of the acoustic output during ultrasound. Clinicians and specialists must know these indices and ensure they are within safe ranges. This study aimed to assess the parameters of acoustic output power employed in obstetric ultrasound (thermal and mechanical index).

A cross-sectional observational study conducted at Maternity and Children's Hospital in Al-Madinah Al-Munawwarah, the data was collected from obstetric scanning of 411 pregnant females using a data collection sheet including gravida and women's age, gestational age, scan mode, scan time, and thermal and mechanical index (TI and MI) values.

The study found that there were significant differences in safety indices measurement between different modes; in Pulsed Doppler, mean Thermal Index Bone (TIb) had the highest value (1.60±0.40), and the Mechanical Index (MI) was the lowest (0.68±0.33). There were insignificant differences in safety indices values in different modes in different trimesters. The thermal indices of soft tissue and bony structure (TIs and TIb) of brightness mode (B-mode) were constant in all trimesters, but the MI in the first trimester was lower than in the other trimesters.

This study concluded that the mean values of thermal indices used in B mode , M mode and Color Doppler lie within the Recommended limit of (BMUS) British Medical Ultrasound Society (below 0.7) except for Pulsed Doppler it was exceed 1.5. While for MI in different ultrasound modes except in pulsed Doppler the average values is higher than 0.7 which was recommended by (BMUS) British Medical Ultrasound Society and lower than 1.9 which was the maximum threshold approved by FDA (Food and drug Administration). The average scanning time is low (6.4 minute) reflect the safe use of ultrasound in obstetrics in this study.

Utilizing ultrasound radiomics, we developed a machine learning (ML) model to construct a nomogram for the non-invasive evaluation of glomerular status in diabetic kidney disease (DKD).

Patients with DKD who underwent renal biopsy were retrospectively enrolled between February 2017 and February 2023. The patients were classified into mild or moderate-severe glomerular severity based on pathological findings. All patients were randomly divided into a training (n = 79) or testing cohort (n = 35). Radiomics features were extracted from ultrasound images, and a logistic regression ML algorithm was applied to construct an ultrasound radiomic model after selecting the most significant features using univariate analysis and the least absolute shrinkage and selection operator algorithm (LASSO). A clinical model was created following univariate and multivariate logistic regression analyses of the patient's clinical characteristics. Then, the clinical-radiomics model was constructed by combining rad scores and independent clinical characteristics and plotting the nomogram. The receiver operating characteristic curve (ROC) and decision curve analysis (DCA), respectively, were used to evaluate the prediction abilities of the clinical model, ultrasound-radiomics model, and clinical-radiomics model.

A total of 114 DKD patients were included in the study, including 43 with mild glomerulopathy and 71 with moderate-severe glomerulopathy. The area under the curve (AUC) for the clinical model based on clinical features and the radiomic model based on 2D ultrasound images in the testing cohort was 0.729 and 0.761, respectively. Further, the AUC for the clinical-radiomics nomogram was constructed by combining clinical features, and the rad score was 0.850 in the testing cohort. The outcomes were better than those of both the radiomic and clinical single-model approaches.

The nomogram constructed by combining ultrasound radiomics and clinical features has good performance in assessing the glomerular status of patients with DKD and will help clinicians monitor the progression of DKD.

This study aimed to evaluate the Pharmacovigilance (PV) and severity of hypersensitivity reactions induced by non-ionic Iodinated Contrast Media (ICM) in the radiology diagnosis reported to the United States Food and Drug Administration Adverse Event Reporting System (FAERS).

We retrospectively reviewed the reports of ICM-induced hypersensitivity reactions submitted to the FAERS database between January 2015 and January 2023 and conducted a disproportionality analysis. The seven most common non-ionic ICM, including iohexol, iopamidol, ioversol, iopromide, iomeprol, iobitridol, and iodixanol, were chiefly analyzed. Our primary endpoint was the PV of non-ionic ICM-induced total hypersensitivity events. STATA 17.0 MP was used for statistical analysis.

