Feature Robustness and Diagnostic Capabilities of Convolutional Neural Networks Against Radiomics Features in Computed Tomography Imaging

Development and validation of intracranial hypertension prediction models based on radiomic features in patients with traumatic brain injury: an exploratory study based on CENTER-TBI data

BACKGROUND

Head computed tomography (CT) is a routinely performed examination to assess the intracranial condition of patients with traumatic brain injury (TBI), and radiological findings can help to indicate the presence of intracranial hypertension. At present, the prediction of intracranial hypertension is mainly based on manual discrimination of imaging characteristics. The aim of our study was to establish a model to predict intracranial hypertension via fully automatic CT image segmentation, rigorous radiomic feature extraction and reliable model development and validation.

METHODS

Patients admitted to the intensive care unit (ICU) who underwent intracranial pressure (ICP) monitoring were included in our study. For the development cohort, we extracted data from the CENTER-TBI database and randomly divided the data into a training group and a test group. For the validation cohort, we extracted data from patients admitted to the Shanghai General Hospital. Patients whose initial recorded ICP value was greater than or equal to 20 mmHg were defined as having intracranial hypertension. Radiological features, including imaging characteristics and three categories of radiomic features, were extracted from the head CT. Feature selection was performed for all radiological findings. A morphological model was built on the basis of selected imaging characteristics. First-order, second-order and third-order models were built on the basis of selected radiomic features. A comprehensive model was built on the basis of all selected radiological findings. The performances of these five models were assessed by four classifiers, including logistic regression (LR), random forest (RF), multilayer perceptron (MLP), and extreme gradient boosting (XGB), from which the best classifier was selected. After the process of model training and external validation, we ultimately used the optimal classifier to generate a prediction model with greater predictive power and stability.

RESULTS

Five models were built, including a morphological model, first-order model, second-order model, third-order model and comprehensive model. The optimal classifier was the logistic regression (LR) classifier, with which the morphological, first-order, second-order, third-order and comprehensive models had AUCs of 0.75, 0.77, 0.76, 0.86, and 0.83 and F1 scores of 0.54, 0.73, 0.63, 0.72, and 0.75, respectively, in the external validation group.

CONCLUSIONS

We successfully established a model for predicting intracranial hypertension on the basis of radiomic features. This model may serve as an approach for intracranial hypertension prediction in TBI patients.

Radiomics models to predict bone marrow metastasis of neuroblastoma using CT

BACKGROUND

Bone marrow is the leading site for metastasis from neuroblastoma and affects the prognosis of patients with neuroblastoma. However, the accurate diagnosis of bone marrow metastasis is limited by the high spatial and temporal heterogeneity of neuroblastoma. Radiomics analysis has been applied in various cancers to build accurate diagnostic models but has not yet been applied to bone marrow metastasis of neuroblastoma.

METHODS

We retrospectively collected information from 187 patients pathologically diagnosed with neuroblastoma and divided them into training and validation sets in a ratio of 7:3. A total of 2632 radiomics features were retrieved from venous and arterial phases of contrast-enhanced computed tomography (CT), and nine machine learning approaches were used to build radiomics models, including multilayer perceptron (MLP), extreme gradient boosting, and random forest. We also constructed radiomics-clinical models that combined radiomics features with clinical predictors such as age, gender, ascites, and lymph gland metastasis. The performance of the models was evaluated with receiver operating characteristics (ROC) curves, calibration curves, and risk decile plots.

RESULTS

The MLP radiomics model yielded an area under the ROC curve (AUC) of 0.97 (95% confidence interval [CI]: 0.95-0.99) on the training set and 0.90 (95% CI: 0.82-0.95) on the validation set. The radiomics-clinical model using an MLP yielded an AUC of 0.93 (95% CI: 0.89-0.96) on the training set and 0.91 (95% CI: 0.85-0.97) on the validation set.

CONCLUSIONS

MLP-based radiomics and radiomics-clinical models can precisely predict bone marrow metastasis in patients with neuroblastoma.

