Deep Learning Electronic Cleansing for Single- and Dual-Energy CT Colonography

Comparison of Diatrizoate and Iohexol for Patient Acceptance and Fecal-Tagging Performance in Noncathartic CT Colonography

OBJECTIVE

The aim of this study was to compare diatrizoate and iohexol regarding patient acceptance and fecal-tagging performance in noncathartic computed tomography colonography.

METHODS

This study enrolled 284 volunteers with fecal tagging by either diatrizoate or iohexol at an iodine concentration of 13.33 mg/mL and an iodine load of 24 g. Patient acceptance was rated on a 4-point scale of gastrointestinal discomfort. Two gastrointestinal radiologists jointly analyzed image quality, fecal-tagging density and homogeneity, and residual contrast agent in the small intestine. The results were compared by the generalized estimating equation method.

RESULTS

Patient acceptance was comparable between the 2 groups (3.95 ± 0.22 vs 3.96 ± 0.20, P = 0.777). The diatrizoate group had less residual fluid and stool than the iohexol group ( P = 0.019, P = 0.004, respectively). There was no significant difference in colorectal distention, residual fluid, and stool tagging quality between the 2 groups (all P 's > 0.05). The mean 2-dimensional image quality score was 4.59 ± 0.68 with diatrizoate and 3.60 ± 1.14 with iohexol ( P < 0.001). The attenuation of tagged feces was 581 ± 66 HU with diatrizoate and 1038 ± 117 HU with iohexol ( P < 0.001). Residual contrast agent in the small intestine was assessed at 55.3% and 62.3% for the diatrizoate group and iohexol group, respectively ( P = 0.003).

CONCLUSIONS

Compared with iohexol, diatrizoate had better image quality, proper fecal-tagging density, and more homogeneous tagging along with comparable excellent patient acceptance, and might be more suitable for fecal tagging in noncathartic computed tomography colonography.

Dual- and multi-energy CT for particle stopping-power estimation: current state, challenges and potential

Range uncertainty has been a key factor preventing particle radiotherapy from reaching its full physical potential. One of the main contributing sources is the uncertainty in estimating particle stopping power (ρ_s) within patients. Currently, theρ_sdistribution in a patient is derived from a single-energy CT (SECT) scan acquired for treatment planning by converting CT number expressed in Hounsfield units (HU) of each voxel toρ_susing a Hounsfield look-up table (HLUT), also known as the CT calibration curve. HU andρ_sshare a linear relationship with electron density but differ in their additional dependence on elemental composition through different physical properties, i.e. effective atomic number and mean excitation energy, respectively. Because of that, the HLUT approach is particularly sensitive to differences in elemental composition between real human tissues and tissue surrogates as well as tissue variations within and among individual patients. The use of dual-energy CT (DECT) forρ_sprediction has been shown to be effective in reducing the uncertainty inρ_sestimation compared to SECT. The acquisition of CT data over different x-ray spectra yields additional information on the material elemental composition. Recently, multi-energy CT (MECT) has been explored to deduct material-specific information with higher dimensionality, which has the potential to further improve the accuracy ofρ_sestimation. Even though various DECT and MECT methods have been proposed and evaluated over the years, these approaches are still only scarcely implemented in routine clinical practice. In this topical review, we aim at accelerating this translation process by providing: (1) a comprehensive review of the existing DECT/MECT methods forρ_sestimation with their respective strengths and weaknesses; (2) a general review of uncertainties associated with DECT/MECT methods; (3) a general review of different aspects related to clinical implementation of DECT/MECT methods; (4) other potential advanced DECT/MECT applications beyondρ_sestimation.

AAPM task group report 273: Recommendations on best practices for AI and machine learning for computer-aided diagnosis in medical imaging

Rapid advances in artificial intelligence (AI) and machine learning, and specifically in deep learning (DL) techniques, have enabled broad application of these methods in health care. The promise of the DL approach has spurred further interest in computer-aided diagnosis (CAD) development and applications using both "traditional" machine learning methods and newer DL-based methods. We use the term CAD-AI to refer to this expanded clinical decision support environment that uses traditional and DL-based AI methods. Numerous studies have been published to date on the development of machine learning tools for computer-aided, or AI-assisted, clinical tasks. However, most of these machine learning models are not ready for clinical deployment. It is of paramount importance to ensure that a clinical decision support tool undergoes proper training and rigorous validation of its generalizability and robustness before adoption for patient care in the clinic. To address these important issues, the American Association of Physicists in Medicine (AAPM) Computer-Aided Image Analysis Subcommittee (CADSC) is charged, in part, to develop recommendations on practices and standards for the development and performance assessment of computer-aided decision support systems. The committee has previously published two opinion papers on the evaluation of CAD systems and issues associated with user training and quality assurance of these systems in the clinic. With machine learning techniques continuing to evolve and CAD applications expanding to new stages of the patient care process, the current task group report considers the broader issues common to the development of most, if not all, CAD-AI applications and their translation from the bench to the clinic. The goal is to bring attention to the proper training and validation of machine learning algorithms that may improve their generalizability and reliability and accelerate the adoption of CAD-AI systems for clinical decision support.