In total, 35357 reports of adverse reaction events in radiology diagnosis were retrieved from the FAERS database. Among them, 6181 reports were on hypersensitivity reaction events (mean age: 57.1 ± 17.8 years). The hypersensitivity reaction-related PV signal was detected for iohexol, ioversol, iopromide, iomeprol, iobitridol, and iodixanol, but not for iopamidol. The proportion of iomeprol-induced hypersensitivity reactions and the probability of ioversol-induced severe hypersensitivity reactions have been found to be significantly increased.

The probability and severity of hypersensitivity reaction events in non-ionic ICM are different. Iohexol, ioversol, iopromide, iomeprol, iobitridol, and iodixanol have higher risks compared to iopamidol. In addition, the constituent ratio of hypersensitivity reactions induced by iomeprol is significantly increased, and the associated probability induced by ioversol is significantly increased.

The increasing longevity of the population has made Alzheimer's disease (AD) a significant public health concern. However, the challenge of accurately distinguishing different disease stages due to limited variability within the same stage and the potential for errors in manual classification highlights the need for more precise approaches to classifying AD stages. In the field of deep learning, the ResNet50V2 model stands as a testament to its exceptional capabilities in image classification tasks.

The dataset employed in this study was sourced from Kaggle and consisted of 6400 MRI images that were meticulously collected and rigorously verified to assure their precision. The selection of images was conducted with great attention to detail, drawing from a diverse array of sources.

This study focuses on harnessing the potential of this model for AD classification, a task that relies on extracting disease-specific features. Furthermore, to achieve this, a multi-class classification methodology is employed, using transfer learning and fine-tuning of layers to adapt the pre-trained ResNet50V2 model for AD classification. Notably, the impact of various input layer sizes on model performance is investigated, meticulously striking a balance between capacity and computational efficiency. The optimal fine-tuning strategy is determined by counting layers within convolution blocks and selectively unfreezing and training individual layers after a designated layer index, ensuring consistency and reproducibility. Custom classification layers, dynamic learning rate reduction, and extensive visualization techniques are incorporated.

The model's performance is evaluated using accuracy, AUC, precision, recall, F1-score, and ROC curves. The comprehensive analysis reveals the model's ability to discriminate between AD stages. Visualization through confusion matrices aided in understanding model behavior. The rounded predicted labels enhanced practical utility.

This approach combined empirical research and iterative refinement, resulting in enhanced accuracy and reliability in AD classification. Our model holds promise for real-world applications, achieving an accuracy of 96.18%, showcasing the potential of deep learning in addressing complex medical challenges.

Accurate segmentation of liver tumor regions in medical images is of great significance for clinical diagnosis and the planning of surgical treatments. Recent advancements in machine learning have shown that convolutional neural networks are powerful in such image processing while largely reducing human labor. However, the variable shape, fuzzy boundary, and discontinuous tumor region of liver tumors in medical images bring great challenges to accurate segmentation. The feature extraction capability of a neural network can be improved by expanding its architecture, but it inevitably demands more computing resources in training and hyperparameter tuning.

This study presents a Dynamic Context Encoder Network (DCE-Net), which incorporates multiple new modules, such as the Involution Layer, Dynamic Residual Module, Context Extraction Module, and Channel Attention Gates, for feature extraction and enhancement.

In the experiment, we used a liver tumor CT dataset of LiTS2017 to train and test the DCE-Net for liver tumor segmentation. The experimental results showed that the four evaluation indexes of the method, precision, recall, dice, and AUC, were 0.8961, 0.9711, 0.9270, and 0.9875, respectively. Furthermore, our ablation study reported that the accuracy and training efficiency of our network were markedly superior to the networks without involution or dynamic residual modules.

Therefore, the DCE-Net proposed in this study has great potential for automatic segmentation of liver lesion tumors in the clinical diagnostic environment.

Patients with diffuse large B-cell lymphoma (DLBCL) often experience a poor prognosis due to cardiac damage induced by anthracycline chemotherapy, with left ventricular diastolic dysfunction manifesting early. Vector Flow Mapping (VFM) is a novel technology, and its effectiveness in detecting left ventricular diastolic dysfunction following anthracycline chemotherapy remains unverified.

