Computer-aided focal liver lesion detection

Deep Learning-Based Approach for Identifying and Measuring Focal Liver Lesions on Contrast-Enhanced MRI

BACKGROUND

The number of focal liver lesions (FLLs) detected by imaging has increased worldwide, highlighting the need to develop a robust, objective system for automatically detecting FLLs.

PURPOSE

To assess the performance of the deep learning-based artificial intelligence (AI) software in identifying and measuring lesions on contrast-enhanced magnetic resonance imaging (MRI) images in patients with FLLs.

STUDY TYPE

Retrospective.

SUBJECTS

395 patients with 1149 FLLs.

FIELD STRENGTH/SEQUENCE

The 1.5 T and 3 T scanners, including T1-, T2-, diffusion-weighted imaging, in/out-phase imaging, and dynamic contrast-enhanced imaging.

ASSESSMENT

The diagnostic performance of AI, radiologist, and their combination was compared. Using 20 mm as the cut-off value, the lesions were divided into two groups, and then divided into four subgroups: <10, 10-20, 20-40, and ≥40 mm, to evaluate the sensitivity of radiologists and AI in the detection of lesions of different sizes. We compared the pathologic sizes of 122 surgically resected lesions with measurements obtained using AI and those made by radiologists.

STATISTICAL TESTS

McNemar test, Bland-Altman analyses, Friedman test, Pearson's chi-squared test, Fisher's exact test, Dice coefficient, and intraclass correlation coefficients. A P-value <0.05 was considered statistically significant.

RESULTS

The average Dice coefficient of AI in segmentation of liver lesions was 0.62. The combination of AI and radiologist outperformed the radiologist alone, with a significantly higher detection rate (0.894 vs. 0.825) and sensitivity (0.883 vs. 0.806). The AI showed significantly sensitivity than radiologists in detecting all lesions <20 mm (0.848 vs. 0.788). Both AI and radiologists achieved excellent detection performance for lesions ≥20 mm (0.867 vs. 0.881, P = 0.671). A remarkable agreement existed in the average tumor sizes among the three measurements (P = 0.174).

DATA CONCLUSION

AI software based on deep learning exhibited practical value in automatically identifying and measuring liver lesions.

LEVEL OF EVIDENCE: 4

TECHNICAL EFFICACY

Stage 2.

Deep learning-based algorithm to detect primary hepatic malignancy in multiphase CT of patients at high risk for HCC

OBJECTIVES

To develop and evaluate a deep learning-based model capable of detecting primary hepatic malignancies in multiphase CT images of patients at high risk for hepatocellular carcinoma (HCC).

METHODS

A total of 1350 multiphase CT scans of 1280 hepatic malignancies (1202 HCCs and 78 non-HCCs) in 1320 patients at high risk for HCC were retrospectively analyzed. Following the delineation of the focal hepatic lesions according to reference standards, the CT scans were categorized randomly into the training (568 scans), tuning (193 scans), and test (589 scans) sets. Multiphase CT information was subjected to multichannel integration, and livers were automatically segmented before model development. A deep learning-based model capable of detecting malignancies was developed using a mask region-based convolutional neural network. The thresholds of the prediction score and the intersection over union were determined on the tuning set corresponding to the highest sensitivity with < 5 false-positive cases per CT scan. The sensitivity and the number of false-positives of the proposed model on the test set were calculated. Potential causes of false-negatives and false-positives on the test set were analyzed.

RESULTS

This model exhibited a sensitivity of 84.8% with 4.80 false-positives per CT scan on the test set. The most frequent potential causes of false-negatives and false-positives were determined to be atypical enhancement patterns for HCC (71.7%) and registration/segmentation errors (42.7%), respectively.

CONCLUSIONS

The proposed deep learning-based model developed to automatically detect primary hepatic malignancies exhibited an 84.8% of sensitivity with 4.80 false-positives per CT scan in the test set.

KEY POINTS

• Image processing, including multichannel integration of multiphase CT and automatic liver segmentation, enabled the application of a deep learning-based model to detect primary hepatic malignancy. • Our model exhibited a sensitivity of 84.8% with a false-positive rate of 4.80 per CT scan.

Diagnostic Performance of Deep Learning-Based Lesion Detection Algorithm in CT for Detecting Hepatic Metastasis from Colorectal Cancer

OBJECTIVE

To compare the performance of the deep learning-based lesion detection algorithm (DLLD) in detecting liver metastasis with that of radiologists.