A multimodal neural network with gradient blending improves predictions of survival and metastasis in sarcoma

The objective of this study is to develop a multimodal neural network (MMNN) model that analyzes clinical variables and MRI images of a soft tissue sarcoma (STS) patient, to predict overall survival and risk of distant metastases. We compare the performance of this MMNN to models based on clinical variables alone, radiomics models, and an unimodal neural network. We include patients aged 18 or older with biopsy-proven STS who underwent primary resection between January 1st, 2005, and December 31st, 2020 with complete outcome data and a pre-treatment MRI with both a T1 post-contrast sequence and a T2 fat-sat sequence available. A total of 9380 MRI slices containing sarcomas from 287 patients are available. Our MMNN accepts the entire 3D sarcoma volume from T1 and T2 MRIs and clinical variables. Gradient blending allows the clinical and image sub-networks to optimally converge without overfitting. Heat maps were generated to visualize the salient image features. Our MMNN outperformed all other models in predicting overall survival and the risk of distant metastases. The C-Index of our MMNN for overall survival is 0.77 and the C-Index for risk of distant metastases is 0.70. The provided heat maps demonstrate areas of sarcomas deemed most salient for predictions. Our multimodal neural network with gradient blending improves predictions of overall survival and risk of distant metastases in patients with soft tissue sarcoma. Future work enabling accurate subtype-specific predictions will likely utilize similar end-to-end multimodal neural network architecture and require prospective curation of high-quality data, the inclusion of genomic data, and the involvement of multiple centers through federated learning.

Deep Learning and Multimodal Artificial Intelligence in Orthopaedic Surgery

This review article focuses on the applications of deep learning with neural networks and multimodal neural networks in the orthopaedic domain. By providing practical examples of how artificial intelligence (AI) is being applied successfully in orthopaedic surgery, particularly in the realm of imaging data sets and the integration of clinical data, this study aims to provide orthopaedic surgeons with the necessary tools to not only evaluate existing literature but also to consider AI's potential in their own clinical or research pursuits. We first review standard deep neural networks which can analyze numerical clinical variables, then describe convolutional neural networks which can analyze image data, and then introduce multimodal AI models which analyze various types of different data. Then, we contrast these deep learning techniques with related but more limited techniques such as radiomics, describe how to interpret deep learning studies, and how to initiate such studies at your institution. Ultimately, by empowering orthopaedic surgeons with the knowledge and know-how of deep learning, this review aspires to facilitate the translation of research into clinical practice, thereby enhancing the efficacy and precision of real-world orthopaedic care for patients.

Comparative analysis of radiomics and deep-learning algorithms for survival prediction in hepatocellular carcinoma

To examine the comparative robustness of computed tomography (CT)-based conventional radiomics and deep-learning convolutional neural networks (CNN) to predict overall survival (OS) in HCC patients. Retrospectively, 114 HCC patients with pretherapeutic CT of the liver were randomized into a development (n = 85) and a validation (n = 29) cohort, including patients of all tumor stages and several applied therapies. In addition to clinical parameters, image annotations of the liver parenchyma and of tumor findings on CT were available. Cox-regression based on radiomics features and CNN models were established and combined with clinical parameters to predict OS. Model performance was assessed using the concordance index (C-index). Log-rank tests were used to test model-based patient stratification into high/low-risk groups. The clinical Cox-regression model achieved the best validation performance for OS (C-index [95% confidence interval (CI)] 0.74 [0.57-0.86]) with a significant difference between the risk groups (p = 0.03). In image analysis, the CNN models (lowest C-index [CI] 0.63 [0.39-0.83]; highest C-index [CI] 0.71 [0.49-0.88]) were superior to the corresponding radiomics models (lowest C-index [CI] 0.51 [0.30-0.73]; highest C-index [CI] 0.66 [0.48-0.79]). A significant risk stratification was not possible (p > 0.05). Under clinical conditions, CNN-algorithms demonstrate superior prognostic potential to predict OS in HCC patients compared to conventional radiomics approaches and could therefore provide important information in the clinical setting, especially when clinical data is limited.

