Focal liver lesions segmentation and classification in nonenhanced T2-weighted MRI

Focal liver lesion diagnosis with deep learning and multistage CT imaging

Diagnosing liver lesions is crucial for treatment choices and patient outcomes. This study develops an automatic diagnosis system for liver lesions using multiphase enhanced computed tomography (CT). A total of 4039 patients from six data centers are enrolled to develop Liver Lesion Network (LiLNet). LiLNet identifies focal liver lesions, including hepatocellular carcinoma (HCC), intrahepatic cholangiocarcinoma (ICC), metastatic tumors (MET), focal nodular hyperplasia (FNH), hemangioma (HEM), and cysts (CYST). Validated in four external centers and clinically verified in two hospitals, LiLNet achieves an accuracy (ACC) of 94.7% and an area under the curve (AUC) of 97.2% for benign and malignant tumors. For HCC, ICC, and MET, the ACC is 88.7% with an AUC of 95.6%. For FNH, HEM, and CYST, the ACC is 88.6% with an AUC of 95.9%. LiLNet can aid in clinical diagnosis, especially in regions with a shortage of radiologists.

Diagnostic Performance of Artificial Intelligence in Detection of Hepatocellular Carcinoma: A Meta-analysis

Due to the increasing interest in the use of artificial intelligence (AI) algorithms in hepatocellular carcinoma detection, we performed a systematic review and meta-analysis to pool the data on diagnostic performance metrics of AI and to compare them with clinicians' performance. A search in PubMed and Scopus was performed in January 2024 to find studies that evaluated and/or validated an AI algorithm for the detection of HCC. We performed a meta-analysis to pool the data on the metrics of diagnostic performance. Subgroup analysis based on the modality of imaging and meta-regression based on multiple parameters were performed to find potential sources of heterogeneity. The risk of bias was assessed using Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) and Prediction Model Study Risk of Bias Assessment Tool (PROBAST) reporting guidelines. Out of 3177 studies screened, 44 eligible studies were included. The pooled sensitivity and specificity for internally validated AI algorithms were 84% (95% CI: 81,87) and 92% (95% CI: 90,94), respectively. Externally validated AI algorithms had a pooled sensitivity of 85% (95% CI: 78,89) and specificity of 84% (95% CI: 72,91). When clinicians were internally validated, their pooled sensitivity was 70% (95% CI: 60,78), while their pooled specificity was 85% (95% CI: 77,90). This study implies that AI can perform as a diagnostic supplement for clinicians and radiologists by screening images and highlighting regions of interest, thus improving workflow.

Radiomics-based distinction of small (≤2 cm) hepatocellular carcinoma and precancerous lesions based on unenhanced MRI

AIM

To assess the feasibility of a radiomics model based on unenhanced magnetic resonance imaging (MRI) to differentiate small hepatocellular carcinoma (S-HCC) (≤2 cm) and pre-hepatocellular carcinoma (Pre-HCC).

MATERIALS AND METHODS

One hundred and fourteen histopathologically confirmed 114 hepatic nodules were analysed retrospectively. All patients had undergone MRI before surgery using a 3 T MRI system. Each nodule was segmented on unenhanced MRI sequences (T1-weighted imaging [T1] and T2WI with fat-suppression [FS-T2]). Radiomics features were extracted and the optimal features were selected using the least absolute shrinkage and selection operator (LASSO). The support vector machine (SVM) was used to establish the radiomics model. One abdominal radiologist performed the conventional qualitative analysis for classification of S-HCC and Pre-HCC. The diagnostic performances of the radiomics and radiologist models were evaluated using receiver operating characteristic (ROC) analysis.

RESULT

Radiomics features (n=1,223) were extracted from each sequence and the optimal features were selected from T1, FS-T2, and T1+FS-T2 to construct the radiomics models. The radiomics model based on T1+FS-T2 showed the best performance among the three models, with areas under the ROC curves (AUCs) of 0.95 (95 % confidence interval [CI], 0.875-0.986) and 0.942 (95 % CI, 0.775-0.985), accuracies of 86 % and 88.5 %, sensitivities of 94.12 % and 100 %, and specificities of 85.48 % and 85.19 %, respectively. The radiomics model on FS-T2 showed better performance on a single sequence than that of the T1-based model. The diagnostic performance for the radiomic model was significantly higher than that for the radiologist (AUC = 0.518, p<0.05).

CONCLUSION

This study suggested that a radiomics model based on unenhanced MRI may serve as a feasible and non-invasive tool to classify S-HCC and Pre-HCC.

Automated Assessment of T2-Weighted MRI to Differentiate Malignant and Benign Primary Solid Liver Lesions in Noncirrhotic Livers Using Radiomics

RATIONALE AND OBJECTIVES

Distinguishing malignant from benign liver lesions based on magnetic resonance imaging (MRI) is an important but often challenging task, especially in noncirrhotic livers. We developed and externally validated a radiomics model to quantitatively assess T2-weighted MRI to distinguish the most common malignant and benign primary solid liver lesions in noncirrhotic livers.

MATERIALS AND METHODS

Data sets were retrospectively collected from three tertiary referral centers (A, B, and C) between 2002 and 2018. Patients with malignant (hepatocellular carcinoma and intrahepatic cholangiocarcinoma) and benign (hepatocellular adenoma and focal nodular hyperplasia) lesions were included. A radiomics model based on T2-weighted MRI was developed in data set A using a combination of machine learning approaches. The model was internally evaluated on data set A through cross-validation, externally validated on data sets B and C, and compared to visual scoring of two experienced abdominal radiologists on data set C.

