Dynamic contrast-enhanced computed tomography diagnosis of primary liver cancers using transfer learning of pretrained convolutional neural networks: Is registration of multiphasic images necessary?

Comparison of machine learning models and CEUS LI-RADS in differentiation of hepatic carcinoma and liver metastases in patients at risk of both hepatitis and extrahepatic malignancy

BACKGROUND

CEUS LI-RADS (Contrast Enhanced Ultrasound Liver Imaging Reporting and Data System) has good diagnostic efficacy for differentiating hepatic carcinoma (HCC) from solid malignant tumors. However, it can be problematic in patients with both chronic hepatitis B and extrahepatic primary malignancy. We explored the diagnostic performance of LI-RADS criteria and CEUS-based machine learning (ML) models in such patients.

METHODS

Consecutive patients with hepatitis and HCC or liver metastasis (LM) who were included in a multicenter liver cancer database between July 2017 and January 2022 were enrolled in this study. LI-RADS and enhancement features were assessed in a training cohort, and ML models were constructed using gradient boosting, random forest, and generalized linear models. The diagnostic performance of the ML models was compared with LI-RADS in a validation cohort of patients with both chronic hepatitis and extrahepatic malignancy.

RESULTS

The mild washout time was adjusted to 54 s from 60 s, increasing accuracy from 76.8 to 79.4%. Through feature screening, washout type II, rim enhancement and unclear border were identified as the top three predictor variables. Using LI-RADS to differentiate HCC from LM, the sensitivity, specificity, and AUC were 68.2%, 88.6%, and 0.784, respectively. In comparison, the random forest and generalized linear model both showed significantly higher sensitivity and accuracy than LI-RADS (0.83 vs. 0.784; all P < 0.001).

CONCLUSIONS

Compared with LI-RADS, the random forest and generalized linear model had higher accuracy for differentiating HCC from LM in patients with chronic hepatitis B and extrahepatic malignancy.

Clinical applications of artificial intelligence in liver imaging

This review outlines the current status and challenges of the clinical applications of artificial intelligence in liver imaging using computed tomography or magnetic resonance imaging based on a topic analysis of PubMed search results using latent Dirichlet allocation. LDA revealed that "segmentation," "hepatocellular carcinoma and radiomics," "metastasis," "fibrosis," and "reconstruction" were current main topic keywords. Automatic liver segmentation technology using deep learning is beginning to assume new clinical significance as part of whole-body composition analysis. It has also been applied to the screening of large populations and the acquisition of training data for machine learning models and has resulted in the development of imaging biomarkers that have a significant impact on important clinical issues, such as the estimation of liver fibrosis, recurrence, and prognosis of malignant tumors. Deep learning reconstruction is expanding as a new technological clinical application of artificial intelligence and has shown results in reducing contrast and radiation doses. However, there is much missing evidence, such as external validation of machine learning models and the evaluation of the diagnostic performance of specific diseases using deep learning reconstruction, suggesting that the clinical application of these technologies is still in development.

Revisiting artificial intelligence diagnosis of hepatocellular carcinoma with DIKWH framework

Hepatocellular carcinoma (HCC) is the most common type of liver cancer with a high morbidity and fatality rate. Traditional diagnostic methods for HCC are primarily based on clinical presentation, imaging features, and histopathology. With the rapid development of artificial intelligence (AI), which is increasingly used in the diagnosis, treatment, and prognosis prediction of HCC, an automated approach to HCC status classification is promising. AI integrates labeled clinical data, trains on new data of the same type, and performs interpretation tasks. Several studies have shown that AI techniques can help clinicians and radiologists be more efficient and reduce the misdiagnosis rate. However, the coverage of AI technologies leads to difficulty in which the type of AI technology is preferred to choose for a given problem and situation. Solving this concern, it can significantly reduce the time required to determine the required healthcare approach and provide more precise and personalized solutions for different problems. In our review of research work, we summarize existing research works, compare and classify the main results of these according to the specified data, information, knowledge, wisdom (DIKW) framework.