MATERIALS AND METHODS

This clinical retrospective study used 4386-slice computed tomography (CT) images and labels from a training cohort (502 patients with colorectal cancer [CRC] from November 2005 to December 2010) to train the DLLD for detecting liver metastasis, and used CT images of a validation cohort (40 patients with 99 liver metastatic lesions and 45 patients without liver metastasis from January 2011 to December 2011) for comparing the performance of the DLLD with that of readers (three abdominal radiologists and three radiology residents). For per-lesion binary classification, the sensitivity and false positives per patient were measured.

RESULTS

A total of 85 patients with CRC were included in the validation cohort. In the comparison based on per-lesion binary classification, the sensitivity of DLLD (81.82%, [81/99]) was comparable to that of abdominal radiologists (80.81%, p = 0.80) and radiology residents (79.46%, p = 0.57). However, the false positives per patient with DLLD (1.330) was higher than that of abdominal radiologists (0.357, p < 0.001) and radiology residents (0.667, p < 0.001).

CONCLUSION

DLLD showed a sensitivity comparable to that of radiologists when detecting liver metastasis in patients initially diagnosed with CRC. However, the false positives of DLLD were higher than those of radiologists. Therefore, DLLD could serve as an assistant tool for detecting liver metastasis instead of a standalone diagnostic tool.

Computer-aided diagnosis of liver lesions using CT images: A systematic review

BACKGROUND

Medical image processing has a strong footprint in radio diagnosis for the detection of diseases from the images. Several computer-aided systems were researched in the recent past to assist the radiologist in diagnosing liver diseases and reducing the interpretation time. The aim of this paper is to provide an overview of the state-of-the-art techniques in computer-assisted diagnosis systems to predict benign and malignant lesions using computed tomography images.

METHODS

The research articles published between 1998 and 2020 obtained from various standard databases were considered for preparing the review. The research papers include both conventional as well as deep learning-based systems for liver lesion diagnosis. The paper initially discusses the various hepatic lesions that are identifiable on computed tomography images, then the computer-aided diagnosis systems and their workflow. The conventional and deep learning-based systems are presented in stages wherein the various methods used for preprocessing, liver and lesion segmentation, radiological feature extraction and classification are discussed.

CONCLUSION

The review suggests the scope for future, work as efficient and effective segmentation methods that work well with diverse images have not been developed. Furthermore, unsupervised and semi-supervised deep learning models were not investigated for liver disease diagnosis in the reviewed papers. Other areas to be explored include image fusion and inclusion of essential clinical features along with the radiological features for better classification accuracy.

Computer-aided diagnosis of cirrhosis and hepatocellular carcinoma using multi-phase abdomen CT

PURPOSE

High mortality rate due to liver cirrhosis has been reported over the globe in the previous years. Early detection of cirrhosis may help in controlling the disease progression toward hepatocellular carcinoma (HCC). The lack of trained CT radiologists and increased patient population delays the diagnosis and further management. This study proposes a computer-aided diagnosis system for detecting cirrhosis and HCC in a very efficient and less time-consuming approach.

METHODS

Contrast-enhanced CT dataset of 40 patients (n = 40; M:F = 5:3; age = 25-55 years) with three groups of subjects: healthy (n = 14), cirrhosis (n = 12) and cirrhosis with HCC (n = 14), were retrospectively analyzed in this study. A novel method for the automatic 3D segmentation of liver using modified region-growing segmentation technique was developed and compared with the state-of-the-art deep learning-based technique. Further, histogram parameters were calculated from segmented CT liver volume for classification between healthy and diseased (cirrhosis and HCC) liver using logistic regression. Multi-phase analysis of CT images was performed to extract 24 temporal features for detecting cirrhosis and HCC liver using support vector machine (SVM).

RESULTS

The proposed method produced improved 3D segmentation with Dice coefficient 90% for healthy liver, 86% for cirrhosis and 81% for HCC subjects compared to the deep learning algorithm (healthy: 82%; cirrhosis: 78%; HCC: 70%). Standard deviation and kurtosis were found to be statistically different (p < 0.05) among healthy and diseased liver, and using logistic regression, classification accuracy obtained was 92.5%. For detecting cirrhosis and HCC liver, SVM with RBF kernel obtained highest slice-wise and patient-wise prediction accuracy of 86.9% (precision = 0.93, recall = 0.7) and 80% (precision = 0.86, recall = 0.75), respectively, than that of linear kernel (slice-wise: accuracy = 85.4%, precision = 0.92, recall = 0.67; patient-wise: accuracy = 73.33%, precision = 0.75, recall = 0.75).

CONCLUSIONS

The proposed computer-aided diagnosis system for detecting cirrhosis and hepatocellular carcinoma (HCC) showed promising results and can be used as effective screening tool in medical image analysis.