RESULTS

The overall data set included 486 patients (A: 187, B: 98, and C: 201). The radiomics model had a mean area under the curve (AUC) of 0.78 upon internal validation on data set A and a similar AUC in external validation (B: 0.74 and C: 0.76). In data set C, the two radiologists showed moderate agreement (Cohen's κ: 0.61) and achieved AUCs of 0.86 and 0.82.

CONCLUSION

Our T2-weighted MRI radiomics model shows potential for distinguishing malignant from benign primary solid liver lesions. External validation indicated that the model is generalizable despite substantial MRI acquisition protocol differences. Pending further optimization and generalization, this model may aid radiologists in improving the diagnostic workup of patients with liver lesions.

Focal Lesions of the Liver and Radiomics: What Do We Know?

Despite differences in pathological analysis, focal liver lesions are not always distinguishable in contrast-enhanced magnetic resonance imaging (MRI), contrast-enhanced computed tomography (CT), and positron emission tomography (PET). This issue can cause problems of differential diagnosis, treatment, and follow-up, especially in patients affected by HBV/HCV chronic liver disease or fatty liver disease. Radiomics is an innovative imaging approach that extracts and analyzes non-visible quantitative imaging features, supporting the radiologist in the most challenging differential diagnosis when the best-known methods are not conclusive. The purpose of this review is to evaluate the most significant CT and MRI texture features, which can discriminate between the main benign and malignant focal liver lesions and can be helpful to predict the response to pharmacological or surgical therapy and the patient's prognosis.

Predicting gene mutation status via artificial intelligence technologies based on multimodal integration (MMI) to advance precision oncology

Personalized treatment strategies for cancer frequently rely on the detection of genetic alterations which are determined by molecular biology assays. Historically, these processes typically required single-gene sequencing, next-generation sequencing, or visual inspection of histopathology slides by experienced pathologists in a clinical context. In the past decade, advances in artificial intelligence (AI) technologies have demonstrated remarkable potential in assisting physicians with accurate diagnosis of oncology image-recognition tasks. Meanwhile, AI techniques make it possible to integrate multimodal data such as radiology, histology, and genomics, providing critical guidance for the stratification of patients in the context of precision therapy. Given that the mutation detection is unaffordable and time-consuming for a considerable number of patients, predicting gene mutations based on routine clinical radiological scans or whole-slide images of tissue with AI-based methods has become a hot issue in actual clinical practice. In this review, we synthesized the general framework of multimodal integration (MMI) for molecular intelligent diagnostics beyond standard techniques. Then we summarized the emerging applications of AI in the prediction of mutational and molecular profiles of common cancers (lung, brain, breast, and other tumor types) pertaining to radiology and histology imaging. Furthermore, we concluded that there truly exist multiple challenges of AI techniques in the way of its real-world application in the medical field, including data curation, feature fusion, model interpretability, and practice regulations. Despite these challenges, we still prospect the clinical implementation of AI as a highly potential decision-support tool to aid oncologists in future cancer treatment management.

Single-breath-hold T2WI MRI with artificial intelligence-assisted technique in liver imaging: As compared with conventional respiratory-triggered T2WI

OBJECTIVE

To investigate the clinical feasibility of single-breath-hold T2-weighted (SBH-T2WI) liver MRI using Artificial Intelligence-assisted Compressed Sensing (ACS) technique in liver imaging as compared with conventional respiratory-triggered T2WI (RT-T2WI).

METHODS

From January 2021 to October 2021, 81 patients suspected of liver lesions were enrolled in this prospective study. The liver MRI was performed, including both RT-T2WI and ACS SBH-T2WI. Two experienced radiologists reviewed all images of each studied sequence, and recorded the lesion location and the largest diameter of the lesions. The image quality was quantitatively and qualitatively analyzed regarding signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), contrast ratio (CR), motion artifact, lesion conspicuity, liver boundary sharpness, and overall image quality. The lesion detection and image quality were compared between two sequences using the Chi-square test or Wilcoxon signed-rank test.

RESULTS

For lesion detection, 64 lesions were identified in 53 enrolled patients as the reference standard. The average size was 12.09 ± 7.4 mm for the benign lesions and 45.89 ± 22.01 mm for the malignant lesions. Of 64 liver lesions, ACS SBH-T2WI detected 60 lesions (93.8%), and RT-T2WI detected 58 lesions (90.6%). For image quality analysis, the motion artifact of ACS SBH-T2WI sequence was significantly reduced compared with the conventional RT-T2WI sequence (p < 0.05). The SNR, liver boundary sharpness, and overall image quality showed no statistical differences between the two sequences. While the CNR, CR, and lesion conspicuity of ACS SBH-T2WI were significantly better than RT-T2WI (all p < 0.05).

CONCLUSIONS

The SBH-T2WI with ACS technique showed promising performance as it provided significantly better image quality and lesion detectability with a considerable decrease in scanning time as compared with the conventional RT-T2WI.