Comparison of the performance of conventional and spectral-based tagged stool cleansing algorithms at CT colonography

OBJECTIVES

To compare the performance of conventional versus spectral-based electronic stool cleansing for iodine-tagged CT colonography (CTC) using a dual-layer spectral detector scanner.

METHODS

We retrospectively evaluated iodine contrast stool-tagged CTC scans of 30 consecutive patients (mean age: 69 ± 8 years) undergoing colorectal cancer screening obtained on a dual-layer spectral detector CT scanner. One reader identified locations of electronic cleansing artifacts (n = 229) on conventional and spectral cleansed images. Three additional independent readers evaluated these locations using a conventional cleansing algorithm (Intellispace Portal) and two experimental spectral cleansing algorithms (i.e., fully transparent and translucent tagged stool). For each cleansed image set, readers recorded the severity of over- and under-cleansing artifacts on a 5-point Likert scale (0 = none to 4 = severe) and readability compared to uncleansed images. Wilcoxon's signed-rank tests were used to assess artifact severity, type, and readability (worse, unchanged, or better).

RESULTS

Compared with conventional cleansing (66% score ≥ 2), the severity of overall cleansing artifacts was lower in transparent (60% score ≥ 2, p = 0.011) and translucent (50% score ≥ 2, p < 0.001) spectral cleansing. Under-cleansing artifact severity was lower in transparent (49% score ≥ 2, p < 0.001) and translucent (39% score ≥ 2, p < 0.001) spectral cleansing compared with conventional cleansing (60% score ≥ 2). Over-cleansing artifact severity was worse in transparent (17% score ≥ 2, p < 0.001) and translucent (14% score ≥ 2, p = 0.023) spectral cleansing compared with conventional cleansing (9% score ≥ 2). Overall readability was significantly improved in transparent (p < 0.001) and translucent (p < 0.001) spectral cleansing compared with conventional cleansing.

CONCLUSIONS

Spectral cleansing provided more robust electronic stool cleansing of iodine-tagged stool at CTC than conventional cleansing.

KEY POINTS

• Spectral-based electronic cleansing of tagged stool at CT colonography provides higher quality images with less perception of artifacts than does conventional cleansing. • Spectral-based electronic cleansing could potentially advance minimally cathartic approach for CT colonography. Further clinical trials are warranted.

Self-Supervised Adversarial Learning with a Limited Dataset for Electronic Cleansing in Computed Tomographic Colonography: A Preliminary Feasibility Study

Existing electronic cleansing (EC) methods for computed tomographic colonography (CTC) are generally based on image segmentation, which limits their accuracy to that of the underlying voxels. Because of the limitations of the available CTC datasets for training, traditional deep learning is of limited use in EC. The purpose of this study was to evaluate the technical feasibility of using a novel self-supervised adversarial learning scheme to perform EC with a limited training dataset with subvoxel accuracy. A three-dimensional (3D) generative adversarial network (3D GAN) was pre-trained to perform EC on CTC datasets of an anthropomorphic phantom. The 3D GAN was then fine-tuned to each input case by use of the self-supervised scheme. The architecture of the 3D GAN was optimized by use of a phantom study. The visually perceived quality of the virtual cleansing by the resulting 3D GAN compared favorably to that of commercial EC software on the virtual 3D fly-through examinations of 18 clinical CTC cases. Thus, the proposed self-supervised 3D GAN, which can be trained to perform EC on a small dataset without image annotations with subvoxel accuracy, is a potentially effective approach for addressing the remaining technical problems of EC in CTC.

Diagnostic accuracy of ultra-low-dose CT colonography for the detection of colorectal polyps: a feasibility study

PURPOSE

The aim of this feasibility study was to evaluate the diagnostic accuracy of ultra-low-dose CT colonography using iterative reconstruction algorithms with reference to standard colonoscopy.