Dual segmentation models for poorly and well-differentiated hepatocellular carcinoma using two-step transfer deep learning on dynamic contrast-enhanced CT images

The aim of this study was to develop dual segmentation models for poorly and well-differentiated hepatocellular carcinoma (HCC), using two-step transfer learning (TSTL) based on dynamic contrast-enhanced (DCE) computed tomography (CT) images. From 2013 to 2019, DCE-CT images of 128 patients with 80 poorly differentiated and 48 well-differentiated HCCs were selected at our hospital. In the first transfer learning (TL) step, a pre-trained segmentation model with 192 CT images of lung cancer patients was retrained as a poorly differentiated HCC model. In the second TL step, a well-differentiated HCC model was built from a poorly differentiated HCC model. The average three-dimensional Dice's similarity coefficient (3D-DSC) and 95th-percentile of the Hausdorff distance (95% HD) were mainly employed to evaluate the segmentation accuracy, based on a nested fourfold cross-validation test. The DSC denotes the degree of regional similarity between the HCC reference regions and the regions estimated using the proposed models. The 95% HD is defined as the 95th-percentile of the maximum measures of how far two subsets of a metric space are from each other. The average 3D-DSC and 95% HD were 0.849 ± 0.078 and 1.98 ± 0.71 mm, respectively, for poorly differentiated HCC regions, and 0.811 ± 0.089 and 2.01 ± 0.84 mm, respectively, for well-differentiated HCC regions. The average 3D-DSC for both regions was 1.2 times superior to that calculated without the TSTL. The proposed model using TSTL from the lung cancer dataset showed the potential to segment poorly and well-differentiated HCC regions on DCE-CT images.

Detecting the presence of supernumerary teeth during the early mixed dentition stage using deep learning algorithms: A pilot study

BACKGROUND

Supernumerary teeth are a common anomaly and are frequently observed in paediatric patients. To prevent or minimize complications, early diagnosis and treatment is ideal in children with supernumerary teeth.

AIM

This study aimed to apply convolutional neural network (CNN)-based deep learning to detect the presence of supernumerary teeth in children during the early mixed dentition stage.

DESIGN

Three CNN models, AlexNet, VGG16-TL, and InceptionV3-TL, were employed in this study. A total of 220 panoramic radiographs (from children aged 6 years 0 months to 9 years 6 months) including supernumerary teeth (cases, n = 120) or no anomalies (controls, n = 100) were retrospectively analyzed. The CNN performances were assessed according to accuracy, sensitivity, specificity, receiver operating characteristic (ROC) curves, and area under the ROC curves for a cross-validation test dataset.

RESULTS

The VGG16-TL model had the highest performance according to accuracy, sensitivity, specificity, and area under the ROC curve, but the other models had similar performance.

CONCLUSION

CNN-based deep learning is a promising approach for detecting the presence of supernumerary teeth during the early mixed dentition stage.

A deep learning-based approach for the diagnosis of adrenal adenoma: a new trial using CT

OBJECTIVE

To develop and validate deep convolutional neural network (DCNN) models for the diagnosis of adrenal adenoma (AA) using CT.

METHODS

This retrospective study enrolled 112 patients who underwent abdominal CT (non-contrast, early, and delayed phases) with 107 adrenal lesions (83 AAs and 24 non-AAs) confirmed pathologically and with 8 lesions confirmed by follow-up as metastatic carcinomas. Three patients had adrenal lesions on both sides. We constructed six DCNN models from six types of input images for comparison: non-contrast images only (Model A), delayed phase images only (Model B), three phasic images merged into a 3-channel (Model C), relative washout rate (RWR) image maps only (Model D), non-contrast and RWR maps merged into a 2-channel (Model E), and delayed phase and RWR maps merged into a 2-channel (Model F). These input images were prepared manually with cropping and registration of CT images. Each DCNN model with six convolutional layers was trained with data augmentation and hyperparameter tuning. The optimal threshold values for binary classification were determined from the receiver-operating characteristic curve analyses. We adopted the nested cross-validation method, in which the outer fivefold cross-validation was used to assess the diagnostic performance of the models and the inner fivefold cross-validation was used to tune hyperparameters of the models.