Quantification of tumor response of cystic vestibular schwannoma to Gamma Knife radiosurgery by using artificial intelligence

OBJECTIVE

Gamma Knife radiosurgery (GKRS) is a common treatment modality for vestibular schwannoma (VS). The ability to predict treatment response is important in patient counseling and decision-making. The authors developed an algorithm that can automatically segment and differentiate cystic and solid tumor components of VS. They also investigated associations between the quantified radiological features of each component and tumor response after GKRS.

METHODS

This is a retrospective study comprising 323 patients with VS treated with GKRS. After preprocessing and generation of pretreatment T2-weighted (T2W)/T1-weighted with contrast (T1WC) images, the authors segmented VSs into cystic and solid components by using fuzzy C-means clustering. Quantitative radiological features of the entire tumor and its cystic and solid components were extracted. Linear regression models were implemented to correlate clinical variables and radiological features with the specific growth rate (SGR) of VS after GKRS.

RESULTS

A multivariable linear regression model of radiological features of the entire tumor demonstrated that a higher tumor mean signal intensity (SI) on T2W/T1WC images (p < 0.001) was associated with a lower SGR after GKRS. Similarly, a multivariable linear regression model using radiological features of cystic and solid tumor components demonstrated that a higher solid component mean SI (p = 0.039) and a higher cystic component mean SI (p = 0.004) on T2W/T1WC images were associated with a lower SGR after GKRS. A larger cystic component proportion (p = 0.085) was associated with a trend toward a lower SGR after GKRS.

CONCLUSIONS

Radiological features of VSs on pretreatment MRI that were quantified using fuzzy C-means were associated with tumor response after GKRS. Tumors with a higher tumor mean SI, a higher solid component mean SI, and a higher cystic component mean SI on T2W/T1WC images were more likely to regress in volume after GKRS. Those with a larger cystic component proportion also trended toward regression after GKRS. Further refinement of the algorithm may allow direct prediction of tumor response.

Tumor Size Measurements for Predicting Hodgkin’s and Non-Hodgkin’s Lymphoma Response to Treatment

The purpose of this study was to investigate the value of tumor size measurements as prognostic indicators of treatment outcome of Hodgkin’s and Non-Hodgkin’s lymphomas. ¹⁸F-FDG PET/CT exams before and after treatment were analyzed and metabolic and anatomic parameters—tumor maximum diameter, tumor maximum area, tumor volume, and maximum standardized uptake value (SUVmax)—were determined manually by an expert and automatically by a computer algorithm on PET and CT images. Results showed that the computer algorithm measurements did not correlate well with the expert’s standard maximum tumor diameter measurements but yielded better three dimensional metrics that could have clinical value. SUVmax was the strongest prognostic indicator of the clinical outcome after treatment, followed by the automated metabolic tumor volume measurements and the expert’s metabolic maximum diameter measurements. Anatomic tumor measurements had poor prognostic value. Metabolic volume measurements, although promising, did not significantly surpass current standard of practice, but automated measurements offered a significant advantage in terms of time and effort and minimized biases and variances in the PET measurements. Overall, considering the limited value of tumor size in predicting response to treatment, a paradigm shift seems necessary in order to identify robust prognostic markers in PET/CT; radiomics, namely combinations of anatomy, metabolism, and imaging, may be an option.

An update on radiomics techniques in primary liver cancers

BACKGROUND

Radiomics is a progressing field of research that deals with the extraction of quantitative metrics from medical images. Radiomic features detention indirectly tissue features such as heterogeneity and shape and can, alone or in combination with demographic, histological, genomic, or proteomic data, be used for decision support system in clinical setting.

METHODS

This article is a narrative review on Radiomics in Primary Liver Cancers. Particularly, limitations and future perspectives are discussed.

RESULTS

In oncology, assessment of tissue heterogeneity is of particular interest: genomic analysis have demonstrated that the degree of tumour heterogeneity is a prognostic determinant of survival and an obstacle to cancer control. Therefore, that Radiomics could support cancer detection, diagnosis, evaluation of prognosis and response to treatment, so as could supervise disease status in hepatocellular carcinoma (HCC) and Intrahepatic Cholangiocarcinoma (ICC) patients. Radiomic analysis is a convenient radiological image analysis technique used to support clinical decisions as it is able to provide prognostic and / or predictive biomarkers that allow a fast, objective and repeatable tool for disease monitoring.

CONCLUSIONS

Although several studies have shown that this analysis is very promising, there is little standardization and generalization of the results, which limits the translation of this method into the clinical context. The limitations are mainly related to the evaluation of data quality, repeatability, reproducibility, overfitting of the model.

TRIAL REGISTRATION

Not applicable.

IRIS—Intelligent Rapid Interactive Segmentation for Measuring Liver Cyst Volumes in Autosomal Dominant Polycystic Kidney Disease

Purpose: To develop and integrate interactive features with automatic methods for accurate liver cyst segmentation in patients with autosomal dominant polycystic kidney and liver disease (ADPKD). Methods: SmartClick and antiSmartClick were developed using iterative region growth guided by spatial and intensity connections and were integrated with automated level set (LS) segmentation and graphical user interface, forming an intelligent rapid interactive segmentation (IRIS) tool. IRIS and LS segmentations of liver cysts on T2 weighted images of patients with ADPKD (n = 17) were compared with manual segmentation as ground truth (GT). Results: Compared to manual GT, IRIS reduced the segmentation time by more than 10-fold. Compared to automated LS, IRIS reduced the mean liver cyst volume error from 42.22% to 13.44% (p < 0.001). IRIS segmentation agreed well with manual GT (79% dice score and 99% intraclass correlation coefficient). Conclusion: IRIS is feasible for fast, accurate liver cyst segmentation in patients with ADPKD.