MATERIALS AND METHODS

Prior to this study, a phantom study was performed to investigate the optimal protocol for ultra-low-dose CT colonography. A total of 206 patients with average/high risk of colorectal cancer were recruited. After undergoing full bowel preparation, the patients were scanned in the prone and supine positions with the CT conditions set to 120 kV, standard deviation 45 to 50, and an adaptive iterative reconstruction algorithm applied. Two expert readers read the images independently. The main outcome measures were the per-patient and per-polyp accuracies for the detection of polyps ≥ 10 mm, with colonoscopy results as the reference standard.

RESULTS

Two hundred patients (102 females, mean age 67.5 years) underwent both ultra-low-dose CT colonography and colonoscopy on the same day. The mean radiation exposure dose was 0.64 ± 0.34 mSv. On colonoscopy, 39 patients had 45 polyps ≥ 10 mm (non-polypoid morphology 7), including 4 cancers. Per-patient sensitivity, specificity, and accuracy of CT colonography for polyps ≥ 10 mm were 0.74, 0.96, and 0.92 for reader one, and 0.74, 0.99, and 0.94 for reader two, respectively. Per-polyp sensitivities for polyps ≥ 10 mm were 0.73 for reader one and 0.71 for reader two. On subgroup analysis by morphology, non-polypoid polyps ≥ 10 mm were not detected by both readers.

CONCLUSION

Extreme ultra-low-dose CT colonography had an insufficient diagnostic performance for the detection of polyps ≥ 10 mm, because it was unable to detect non-polypoid polyps. This study showed that the problem with ultra-low-dose CT colonography was the lack of detectability of small-size polyps, especially non-polypoid polyps. To use ultra-low-dose CT colonography clinically, it is necessary to resolve the problems identified by this study.

Automatic Detection and Segmentation of Colorectal Cancer with Deep Residual Convolutional Neural Network

Early and automatic detection of colorectal tumors is essential for cancer analysis, and the same is implemented using computer-aided diagnosis (CAD). A computerized tomography (CT) image of the colon is being used to identify colorectal carcinoma. Digital imaging and communication in medicine (DICOM) is a standard medical imaging format to process and analyze images digitally. Accurate detection of tumor cells in the complex digestive tract is necessary for optimal treatment. The proposed work is divided into two phases. The first phase involves the segmentation, and the second phase is the extraction of the colon lesions with the observed segmentation parameters. A deep convolutional neural network (DCNN) based residual network approach for the colon and polyps' segmentation from the CT images is applied over the 2D CT images. The residual stack block is being added to the hidden layers with short skip nuance, which helps to retain spatial information. ResNet-enabled CNN is employed in the current work to achieve complete boundary segmentation of the colon cancer region. The results obtained through segmentation serve as features for further extraction and classification of benign as well as malignant colon cancer. Performance evaluation metrics indicate that the proposed network model has effectively segmented and classified colorectal tumors with dice scores of 91.57% (on average), sensitivity = 98.28, specificity = 98.68, and accuracy = 98.82.

Spectral CT of the abdomen: Where are we now?

Spectral CT adds a new dimension to radiological evaluation, beyond assessment of anatomical abnormalities. Spectral data allows for detection of specific materials, improves image quality while at the same time reducing radiation doses and contrast media doses, and decreases the need for follow up evaluation of indeterminate lesions. We review the different acquisition techniques of spectral images, mainly dual-source, rapid kV switching and dual-layer detector, and discuss the main spectral results available. We also discuss the use of spectral imaging in abdominal pathologies, emphasizing the strengths and pitfalls of the technique and its main applications in general and in specific organs.

A review of deep learning in medical imaging: Imaging traits, technology trends, case studies with progress highlights, and future promises

Since its renaissance, deep learning has been widely used in various medical imaging tasks and has achieved remarkable success in many medical imaging applications, thereby propelling us into the so-called artificial intelligence (AI) era. It is known that the success of AI is mostly attributed to the availability of big data with annotations for a single task and the advances in high performance computing. However, medical imaging presents unique challenges that confront deep learning approaches. In this survey paper, we first present traits of medical imaging, highlight both clinical needs and technical challenges in medical imaging, and describe how emerging trends in deep learning are addressing these issues. We cover the topics of network architecture, sparse and noisy labels, federating learning, interpretability, uncertainty quantification, etc. Then, we present several case studies that are commonly found in clinical practice, including digital pathology and chest, brain, cardiovascular, and abdominal imaging. Rather than presenting an exhaustive literature survey, we instead describe some prominent research highlights related to these case study applications. We conclude with a discussion and presentation of promising future directions.