This study evaluates left ventricular diastolic function in DLBCL patients after anthracycline chemotherapy using vector flow mapping (VFM).

We prospectively enrolled 54 DLBCL patients who had undergone anthracycline chemotherapy (receiving a minimum of 4 cycles) as the case group and 54 age- and sex-matched individuals as controls. VFM assessments were conducted in the case group pre-chemotherapy (T0), post-4 chemotherapy cycles (T4), and in the control group. Measurements included basal, middle, and apical segment energy loss (ELb, ELm, ELa) and intraventricular pressure differences (IVPDb, IVPDm, IVPDa) across four diastolic phases: isovolumic relaxation (D1), rapid filling (D2), slow filling (D3), and atrial contraction (D4).

When comparing parameters between the control and case groups at T0, no significant differences were observed in general data, conventional ultrasound parameters, and VFM parameters (all P > 0.05). From T0 to T4, ELa significantly increased throughout the diastole cycle (all P < 0.05); ELm increased only during D4 (all P < 0.05); and ELb increased during D1, D2, and D4 (all P < 0.05). All IVPD measurements (IVPDa, IVPDm, IVPDb) increased during D1 and D4 (all P < 0.05) but decreased during D2 and D3 (all P < 0.05). Significant positive correlations were identified between ELa-D4, IVPDa-D4, and parameters A, e’, E/e,’ and LAVI (all r > 0.5, all P < 0.001). Negative correlations were noted with E/A for ELa-D4 IVPDa-D4 (all r < -0.5, all P < 0.001). Positive correlations were observed for IVPDa-D1, IVPDa-D2 with E, E/e’, and LAVI (0.3

VFM parameters demonstrate a certain correlation with conventional diastolic function parameters and show promise in assessing left ventricular diastolic function. Furthermore, VFM parameters exhibit greater sensitivity to early diastolic function changes, suggesting that VFM could be a novel method for evaluating differences in left ventricular diastolic function in DLBCL patients before and after chemotherapy.

This study aimed to investigate the impact of puerarin early intervention on growth parameters and Hepatic Fat Signal Fraction (HFF) quantification in Intrauterine Growth Restricted (IUGR) rats through Proton Magnetic resonance spectroscopy (¹H-MRS).

Pregnant rats were divided into three groups: control, IUGR with puerarin treatment, and IUGR without treatment. The treatment and non-treatment groups were received a low-protein diet during pregnancy, while the control group received a normal diet. After birth, pups in the treatment group received a unilateral intraperitoneal injection of 50 mg/kg/d puerarin. Male rats were evaluated at 3,8 and 12 weeks, including measurements of weight, body length and waist circumference and body mass index (BMI). Conventional magnetic resonance imaging and ¹H-MRS were conducted using a 3.0 T whole-body MR scanner.

Newborn pups in the treatment and non-treatment groups showed significantly lower body weight, BMI, and body length at 3 weeks compared to the control group. However, there were no significant differences in HFF and waist circumference between the three groups at 3 weeks. At 8 and 12 weeks post-delivery, significant differences in body weight, BMI, waist circumference were observed in newborn pups of IUGR non-treatment rats compared to the control group. In contrast, there were no significant differences in body weight, BMI, waist circumference between the treatment group and the control group at 8 and 12 weeks. Moreover, the treatment group exhibited notably higher HFF compared to the control group at both time points. At 12 weeks post-birth, a significant difference in HFF was observed between the IUGR non-treatment and treatment groups, although no significant difference was found at 8 weeks.

Early intervention with puerarin following birth has a significant impact on liver fat content and may potentially reduce adult obesity among IUGR rats.

Transorbital Ultrasonography (TOS) is a promising imaging technology that can be used to characterize the structures of the optic nerve and the potential alterations that may occur in those structures as a result of an increase in intracranial pressure (ICP) or the presence of other disorders such as multiple sclerosis (MS) and hydrocephalus.

In this paper, the primary objective is to develop a fully automated system that is capable of segmenting and calculating the diameters of structures that are associated with the optic nerve in TOS images. These structures include the optic nerve diameter sheath (ONSD) and the optic nerve diameter (OND).