Radiogenomics: Hunting Down Liver Metastasis in Colorectal Cancer Patients

Simple Summary

Colorectal cancer (CRC) is the third leading cause of cancer and the second most deadly tumor type in the world. The liver is the most common site of metastasis in CRC patients. The conversion of new imaging biomarkers into diagnostic, prognostic and predictive signatures, by the development of radiomics and radiogenomics, is an important potential new tool for the clinical management of cancer patients. In this review, we assess the knowledge gained from radiomics and radiogenomics studies in liver metastatic colorectal cancer patients and their possible use for early diagnosis, response assessment and treatment decisions.

Abstract

Radiomics is a developing new discipline that analyzes conventional medical images to extract quantifiable data that can be mined for new biomarkers that show the biology of pathological processes at microscopic levels. These data can be converted into image-based signatures to improve diagnostic, prognostic and predictive accuracy in cancer patients. The combination of radiomics and molecular data, called radiogenomics, has clear implications for cancer patients’ management. Though some studies have focused on radiogenomics signatures in hepatocellular carcinoma patients, only a few have examined colorectal cancer metastatic lesions in the liver. Moreover, the need to differentiate between liver lesions is fundamental for accurate diagnosis and treatment. In this review, we summarize the knowledge gained from radiomics and radiogenomics studies in hepatic metastatic colorectal cancer patients and their use in early diagnosis, response assessment and treatment decisions. We also investigate their value as possible prognostic biomarkers. In addition, the great potential of image mining to provide a comprehensive view of liver niche formation is examined thoroughly. Finally, new challenges and current limitations for the early detection of the liver premetastatic niche, based on radiomics and radiogenomics, are also discussed.

A deep-learning-based prediction model for the biodistribution of 90 Y microspheres in liver radioembolization

BACKGROUND

Radioembolization with ⁹⁰ Y microspheres is a treatment approach for liver cancer. Currently, employed dosimetric calculations exhibit low accuracy, lacking consideration of individual patient, and tissue characteristics.

PURPOSE

The purpose of the present study was to employ deep learning (DL) algorithms to differentiate patterns of pretreatment distribution of ^99m Tc-macroaggregated albumin on SPECT/CT and post-treatment distribution of ⁹⁰ Y microspheres on PET/CT and to accurately predict how the ⁹⁰ Y-microspheres will be distributed in the liver tissue by radioembolization therapy.

METHODS

Data for 19 patients with liver cancer (10 with hepatocellular carcinoma, 5 with intrahepatic cholangiocarcinoma, 4 with liver metastases) who underwent radioembolization with ⁹⁰ Y microspheres were used for the DL training. We developed a 3D voxel-based variation of the Pix2Pix model, which is a special type of conditional GANs designed to perform image-to-image translation. SPECT and CT scans along with the clinical target volume for each patient were used as inputs, as were their corresponding post-treatment PET scans. The real and predicted absorbed PET doses for the tumor and the whole liver area were compared. Our model was evaluated using the leave-one-out method, and the dose calculations were measured using a tissue-specific dose voxel kernel.

RESULTS

The comparison of the real and predicted PET/CT scans showed an average absorbed dose difference of 5.42% ± 19.31% and 0.44% ± 1.64% for the tumor and the liver area, respectively. The average absorbed dose differences were 7.98 ± 31.39 Gy and 0.03 ± 0.25 Gy for the tumor and the non-tumor liver parenchyma, respectively. Our model had a general tendency to underpredict the dosimetric results; the largest differences were noticed in one case, where the model underestimated the dose to the tumor area by 56.75% or 72.82 Gy.

CONCLUSIONS

The proposed deep-learning-based pretreatment planning method for liver radioembolization accurately predicted ⁹⁰ Y microsphere biodistribution. Its combination with a rapid and accurate 3D dosimetry method will render it clinically suitable and could improve patient-specific pretreatment planning.

Artificial intelligence assists identifying malignant versus benign liver lesions using contrast-enhanced ultrasound

BACKGROUND AND AIM

This study aims to construct a strategy that uses assistance from artificial intelligence (AI) to assist radiologists in the identification of malignant versus benign focal liver lesions (FLLs) using contrast-enhanced ultrasound (CEUS).

METHODS

A training set (patients = 363) and a testing set (patients = 211) were collected from our institute. On four-phase CEUS images in the training set, a composite deep learning architecture was trained and tuned for differentiating malignant and benign FLLs. In the test dataset, AI performance was evaluated by comparison with radiologists with varied levels of experience. Based on the comparison, an AI assistance strategy was constructed, and its usefulness in reducing CEUS interobserver heterogeneity was further tested.

RESULTS

In the test set, to identify malignant versus benign FLLs, AI achieved an area under the curve of 0.934 (95% CI 0.890-0.978) with an accuracy of 91.0%. Comparing with radiologists reviewing videos along with complementary patient information, AI outperformed residents (82.9-84.4%, P = 0.038) and matched the performance of experts (87.2-88.2%, P = 0.438). Due to the higher positive predictive value (PPV) (AI: 95.6% vs residents: 88.6-89.7%, P = 0.056), an AI strategy was defined to improve the malignant diagnosis. With the assistance of AI, radiologists exhibited a sensitivity improvement of 97.0-99.4% (P < 0.05) and an accuracy of 91.0-92.9% (P = 0.008-0.189), which was comparable with that of the experts (P = 0.904).