Deep Features from Pretrained Networks Do Not Outperform Hand-Crafted Features in Radiomics

In radiomics, utilizing features extracted from pretrained deep networks could result in models with a higher predictive performance than those relying on hand-crafted features. This study compared the predictive performance of models trained with either deep features, hand-crafted features, or a combination of these features in terms of the area under the receiver-operating characteristic curve (AUC) and other metrics. We trained models on ten radiological datasets using five feature selection methods and three classifiers. Our results indicate that models based on deep features did not show an improved AUC compared to those utilizing hand-crafted features (deep: AUC 0.775, hand-crafted: AUC 0.789; p = 0.28). Including morphological features alongside deep features led to overall improvements in prediction performance for all models (+0.02 gain in AUC; p < 0.001); however, the best model did not benefit from this (+0.003 gain in AUC; p = 0.57). Using all hand-crafted features in addition to the deep features resulted in a further overall improvement (+0.034 in AUC; p < 0.001), but only a minor improvement could be observed for the best model (deep: AUC 0.798, hand-crafted: AUC 0.789; p = 0.92). Furthermore, our results show that models based on deep features extracted from networks pretrained on medical data have no advantage in predictive performance over models relying on features extracted from networks pretrained on ImageNet data. Our study contributes a benchmarking analysis of models trained on hand-crafted and deep features from pretrained networks across multiple datasets. It also provides a comprehensive understanding of their applicability and limitations in radiomics. Our study shows, in conclusion, that models based on features extracted from pretrained deep networks do not outperform models trained on hand-crafted ones.

Artificial intelligence-based radiomics in bone tumors: Technical advances and clinical application

Radiomics is the extraction of predefined mathematic features from medical images for predicting variables of clinical interest. Recent research has demonstrated that radiomics can be processed by artificial intelligence algorithms to reveal complex patterns and trends for diagnosis, and prediction of prognosis and response to treatment modalities in various types of cancer. Artificial intelligence tools can utilize radiological images to solve next-generation issues in clinical decision making. Bone tumors can be classified as primary and secondary (metastatic) tumors. Osteosarcoma, Ewing sarcoma, and chondrosarcoma are the dominating primary tumors of bone. The development of bone tumor model systems and relevant research, and the assessment of novel treatment methods are ongoing to improve clinical outcomes, notably for patients with metastases. Artificial intelligence and radiomics have been utilized in almost full spectrum of clinical care of bone tumors. Radiomics models have achieved excellent performance in the diagnosis and grading of bone tumors. Furthermore, the models enable to predict overall survival, metastases, and recurrence. Radiomics features have exhibited promise in assisting therapeutic planning and evaluation, especially neoadjuvant chemotherapy. This review provides an overview of the evolution and opportunities for artificial intelligence in imaging, with a focus on hand-crafted features and deep learning-based radiomics approaches. We summarize the current application of artificial intelligence-based radiomics both in primary and metastatic bone tumors, and discuss the limitations and future opportunities of artificial intelligence-based radiomics in this field. In the era of personalized medicine, our in-depth understanding of emerging artificial intelligence-based radiomics approaches will bring innovative solutions to bone tumors and achieve clinical application.

Transfer-Learning Deep Radiomics and Hand-Crafted Radiomics for Classifying Lymph Nodes from Contrast-Enhanced Computed Tomography in Lung Cancer

OBJECTIVES

Positron emission tomography (PET) is currently considered the non-invasive reference standard for lymph node (N-)staging in lung cancer. However, not all patients can undergo this diagnostic procedure due to high costs, limited availability, and additional radiation exposure. The purpose of this study was to predict the PET result from traditional contrast-enhanced computed tomography (CT) and to test different feature extraction strategies.

METHODS

In this study, 100 lung cancer patients underwent a contrast-enhanced 18F-fluorodeoxyglucose (FDG) PET/CT scan between August 2012 and December 2019. We trained machine learning models to predict FDG uptake in the subsequent PET scan. Model inputs were composed of (i) traditional "hand-crafted" radiomics features from the segmented lymph nodes, (ii) deep features derived from a pretrained EfficientNet-CNN, and (iii) a hybrid approach combining (i) and (ii).