A Survey on Machine Learning and Deep Learning-based Computer-Aided Methods for Detection of Polyps in CT Colonography

BACKGROUND

Colon cancer generally begins as a neoplastic growth of tissue, called polyps, originating from the inner lining of the colon wall. Most colon polyps are considered harmless but over the time, they can evolve into colon cancer, which, when diagnosed in later stages, is often fatal. Hence, time is of the essence in the early detection of polyps and the prevention of colon cancer.

METHODS

To aid this endeavor, many computer-aided methods have been developed, which use a wide array of techniques to detect, localize and segment polyps from CT Colonography images. In this paper, a comprehensive state-of-the-art method is proposed and categorize this work broadly using the available classification techniques using Machine Learning and Deep Learning.

CONCLUSION

The performance of each of the proposed approach is analyzed with existing methods and also how they can be used to tackle the timely and accurate detection of colon polyps.

A generative flow-based model for volumetric data augmentation in 3D deep learning for computed tomographic colonography

Purpose

Deep learning can be used for improving the performance of computer-aided detection (CADe) in various medical imaging tasks. However, in computed tomographic (CT) colonography, the performance is limited by the relatively small size and the variety of the available training datasets. Our purpose in this study was to develop and evaluate a flow-based generative model for performing 3D data augmentation of colorectal polyps for effective training of deep learning in CADe for CT colonography.

Methods

We developed a 3D-convolutional neural network (3D CNN) based on a flow-based generative model (3D Glow) for generating synthetic volumes of interest (VOIs) that has characteristics similar to those of the VOIs of its training dataset. The 3D Glow was trained to generate synthetic VOIs of polyps by use of our clinical CT colonography case collection. The evaluation was performed by use of a human observer study with three observers and by use of a CADe-based polyp classification study with a 3D DenseNet.

Results

The area-under-the-curve values of the receiver operating characteristic analysis of the three observers were not statistically significantly different in distinguishing between real polyps and synthetic polyps. When trained with data augmentation by 3D Glow, the 3D DenseNet yielded a statistically significantly higher polyp classification performance than when it was trained with alternative augmentation methods.

Conclusion

The 3D Glow-generated synthetic polyps are visually indistinguishable from real colorectal polyps. Their application to data augmentation can substantially improve the performance of 3D CNNs in CADe for CT colonography. Thus, 3D Glow is a promising method for improving the performance of deep learning in CADe for CT colonography.

Electronic cleansing of tagged residue in CT colonography: what radiologists need to know

CT colonography (CTC) is the radiological examination of choice for the diagnosis of colorectal neoplasia. Faecal tagging is considered a mandatory part of bowel preparation. However, the colonic mucosa, obscured by tagged residue, is not accessible to endoluminal 3D views and requires time-consuming 2D evaluation. Electronic cleansing (EC) software algorithms can overcome this limitation by digitally subtracting tagged residue from the colonic lumen. Ideally, this enables a seamless 3D endoluminal evaluation. Despite this benefit, EC is a potential source of a wide range of artefacts. Accurate EC requires proper CTC examination technique and faecal tagging. The digital subtraction process has been shown to affect the relevant morphological features of both colonic anatomy and colonic lesions, if submerged under faecal residue. This article summarises the potential effects of EC on CTC imaging, the consequences for reporting and patient management, and strategies to avoid pitfalls. Furthermore, potentially negative effects on clinical reporting and patient management are shown, and problem-solving techniques, as well as recommendations for the appropriate use of EC techniques, are presented. Radiologists using EC should be familiar with EC-related effects on polyp size and also with correct measurement techniques.

Enhancement of Deep Learning in Image Classification Performance Using Xception with the Swish Activation Function for Colorectal Polyp Preliminary Screening

One of the leading forms of cancer is colorectal cancer (CRC), which is responsible for increasing mortality in young people. The aim of this paper is to provide an experimental modification of deep learning of Xception with Swish and assess the possibility of developing a preliminary colorectal polyp screening system by training the proposed model with a colorectal topogram dataset in two and three classes. The results indicate that the proposed model can enhance the original convolutional neural network model with evaluation classification performance by achieving accuracy of up to 98.99% for classifying into two classes and 91.48% for three classes. For testing of the model with another external image, the proposed method can also improve the prediction compared to the traditional method, with 99.63% accuracy for true prediction of two classes and 80.95% accuracy for true prediction of three classes.