CONCLUSIONS

The CEUS-based AI strategy improved the performance of residents and reduced CEUS's interobserver heterogeneity in the differentiation of benign and malignant FLLs.

Single-breath-hold T2WI liver MRI with deep learning-based reconstruction: A clinical feasibility study in comparison to conventional multi-breath-hold T2WI liver MRI

OBJECTIVE

To investigate the clinical feasibility of single-breath-hold (SBH) T2-weighted (T2WI) liver MRI with deep learning-based reconstruction in the evaluation of image quality and lesion delineation, compared with conventional multi-breath-hold (MBH) T2WI.

METHODS

One hundred and fifty-two adult patients with suspected liver disease were prospectively enrolled. Two independent readers reviewed images acquired with conventional MBH-T2WI and SBH-T2WI at 3.0 T MR scanner. For image quality analyses, motion artifacts scores and boundary sharpness scores were compared using nonparametric Wilcoxon matched pairs tests between MBH-T2WI and SBH-T2WI. With the reference standard, 89 patients with 376 index lesions were included for lesion analyses. The lesion detection rates were compared by chi-square test, the lesion conspicuity scores and lesion-liver contrast ratio (CR) were compared using nonparametric Wilcoxon matched pairs tests between the two sequences.

RESULTS

For both readers, motion artifacts scores of SBH-T2WI were significantly lower than MBH-T2WI (P < 0.001). Boundary sharpness scores of SBH-T2WI were significantly higher than MBH-T2WI (P < 0.001). The lesion detection rates for SBH-T2WI were significantly higher than MBH-T2WI (P < 0.001); the differences of lesion detection rates between the two sequences were statistically significant for small (≤ 10 mm) liver lesions (P < 0.001), while not significant for larger (> 10 mm) lesions (P > 0.05). Lesion conspicuity scores were significantly higher on SBH-T2WI than MBH-T2WI in the entire cohort as well as in both stratified subgroups of lesions ≤10 mm and > 10 mm (P < 0.001 for all). CRs for focal liver lesions were also significantly higher with SBH-T2WI (P < 0.001).

CONCLUSION

The SBH-T2WI sequence with deep-learning based reconstruction showed promising performance as it provided significantly better image quality, lesion detectability, lesion conspicuity and contrast within a single breath-hold, compared with the conventional MBH-T2WI.

Radiomics in hepatic metastasis by colorectal cancer

Background

Radiomics is an emerging field and has a keen interest, especially in the oncology field. The process of a radiomics study consists of lesion segmentation, feature extraction, consistency analysis of features, feature selection, and model building. Manual segmentation is one of the most critical parts of radiomics. It can be time-consuming and suffers from variability in tumor delineation, which leads to the reproducibility problem of calculating parameters and assessing spatial tumor heterogeneity, particularly in large or multiple tumors. Radiomic features provides data on tumor phenotype as well as cancer microenvironment. Radiomics derived parameters, when associated with other pertinent data and correlated with outcomes data, can produce accurate robust evidence based clinical decision support systems. The principal challenge is the optimal collection and integration of diverse multimodal data sources in a quantitative manner that delivers unambiguous clinical predictions that accurately and robustly enable outcome prediction as a function of the impending decisions.

Methods

The search covered the years from January 2010 to January 2021. The inclusion criterion was: clinical study evaluating radiomics of liver colorectal metastases. Exclusion criteria were studies with no sufficient reported data, case report, review or editorial letter.

Results

We recognized 38 studies that assessed radiomics in mCRC from January 2010 to January 2021. Twenty were on different tpics, 5 corresponded to most criteria; 3 are review, or letter to editors; so 10 articles were included.

Conclusions

In colorectal liver metastases radiomics should be a valid tool for the characterization of lesions, in the stratification of patients based on the risk of relapse after surgical treatment and in the prediction of response to chemotherapy treatment.

Quantitative magnetic resonance imaging for focal liver lesions: bridging the gap between research and clinical practice

Magnetic resonance imaging (MRI) is highly important for the detection, characterization, and follow-up of focal liver lesions. Several quantitative MRI-based methods have been proposed in addition to qualitative imaging interpretation to improve the diagnostic work-up and prognostics in patients with focal liver lesions. This includes DWI with apparent diffusion coefficient measurements, intravoxel incoherent motion, perfusion imaging, MR elastography, and radiomics. Multiple research studies have reported promising results with quantitative MRI methods in various clinical settings. Nevertheless, applications in everyday clinical practice are limited. This review describes the basic principles of quantitative MRI-based techniques and discusses the main current applications and limitations for the assessment of focal liver lesions.

Machine Learning-Based Ultrasomics Improves the Diagnostic Performance in Differentiating Focal Nodular Hyperplasia and Atypical Hepatocellular Carcinoma

Background

The typical enhancement patterns of hepatocellular carcinoma (HCC) on contrast-enhanced ultrasound (CEUS) are hyper-enhanced in the arterial phase and washed out during the portal venous and late phases. However, atypical variations make a differential diagnosis both challenging and crucial. We aimed to investigate whether machine learning-based ultrasonic signatures derived from CEUS images could improve the diagnostic performance in differentiating focal nodular hyperplasia (FNH) and atypical hepatocellular carcinoma (aHCC).