Novel Mahalanobis-based feature selection improves one-class classification of early hepatocellular carcinoma

Detection of early hepatocellular carcinoma (HCC) is responsible for increasing survival rates in up to 40%. One-class classifiers can be used for modeling early HCC in multidetector computed tomography (MDCT), but demand the specific knowledge pertaining to the set of features that best describes the target class. Although the literature outlines several features for characterizing liver lesions, it is unclear which is most relevant for describing early HCC. In this paper, we introduce an unconstrained GA feature selection algorithm based on a multi-objective Mahalanobis fitness function to improve the classification performance for early HCC. We compared our approach to a constrained Mahalanobis function and two other unconstrained functions using Welch's t-test and Gaussian Data Descriptors. The performance of each fitness function was evaluated by cross-validating a one-class SVM. The results show that the proposed multi-objective Mahalanobis fitness function is capable of significantly reducing data dimensionality (96.4%) and improving one-class classification of early HCC (0.84 AUC). Furthermore, the results provide strong evidence that intensity features extracted at the arterial to portal and arterial to equilibrium phases are important for classifying early HCC.

Shape-intensity prior level set combining probabilistic atlas and probability map constrains for automatic liver segmentation from abdominal CT images

PURPOSE

Propose a fully automatic 3D segmentation framework to segment liver on challenging cases that contain the low contrast of adjacent organs and the presence of pathologies from abdominal CT images.

METHODS

First, all of the atlases are weighted in the selected training datasets by calculating the similarities between the atlases and the test image to dynamically generate a subject-specific probabilistic atlas for the test image. The most likely liver region of the test image is further determined based on the generated atlas. A rough segmentation is obtained by a maximum a posteriori classification of probability map, and the final liver segmentation is produced by a shape-intensity prior level set in the most likely liver region. Our method is evaluated and demonstrated on 25 test CT datasets from our partner site, and its results are compared with two state-of-the-art liver segmentation methods. Moreover, our performance results on 10 MICCAI test datasets are submitted to the organizers for comparison with the other automatic algorithms.

RESULTS

Using the 25 test CT datasets, average symmetric surface distance is [Formula: see text] mm (range 0.62-2.12 mm), root mean square symmetric surface distance error is [Formula: see text] mm (range 0.97-3.01 mm), and maximum symmetric surface distance error is [Formula: see text] mm (range 12.73-26.67 mm) by our method. Our method on 10 MICCAI test data sets ranks 10th in all the 47 automatic algorithms on the site as of July 2015. Quantitative results, as well as qualitative comparisons of segmentations, indicate that our method is a promising tool to improve the efficiency of both techniques.

CONCLUSION

The applicability of the proposed method to some challenging clinical problems and the segmentation of the liver are demonstrated with good results on both quantitative and qualitative experimentations. This study suggests that the proposed framework can be good enough to replace the time-consuming and tedious slice-by-slice manual segmentation approach.

Improved Patch-Based Automated Liver Lesion Classification by Separate Analysis of the Interior and Boundary Regions

The bag-of-visual-words (BoVW) method with construction of a single dictionary of visual words has been used previously for a variety of classification tasks in medical imaging, including the diagnosis of liver lesions. In this paper, we describe a novel method for automated diagnosis of liver lesions in portal-phase computed tomography (CT) images that improves over single-dictionary BoVW methods by using an image patch representation of the interior and boundary regions of the lesions. Our approach captures characteristics of the lesion margin and of the lesion interior by creating two separate dictionaries for the margin and the interior regions of lesions ("dual dictionaries" of visual words). Based on these dictionaries, visual word histograms are generated for each region of interest within the lesion and its margin. For validation of our approach, we used two datasets from two different institutions, containing CT images of 194 liver lesions (61 cysts, 80 metastasis, and 53 hemangiomas). The final diagnosis of each lesion was established by radiologists. The classification accuracy for the images from the two institutions was 99% and 88%, respectively, and 93% for a combined dataset. Our new BoVW approach that uses dual dictionaries shows promising results. We believe the benefits of our approach may generalize to other application domains within radiology.

Three-dimensional spatiotemporal features for fast content-based retrieval of focal liver lesions

Content-based image retrieval systems for 3-D medical datasets still largely rely on 2-D image-based features extracted from a few representative slices of the image stack. Most 2 -D features that are currently used in the literature not only model a 3-D tumor incompletely but are also highly expensive in terms of computation time, especially for high-resolution datasets. Radiologist-specified semantic labels are sometimes used along with image-based 2-D features to improve the retrieval performance. Since radiological labels show large interuser variability, are often unstructured, and require user interaction, their use as lesion characterizing features is highly subjective, tedious, and slow. In this paper, we propose a 3-D image-based spatiotemporal feature extraction framework for fast content-based retrieval of focal liver lesions. All the features are computer generated and are extracted from four-phase abdominal CT images. Retrieval performance and query processing times for the proposed framework is evaluated on a database of 44 hepatic lesions comprising of five pathological types. Bull's eye percentage score above 85% is achieved for three out of the five lesion pathologies and for 98% of query lesions, at least one same type of lesion is ranked among the top two retrieved results. Experiments show that the proposed system's query processing is more than 20 times faster than other already published systems that use 2-D features. With fast computation time and high retrieval accuracy, the proposed system has the potential to be used as an assistant to radiologists for routine hepatic tumor diagnosis.