RESULTS

In total, 2734 lymph nodes [555 (20.3%) PET-positive] from 100 patients [49% female; mean age 65, SD: 14] with lung cancer (60% adenocarcinoma, 21% plate epithelial carcinoma, 8% small-cell lung cancer) were included in this study. The area under the receiver operating characteristic curve (AUC) ranged from 0.79 to 0.87, and the scaled Brier score (SBS) ranged from 16 to 36%. The random forest model (iii) yielded the best results [AUC 0.871 (0.865-0.878), SBS 35.8 (34.2-37.2)] and had significantly higher model performance than both approaches alone (AUC: p < 0.001, z = 8.8 and z = 22.4; SBS: p < 0.001, z = 11.4 and z = 26.6, against (i) and (ii), respectively).

CONCLUSION

Both traditional radiomics features and transfer-learning deep radiomics features provide relevant and complementary information for non-invasive N-staging in lung cancer.

РОЛЬ АЛГОРИТМОВ ИСКУССТВЕННОГО ИНТЕЛЛЕКТА В ДИАГНОСТИКЕ

Hepatosellülyar karsinoma (HSK) ən çox yayılan bədxassəli törəmələr arasında beşinci yeri tutur və dünyada xərçənglə əlaqəli ölümün üçüncü ən çox yayılmış səbəbidir. Süni intellekt (Sİ) sürətlə artan maraq sahəsidir. Müəlliflər HSK-ın diaqnostikasında və qiymətləndirilməsində Sİ-nin tətbiqi barədə məlumat verən məqalələri araşdırmışlar. Bu məqsədlə 27 məqalə təhlil edilmişdir. Təhlil edilmiş məqalələrdən KT görüntülərinin tədqiqinə dair 19 məqalədə (41,30%), USQ görüntülərinin öyrənilməsini əks etdirən 20 (43,47%) və MRT görüntülərindən bəhs edən 7 məqalədə (15,21%) müxtəlif Sİ alqoritmləri qəbul edilmişdir. Heç bir məqalədə PET və rentgen texnologiyasında süni intellektin istifadəsi müzakirə edilməyib. Sistematik yanaşma göstərmişdir ki, HSK-nin diaqnostikası və qiymətləndirilməsi üzrə əvvəlki işlərdə USQ, KT və MRT istifadə edilərək ənənəvi şərhin maşın öyrənməsi ilə müqayisəliliyi qiymətləndirilmişdir. Təhlillərimizdə görüntüləmə üsullarının istifadəsi HSK diaqnostikası üçün tibbi görüntüləmənin faydalılığını və təkamülünü əks etdirir. Bundan əlavə, nəticələrimiz lazımsız təkrarlanmanı və resursların israfını minimuma endirmək üçün birgə məlumat bazasında məlumat mübadiləsinə qaçılmaz ehtiyac olduğunu vurğulayır.

Гепатоцеллюлярная карцинома является пятым по распространенности злокачественным новообразованием и третьей по частоте причиной смерти от рака во всём мире. Искусственный интеллект — это быстрорастущая область интересов. Авторами были рассмотрены статьи, в которых сообщается о применении алгоритмов ИИ в диагностике и оценке ГЦК. Для этого проанализированы 27 статей. В проанализированных статьях в 19 статьях, посвящённых КТ-изображениям (41,30%), в 20 статьях, посвящённых изображениям УЗИ (43,47%), и в 7 статьях, посвящённым МРТ-изображениям (15,21%), использовали разные алгоритмы ИИ. Ни в одной статье не обсуждалось использование искусственного интеллекта в ПЭТ и рентгеновские технологии. Системный подход показал, что предыдущая работа по диагностике и оценке ГЦК оценивала сопоставимость традиционной интерпретации с машинным обучением с использованием УЗИ, КТ и МРТ. Использование методов визуализации в проведенном анализе отражает полезность и эволюцию медицинской визуализации для диагностики ГЦК. Кроме того, результаты поиска литературы подчёркивают острую необходимость совместного использования данных в совместных базах данных, чтобы свести к минимуму ненужное дублирование и растрату ресурсов.