Patients and Methods

A total of 226 focal liver lesions, including 107 aHCC and 119 FNH lesions, examined by CEUS were reviewed retrospectively. For machine learning-based ultrasomics, 3,132 features were extracted from the images of the baseline, arterial, and portal phases. An ultrasomics signature was generated by a machine learning model. The predictive model was constructed using the support vector machine method trained with the following groups: ultrasomics features, radiologist’s score, and combination of ultrasomics features and radiologist’s score. The diagnostic performance was explored using the area under the receiver operating characteristic curve (AUC).

Results

A total of 14 ultrasomics features were chosen to build an ultrasomics model, and they presented good performance in differentiating FNH and aHCC with an AUC of 0.86 (95% confidence interval [CI]: 0.80, 0.89), a sensitivity of 76.6% (95% CI: 67.5%, 84.3%), and a specificity of 80.5% (95% CI: 70.6%, 85.9%). The model trained with a combination of ultrasomics features and the radiologist’s score achieved a significantly higher AUC (0.93, 95% CI: 0.89, 0.96) than that trained with the radiologist’s score (AUC: 0.84, 95% CI: 0.79, 0.89, P < 0.001). For the sub-group of HCC with normal AFP value, the model trained with a combination of ultrasomics features, and the radiologist’s score remain achieved the highest AUC of 0.92 (95% CI: 0.87, 0.96) compared to that with the ultrasomics features (AUC: 0.86, 95% CI: 0.74, 0.89, P < 0.001) and radiologist’s score (AUC: 0.86, 95% CI: 0.79, 0.91, P < 0.001).

Conclusions

Machine learning-based ultrasomics performs as well as the staff radiologist in predicting the differential diagnosis of FNH and aHCC. Incorporating an ultrasomics signature into the radiologist’s score improves the diagnostic performance in differentiating FNH and aHCC.

Preoperative classification of primary and metastatic liver cancer via machine learning-based ultrasound radiomics

OBJECTIVE

To investigate the application of machine learning-based ultrasound radiomics in preoperative classification of primary and metastatic liver cancer.

METHODS

Data of 114 consecutive histopathologically confirmed patients with liver cancer from January 2018 to November 2019 were retrospectively analyzed. All patients underwent liver ultrasonography within 1 week before hepatectomy or fine-needle biopsy. The liver lesions were manually segmented by two experts using ITK-SNAP software. Seven categories of radiomics features, including first-order, two-dimensional shape, gray-level co-occurrence matrices, gray-level run-length matrix, gray-level size-zone matrix, neighboring gray tone difference matrix, and gray-level dependence matrix, were extracted on the Pyradiomics platform. Fourteen filters were applied to the original images, and derived images were obtained. Then, the dimensions of radiomics features were reduced by least absolute shrinkage and selection operator (Lasso) method. Finally, k-nearest neighbor (KNN), logistic regression (LR), multilayer perceptron (MLP), random forest (RF), and support vector machine (SVM) were employed to distinguish primary liver cancer from metastatic liver cancer by a fivefold cross-validation strategy. The performance of the established model was mainly evaluated by the area under the receiver operating characteristic (ROC) curve (AUC) and accuracy.

RESULTS

One thousand four hundred nine radiomics features were extracted from the original images and/or derived images for each patient. The mentioned five machine learning classifiers were able to differentiate primary liver cancer from metastatic liver cancer. LR outperformed other classifiers, with the accuracy of 0.843 ± 0.078 (AUC, 0.816 ± 0.088; sensitivity, 0.768 ± 0.232; specificity, 0.880 ± 0.117).

CONCLUSIONS

Machine learning-based ultrasound radiomics features are able to non-invasively distinguish primary liver tumors from metastatic liver tumors.

KEY POINTS

• Ultrasound-based radiomics was initially used for preoperative classification of primary versus metastatic liver cancer. • Multiple machine learning-based algorithms with cross-validation strategy were applied to extract machine learning-based ultrasound radiomics features. • Distinction between primary and metastatic tumors was obtained with a sensitivity of 0.768 and a specificity of 0.880.

Radiomics of Liver Metastases: A Systematic Review

Simple Summary

Patients with liver metastases can be scheduled for different therapies (e.g., chemotherapy, surgery, radiotherapy, and ablation). The choice of the most appropriate treatment should rely on adequate understanding of tumor biology and prediction of survival, but reliable biomarkers are lacking. Radiomics is an innovative approach to medical imaging: it identifies invisible-to-the-human-eye radiological patterns that can predict tumor aggressiveness and patients outcome. We reviewed the available literature to elucidate the role of radiomics in patients with liver metastases. Thirty-two papers were analyzed, mostly (56%) concerning metastases from colorectal cancer. Even if available studies are still preliminary, radiomics provided effective prediction of response to chemotherapy and of survival, allowing more accurate and earlier prediction than standard predictors. Entropy and homogeneity were the radiomic features with the strongest clinical impact. In the next few years, radiomics is expected to give a consistent contribution to the precision medicine approach to patients with liver metastases.