RESULTS

The areas under the curve with 95% confidence intervals of Models A-F were 0.94 [0.90, 0.98], 0.80 [0.69, 0.89], 0.97 [0.94, 1.00], 0.92 [0.85, 0.97], 0.99 [0.97, 1.00] and 0.94 [0.86, 0.99], respectively. Model E showed high area under the curve greater than 0.95.

CONCLUSION

DCNN models may be a useful tool for the diagnosis of AA using CT.

ADVANCES IN KNOWLEDGE

The current study demonstrates a deep learning-based approach could differentiate adrenal adenoma from non-adenoma using multiphasic CT.

Transfer learning for medical image classification: a literature review

BACKGROUND

Transfer learning (TL) with convolutional neural networks aims to improve performances on a new task by leveraging the knowledge of similar tasks learned in advance. It has made a major contribution to medical image analysis as it overcomes the data scarcity problem as well as it saves time and hardware resources. However, transfer learning has been arbitrarily configured in the majority of studies. This review paper attempts to provide guidance for selecting a model and TL approaches for the medical image classification task.

METHODS

425 peer-reviewed articles were retrieved from two databases, PubMed and Web of Science, published in English, up until December 31, 2020. Articles were assessed by two independent reviewers, with the aid of a third reviewer in the case of discrepancies. We followed the PRISMA guidelines for the paper selection and 121 studies were regarded as eligible for the scope of this review. We investigated articles focused on selecting backbone models and TL approaches including feature extractor, feature extractor hybrid, fine-tuning and fine-tuning from scratch.

RESULTS

The majority of studies (n = 57) empirically evaluated multiple models followed by deep models (n = 33) and shallow (n = 24) models. Inception, one of the deep models, was the most employed in literature (n = 26). With respect to the TL, the majority of studies (n = 46) empirically benchmarked multiple approaches to identify the optimal configuration. The rest of the studies applied only a single approach for which feature extractor (n = 38) and fine-tuning from scratch (n = 27) were the two most favored approaches. Only a few studies applied feature extractor hybrid (n = 7) and fine-tuning (n = 3) with pretrained models.

CONCLUSION

The investigated studies demonstrated the efficacy of transfer learning despite the data scarcity. We encourage data scientists and practitioners to use deep models (e.g. ResNet or Inception) as feature extractors, which can save computational costs and time without degrading the predictive power.

The application of artificial intelligence in hepatology: A systematic review

The integration of human and artificial intelligence (AI) in medicine has only recently begun but it has already become obvious that intelligent systems can dramatically improve the management of liver diseases. Big data made it possible to envisage transformative developments of the use of AI for diagnosing, predicting prognosis and treating liver diseases, but there is still a lot of work to do. If we want to achieve the 21^st century digital revolution, there is an urgent need for specific national and international rules, and to adhere to bioethical parameters when collecting data. Avoiding misleading results is essential for the effective use of AI. A crucial question is whether it is possible to sustain, technically and morally, the process of integration between man and machine. We present a systematic review on the applications of AI to hepatology, highlighting the current challenges and crucial issues related to the use of such technologies.

State of the Art in Artificial Intelligence and Radiomics in Hepatocellular Carcinoma

The most common liver malignancy is hepatocellular carcinoma (HCC), which is also associated with high mortality. Often HCC develops in a chronic liver disease setting, and early diagnosis as well as accurate screening of high-risk patients is crucial for appropriate and effective management of these patients. While imaging characteristics of HCC are well-defined in the diagnostic phase, challenging cases still occur, and current prognostic and predictive models are limited in their accuracy. Radiomics and machine learning (ML) offer new tools to address these issues and may lead to scientific breakthroughs with the potential to impact clinical practice and improve patient outcomes. In this review, we will present an overview of these technologies in the setting of HCC imaging across different modalities and a range of applications. These include lesion segmentation, diagnosis, prognostic modeling and prediction of treatment response. Finally, limitations preventing clinical application of radiomics and ML at the present time are discussed, together with necessary future developments to bring the field forward and outside of a purely academic endeavor.