Hepatocellular carcinoma (HCC) is the fifth most common malignancy and the third leading cause of cancer death worldwide. Artificial intelligence (AI) is a rapidly growing area of interest. We have reviewed articles reporting the application of AI algorithms in the diagnosis and evaluation of HCC. To do this, we analyzed 27 articles. In the analyzed articles, 19 articles on CT images (41.30%), 20 articles on ultrasound images (43.47%), and 7 articles on MRI images (15.21%) used different AI algorithms. None of the articles discussed the use of artificial intelligence in PET and X-ray technologies. Our systematic approach showed that previous work on the diagnosis and evaluation of HCC assessed the comparability of traditional interpretation with machine learning using ultrasound, CT, and MRI. The use of imaging modalities in our analysis reflects the usefulness and evolution of medical imaging for diagnosing HCC. In addition, our results highlight the critical need to share data across collaborative databases to minimize unnecessary duplication and waste of resources.

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.

Radiomics: A radiological evidence-based artificial intelligence technique to facilitate personalized precision medicine in hepatocellular carcinoma

The high postoperative recurrence rates in hepatocellular carcinoma (HCC) remain a major hurdle in its management. Appropriate staging and treatment selection may alleviate the extent of fatal recurrence. However, effective methods to preoperatively evaluate pathophysiologic and molecular characteristics of HCC are lacking. Imaging plays a central role in HCC diagnosis and stratification due to the non-invasive diagnostic criteria. Vast and crucial information is hidden within image data. Other than providing a morphological sketch for lesion diagnosis, imaging could provide new insights to describe the pathophysiological and genetic landscape of HCC. Radiomics aims to facilitate diagnosis and prognosis of HCC using artificial intelligence techniques to harness the immense information contained in medical images. Radiomics produces a set of archetypal and robust imaging features that are correlated to key pathological or molecular biomarkers to preoperatively risk-stratify HCC patients. Inferred with outcome data, comprehensive combination of radiomic, clinical and/or multi-omics data could also improve direct prediction of response to treatment and prognosis. The evolution of radiomics is changing our understanding of personalized precision medicine in HCC management. Herein, we review the key techniques and clinical applications in HCC radiomics and discuss current limitations and future opportunities to improve clinical decision making.

Artificial Intelligence in the Diagnosis of Hepatocellular Carcinoma: A Systematic Review

Hepatocellular carcinoma ranks fifth amongst the most common malignancies and is the third most common cause of cancer-related death globally. Artificial Intelligence is a rapidly growing field of interest. Following the PRISMA reporting guidelines, we conducted a systematic review to retrieve articles reporting the application of AI in HCC detection and characterization. A total of 27 articles were included and analyzed with our composite score for the evaluation of the quality of the publications. The contingency table reported a statistically significant constant improvement over the years of the total quality score (p = 0.004). Different AI methods have been adopted in the included articles correlated with 19 articles studying CT (41.30%), 20 studying US (43.47%), and 7 studying MRI (15.21%). No article has discussed the use of artificial intelligence in PET and X-ray technology. Our systematic approach has shown that previous works in HCC detection and characterization have assessed the comparability of conventional interpretation with machine learning using US, CT, and MRI. The distribution of the imaging techniques in our analysis reflects the usefulness and evolution of medical imaging for the diagnosis of HCC. Moreover, our results highlight an imminent need for data sharing in collaborative data repositories to minimize unnecessary repetition and wastage of resources.

Evaluation of the dependence of radiomic features on the machine learning model

BACKGROUND

In radiomic studies, several models are often trained with different combinations of feature selection methods and classifiers. The features of the best model are usually considered relevant to the problem, and they represent potential biomarkers. Features selected from statistically similarly performing models are generally not studied. To understand the degree to which the selected features of these statistically similar models differ, 14 publicly available datasets, 8 feature selection methods, and 8 classifiers were used in this retrospective study. For each combination of feature selection and classifier, a model was trained, and its performance was measured with AUC-ROC. The best-performing model was compared to other models using a DeLong test. Models that were statistically similar were compared in terms of their selected features.

RESULTS

Approximately 57% of all models analyzed were statistically similar to the best-performing model. Feature selection methods were, in general, relatively unstable (0.58; range 0.35-0.84). The features selected by different models varied largely (0.19; range 0.02-0.42), although the selected features themselves were highly correlated (0.71; range 0.4-0.92).

CONCLUSIONS

Feature relevance in radiomics strongly depends on the model used, and statistically similar models will generally identify different features as relevant. Considering features selected by a single model is misleading, and it is often not possible to directly determine whether such features are candidate biomarkers.