Abstract

Multidisciplinary management of patients with liver metastases (LM) requires a precision medicine approach, based on adequate profiling of tumor biology and robust biomarkers. Radiomics, defined as the high-throughput identification, analysis, and translational applications of radiological textural features, could fulfill this need. The present review aims to elucidate the contribution of radiomic analyses to the management of patients with LM. We performed a systematic review of the literature through the most relevant databases and web sources. English language original articles published before June 2020 and concerning radiomics of LM extracted from CT, MRI, or PET-CT were considered. Thirty-two papers were identified. Baseline higher entropy and lower homogeneity of LM were associated with better survival and higher chemotherapy response rates. A decrease in entropy and an increase in homogeneity after chemotherapy correlated with radiological tumor response. Entropy and homogeneity were also highly predictive of tumor regression grade. In comparison with RECIST criteria, radiomic features provided an earlier prediction of response to chemotherapy. Lastly, texture analyses could differentiate LM from other liver tumors. The commonest limitations of studies were small sample size, retrospective design, lack of validation datasets, and unavailability of univocal cut-off values of radiomic features. In conclusion, radiomics can potentially contribute to the precision medicine approach to patients with LM, but interdisciplinarity, standardization, and adequate software tools are needed to translate the anticipated potentialities into clinical practice.

Improving the malignancy characterization of hepatocellular carcinoma using deeply supervised cross modal transfer learning for non-enhanced MR

The malignancy characterization of hepatocellular carcinoma (HCC) is remarkably significant in clinical practice. In this work, we propose a deeply supervised cross modal transfer learning method to remarkably improve the malignancy characterization of HCC based on non-enhanced MR. First, we use samples of non-enhanced and contrast-enhanced MR for pre-training a deep learning network to learn the cross modal relationship between the non-enhanced modal and enhanced modal. Then, the parameters of the pre-trained across modal representation are transferred to a second deep learning model for fine-tuning based only on non-enhanced MR for malignancy characterization of HCC. Specifically, a deeply supervised network is designed to enhance the stability of the second deep learning model and further improve the performance of lesion characterization. Importantly, only non-enhanced MR of HCC is required for the malignancy characterization in the training and test phase of the second deep learning model. Experiments of one hundred and fifteen clinical HCCs demonstrate that the proposed deeply supervised cross modal transfer learning method can significantly improve the malignancy characterization of HCC based on non-enhanced MR.

Grading of hepatocellular carcinoma based on diffusion weighted images with multiple b-values using convolutional neural networks

PURPOSE

To effectively grade hepatocellular carcinoma (HCC) based on deep features derived from diffusion weighted images (DWI) with multiple b-values using convolutional neural networks (CNN).

MATERIALS AND METHODS

Ninety-eight subjects with 100 pathologically confirmed HCC lesions from July 2012 to October 2018 were included in this retrospective study, including 47 low-grade and 53 high-grade HCCs. DWI was performed for each subject with a 3.0T MR scanner in a breath-hold routine with three b-values (0,100, and 600 s/mm² ). First, logarithmic transformation was performed on original DWI images to generate log maps (logb0, logb100, and logb600). Then, a resampling method was performed to extract multiple 2D axial planes of HCCs from the log map to increase the dataset for training. Subsequently, 2D CNN was used to extract deep features of the log map for HCCs. Finally, fusion of deep features derived from three b-value log maps was conducted for HCC malignancy classification. Specifically, a deeply supervised loss function was devised to further improve the performance of lesion characterization. The data set was split into two parts: the training and validation set (60 HCCs) and the fixed test set (40 HCCs). Four-fold cross validation with 10 repetitions was performed to assess the performance of deep features extracted from single b-value images for HCC grading using the training and validation set. Receiver operating characteristic curve (ROC) and area under the curve (AUC) values were used to assess the characterization performance of the proposed deep feature fusion method to differentiate low-grade and high-grade in the fixed test set.

RESULTS

The proposed fusion of deep features derived from logb0, logb100, and logb600 with deeply supervised loss function generated the highest accuracy for HCC grading (80%), thus outperforming the method of deep feature derived from the ADC map directly (72.5%), the original b0 (65%), b100 (68%), and b600 (70%) images. Furthermore, AUC values of the deep features of the ADC map, the deep feature fusion with concatenation, and the proposed deep feature fusion with deeply supervised loss function were 0.73, 0.78, and 0.83, respectively.

CONCLUSION

The proposed fusion of deep features derived from the logarithm of the three b-value images yields high performance for HCC grading, thus providing a promising approach for the assessment of DWI in lesion characterization.

Automatic classification of focal liver lesions based on MRI and risk factors

OBJECTIVES

Accurate classification of focal liver lesions is an important part of liver disease diagnostics. In clinical practice, the lesion type is often determined from the abdominal MR examination, which includes T2-weighted and dynamic contrast enhanced (DCE) MR images. To date, only T2-weighted images are exploited for automatic classification of focal liver lesions. In this study additional MR sequences and risk factors are used for automatic classification to improve the results and to make a step forward to a clinically useful aid for radiologists.

MATERIALS AND METHODS

Clinical MRI data sets of 95 patients with in total 125 benign lesions (40 adenomas, 29 cysts and 56 hemangiomas) and 88 malignant lesions (30 hepatocellular carcinomas (HCC) and 58 metastases) were included in this study. Contrast curve, gray level histogram, and gray level co-occurrence matrix texture features were extracted from the DCE-MR and T2-weighted images. In addition, risk factors including the presence of steatosis, cirrhosis, and a known primary tumor were used as features. Fifty features with the highest ANOVA F-score were selected and fed to an extremely randomized trees classifier. The classifier evaluation was performed using the leave-one-out principle and receiver operating characteristic (ROC) curve analysis.