Content-based image retrieval of multiphase CT images for focal liver lesion characterization

PURPOSE

Characterization of focal liver lesions with various imaging modalities can be very challenging in the clinical practice and is experience-dependent. The authors' aim is to develop an automatic method to facilitate the characterization of focal liver lesions (FLLs) using multiphase computed tomography (CT) images by radiologists.

METHODS

A multiphase-image retrieval system is proposed to retrieve a preconstructed database of FLLs with confirmed diagnoses, which can assist radiologists' decision-making in FLL characterization. It first localizes the FLL on multiphase CT scans using a hybrid generative-discriminative FLL detection method and a nonrigid B-spline registration method. Then, it extracts the multiphase density and texture features to numerically represent the FLL. Next, it compares the query FLL with the model FLLs in the database in terms of the feature and measures their similarities using the L1-norm based similarity scores. The model FLLs are ranked by similarities and the top results are finally provided to the users for their evidence studies.

RESULTS

The system was tested on a database of 69 four-phase contrast-enhanced CT scans, consisting of six classes of liver lesions, and evaluated in terms of the precision-recall curve and the Bull's Eye Percentage Score (BEP). It obtained a BEP score of 78%. Compared with any single-phase based representation, the multiphase-based representation increased the BEP scores of the system, from 63%-65% to 78%. In a pilot study, two radiologists performed characterization of FLLs without and with the knowledge of the top five retrieved results. The results were evaluated in terms of the diagnostic accuracy, the receiver operating characteristic (ROC) curve and the mean diagnostic confidence. One radiologist's accuracy improved from 75% to 92%, the area under ROC curves (AUC) from 0.85 to 0.95 (p = 0.081), and the mean diagnostic confidence from 4.6 to 7.3 (p = 0.039). The second radiologist's accuracy did not change, at 75%, with AUC increasing from 0.72 to 0.75 (p = 0.709), and the mean confidence from 4.5 to 4.9 (p = 0.607).

CONCLUSIONS

Multiphase CT images can be used in content-based image retrieval for FLL's categorization and result in good performance in comparison with single-phase CT images. The proposed method has the potential to improve the radiologists' diagnostic accuracy and confidence by providing visually similar lesions with confirmed diagnoses for their interpretation of clinical studies.

Automated liver lesion detection in CT images based on multi-level geometric features

PURPOSE

Due to the increasing number of liver cancer cases in clinical practice, there is a significant need for efficient tools for computer-assisted liver lesion analysis. A wide range of clinical applications, such as lesion characterization, quantification and follow-up, can be facilitated by automated liver lesion detection. Liver lesions vary significantly in size, shape, density and heterogeneity, which make them difficult to detect automatically. The goal of this work was to develop a method that can detect all types of liver lesions with high sensitivity and low false positive rate within a short run time.

METHODS

The proposed method identifies abnormal regions in liver CT images based on their intensity using a multi-level segmentation approach. The abnormal regions are analyzed from the inside-out using basic geometric features (such as asymmetry, compactness or volume). Using this multi-level shape characterization, the abnormal regions are classified into lesions and other region types (including vessel, liver boundary). The proposed analysis also allows defining the contour of each finding. The method was trained on a set of 55 cases involving 120 lesions and evaluated on a set of 30 images involving 59 (various types of) lesions, which were manually contoured by a physician.

RESULTS

The proposed algorithm demonstrated a high detection rate (92 %) at a low (1.7) false positive per case (precision 51 %), when the method was started from a manually contoured liver. The same level of false positive per case (1.6) and precision (51 %) was achieved at a somewhat lower detection rate (85 %), when the volume of interest was defined by a fully automated liver segmentation.

CONCLUSIONS

The proposed method can efficiently detect liver lesions irrespective of their size, shape, density and heterogeneity within half a minute. According to the evaluation, its accuracy is competitive with the actual state-of-the-art approaches, and the contour of the detected findings is acceptable in most of the cases. Future work shall focus on more precise lesion contouring so that the proposed method can be a solid basis for fully automated liver tumour burden estimation.