A fully convolutional neural network (FCN) model that has been pre-trained serves as the foundation for the segmentation method. The method that was developed was utilized to collect 464 different photographs from 110 different people, and it was accomplished with the assistance of four distinct pieces of apparatus.

An examination was carried out to compare the outcomes of the automatic measurements with those of a manual operator. Both OND and ONSD have a typical inaccuracy of -0.12 0.32 mm and 0.14 0.58 mm, respectively, when compared to the operator. The Pearson correlation coefficient (PCC) for OND is 0.71, while the coefficient for ONSD is 0.64, showing that there is a positive link between the two measuring tools.

A conclusion may be drawn that the technique that was developed is automatic, and the average error (AE) that was reached for the ONSD measurement is compatible with the ranges of inter-operator variability that have been discovered in the literature.

This study introduces SkinLiTE, a lightweight supervised contrastive learning model tailored to enhance the detection and typification of skin lesions in dermoscopic images. The core of SkinLiTE lies in its unique integration of supervised and contrastive learning approaches, which leverages labeled data to learn generalizable representations. This approach is particularly adept at handling the challenge of complexities and imbalances inherent in skin lesion datasets.

The methodology encompasses a two-phase learning process. In the first phase, SkinLiTE utilizes an encoder network and a projection head to transform and project dermoscopic images into a feature space where contrastive loss is applied, focusing on minimizing intra-class variations while maximizing inter-class differences. The second phase freezes the encoder's weights, leveraging the learned representations for classification through a series of dense and dropout layers. The model was evaluated using three datasets from Skin Cancer ISIC 2019-2020, covering a wide range of skin conditions.

SkinLiTE demonstrated superior performance across various metrics, including accuracy, AUC, and F1 scores, particularly when compared with traditional supervised learning models. Notably, SkinLiTE achieved an accuracy of 0.9087 using AugMix augmentation for binary classification of skin lesions. It also showed comparable results with the state-of-the-art approaches of ISIC challenge without relying on external data, underscoring its efficacy and efficiency. The results highlight the potential of SkinLiTE as a significant step forward in the field of dermatological AI, offering a robust, efficient, and accurate tool for skin lesion detection and classification. Its lightweight architecture and ability to handle imbalanced datasets make it particularly suited for integration into Internet of Medical Things environments, paving the way for enhanced remote patient monitoring and diagnostic capabilities.

This research contributes to the evolving landscape of AI in healthcare, demonstrating the impact of innovative learning methodologies in medical image analysis.

Radiomics can quantify pulmonary nodule characteristics non-invasively by applying advanced imaging feature algorithms. Radiomic textural features derived from Computed Tomography (CT) imaging are broadly used to predict benign and malignant pulmonary nodules. However, few studies have reported on the radiomics-based identification of nodular Pulmonary Cryptococcosis (PC).

This study aimed to evaluate the diagnostic and differential diagnostic value of radiomic features extracted from CT images for nodular PC.

This retrospective analysis included 44 patients with PC (29 males, 15 females), 58 with Tuberculosis (TB) (39 males, 19 females), and 60 with Lung Cancer (LC) (20 males, 40 females) confirmed pathologically. Models 1 (PC vs. non-PC), 2 (PC vs. TB), and 3 (PC vs. LC) were established using radiomic features. Models 4 (PC vs. TB) and 5 (PC vs. LC) were established based on radiomic and CT features.

Five radiomic features were predictive of PC vs. non-PC model, but accuracy and Area Under the Curve (AUC) were 0.49 and 0.472, respectively. In model 2 (PC vs. TB) involving six radiomic features, the accuracy and AUC were 0.80 and 0.815, respectively. Model 3 (PC vs. LC) with six radiomic features performed well, with AUC=0.806 and an accuracy of 0.76. Between the PC and TB groups, model 4 combining radiomics, distribution, and PI, showed AUC=0.870. In differentiating PC from LC, the combination of radiomics, distribution, PI, and RBNAV achieved AUC=0.926 and an accuracy of 0.90.

The prediction models based on radiomic features from CT images performed well in discriminating PC from TB and LC. The individualized prediction models combining radiomic and CT features achieved the best diagnostic performance.