RESULTS

The overall accuracy for the classification of the five major focal liver lesion types is 0.77. The sensitivity/specificity is 0.80/0.78, 0.93/0.93, 0.84/0.82, 0.73/0.56, and 0.62/0.77 for adenoma, cyst, hemangioma, HCC, and metastasis, respectively.

CONCLUSION

The proposed classification system using features derived from clinical DCE-MR and T2-weighted images, with additional risk factors is able to differentiate five common types of lesions and is a step forward to a clinically useful aid for focal liver lesion diagnosis.

Radiomics-based classification of hepatocellular carcinoma and hepatic haemangioma on precontrast magnetic resonance images

Background

To evaluate the feasibility of using radiomics with precontrast magnetic resonance imaging for classifying hepatocellular carcinoma (HCC) and hepatic haemangioma (HH).

Methods

This study enrolled 369 consecutive patients with 446 lesions (a total of 222 HCCs and 224 HHs). A training set was constituted by randomly selecting 80% of the samples and the remaining samples were used to test. On magnetic resonance (MR) images of HCC and HH obtained with in-phase, out-phase, T2-weighted imaging (T2WI), and diffusion-weighted imaging (DWI) sequences, we outlined the target lesions and extracted 1029 radiomics features, which were classified as first-, second-, higher-order statistics and shape features. Then, the variance threshold, select k best, and least absolute shrinkage and selection operator algorithms were explored for dimensionality reduction of the features. We used four classifiers (decision tree, random forest, K nearest neighbours, and logistic regression) to identify HCC and HH on the basis of radiomics features. Two abdominal radiologists also performed the conventional qualitative analysis for classification of HCC and HH. Diagnostic performances of radiomics and radiologists were evaluated by receiver operating characteristic (ROC) analysis.

Results

Valuable radiomics features for building a radiomics signature were extracted from in-phase (n = 22), out-phase (n = 24), T2WI (n = 34) and DWI (n = 24) sequences. In comparison, the logistic regression classifier showed better predictive ability by combining four sequences. In the training set, the area under the ROC curve (AUC) was 0.86 (sensitivity: 0.76; specificity: 0.78), and in the testing set, the AUC was 0.89 (sensitivity: 0.822; specificity: 0.714). The diagnostic performance for the optimal radiomics-based combined model was significantly higher than that for the less experienced radiologist (2-years experience) (AUC = 0.702, p < 0.05), and had no statistic difference with the experienced radiologist (10-years experience) (AUC = 0.908, p>0.05).

Conclusions

We developed and validated a radiomics signature as an adjunct tool to distinguish HCC and HH by combining in-phase, out-phase, T2W, and DW MR images, which outperformed the less experienced radiologist (2-years experience), and was nearly equal to the experienced radiologist (10-years experience).

Fully automated tissue segmentation of the prescription isodose region delineated through the Gamma knife plan for cerebral arteriovenous malformation (AVM) using fuzzy C-means (FCM) clustering

Background

Gamma knife radiosurgery (GKRS) is a common treatment for cerebral arterio-venous malformations (AVMs), particularly in cases where the malformation is deep-seated, large, or in eloquent areas of the brain. Unfortunately, these procedures can result in radiation injury to brain parenchyma. The fact that every AVM is unique in its vascular morphology makes it nearly impossible to exclude brain parenchyma from isodose radiation exposure during the formulation of a GKRS plan. Calculating the percentages of the various forms of tissue exposed to specific doses of radiation is crucial to understanding the clinical responses and causes of brain parenchyma injury following GKRS for AVM.

Methods

In this study, we developed a fully automated algorithm using unsupervised classification via fuzzy c-means clustering for the analysis of T2 weighted images used in a Gamma knife plan. This algorithm is able to calculate the percentages of nidus, brain tissue, and cerebrospinal fluid (CSF) within the prescription isodose radiation exposure region.

Results

The proposed algorithm was used to assess the treatment plan of 25 patients with AVM who had undergone GKRS. The Dice similarity index (SI) was used to determine the degree of agreement between the results obtained using the algorithm and a visually guided manual method (the gold standard) performed by an experienced neurosurgeon. In the nidus, the SI was (74.86 ± 1.30%) (mean ± standard deviation), the sensitivity was (83.05 ± 11.91)%, and the specificity was (86.73 ± 10.31)%. In brain tissue, the SI was (79.50 ± 6.01)%, the sensitivity was (73.05 ± 9.77)%, and the specificity was (85.53 ± 7.13)%. In the CSF, the SI was (69.57 ± 15.26)%, the sensitivity was (89.86 ± 5.87)%, and the specificity was (92.36 ± 4.35)%.

Conclusions

The proposed clustering algorithm provides precise percentages of the various types of tissue within the prescription isodose region in the T2 weighted images used in the GKRS plan for AVM. Our results shed light on the causes of brain radiation injury after GKRS for AVM. In the future, this system could be used to improve outcomes and avoid complications associated with GKRS treatment.

•

A novel image analytical method for the analysis of images of an AVM in a GKRS plan

•

Fuzzy c-means clustering was used for analyses of T2w images in the GKRS plan.

•

Automatic calculation of percentages of tissue inside the isodose line

•

Brain tissue percentages of the nidus of the AVM predict risk of complication.

•

Proposed method could be used to avoid complications associated after GKRS.