Computer-assisted Diagnosis System Research Articles

Classification of orbital tumors using convolutional neural networks

AbstractOrbital tumors are the most common eye tumors that affect people all over the world. Early detection prevents the progression to other regions of the eye and the body. Also, early identification and treatment could reduce mortality. A computer-assisted diagnosis (CAD) system to help physicians diagnose tumors is in great demand in ophthalmology. In recent years, deep learning has demonstrated promising outcomes in computer vision systems. This work proposes a CAD system for detecting various forms of orbital tumors using convolutional neural networks. The system has three stages: preprocessing, data augmentation and classification. The proposed system was evaluated on two datasets of magnetic resonance imaging (MRI) images containing 1404 MRI T1-weighted images and 1560 MRI T2-weighted images. The results have shown that the system is capable of detecting and classifying the tumor in each image type, and the recognition rate for the T1-weighted image is 98% and for the T2-weighted image is 97%.

Optimal deep transfer learning driven computer‐aided breast cancer classification using ultrasound images

AbstractBreast cancer (BC) is regarded as the second leading type of cancer among women globally. Ultrasound images are typically used for the identification and classification of abnormalities that exist in the breast. To enhance diagnosis performance, the computer assisted diagnosis (CAD) model finds it effective for identifying and classifying BC. Generally, the CAD technique contains distinct procedures like feature extraction, preprocessing, segmentation, and classification. The recent developments of deep learning (DL) algorithms in the form of CAD system helps to minimize the cost and enhance the ability of radiologists to interpret medical images. Therefore, this study develops an optimal deep transfer learning driven computer aided BC classification (ODTLD‐CABCC) technique on ultrasound images. The presented ODTLD‐CABCC algorithm undergoes pre‐processing in two levels such as median filtering based noise removal and graph cut segmentation. Furthermore, the residual network (ResNet101) model can be used as a feature extractor. Finally, the sailfish optimizer (SFO) with a labelled weighted extreme learning machine (LWELM) algorithm is used for the classification process. The SFO technique is employed to choose optimal parameters involved in the LWELM algorithm. A comprehensive set of simulations are conducted on the benchmark data and the experimental outcomes are examined under numerous aspects. The comparative examination represents the supremacy of the ODTLD‐CABCC technique over the other approaches.

Deep Learning-based Assessment of Facial Asymmetry Using U-Net Deep Convolutional Neural Network Algorithm.

This study aimed to evaluate the diagnostic performance of a deep convolutional neural network (DCNN)-based computer-assisted diagnosis (CAD) system to detect facial asymmetry on posteroanterior (PA) cephalograms and compare the results of the DCNN with those made by the orthodontist. PA cephalograms of 1020 patients with orthodontics were used to train the DCNN-based CAD systems for autoassessment of facial asymmetry, the degree of menton deviation, and the coordinates of its regarding landmarks. Twenty-five PA cephalograms were used to test the performance of the DCNN in analyzing facial asymmetry. The diagnostic performance of the DCNN-based CAD system was assessed using independent t -tests and Bland-Altman plots. Comparison between the DCNN-based CAD system and conventional analysis confirmed no significant differences. Bland-Altman plots showed good agreement for all the measurements. The DCNN-based CAD system might offer a clinically acceptable diagnostic evaluation of facial asymmetry on PA cephalograms.

Diagnostic value of a computer-assisted diagnosis system for the ultrasound features in thyroid nodules.

To explore the diagnostic value of the TUIAS (SW_TH01/II) computer-aided diagnosis (CAD) software system for the ultrasound Thyroid Imaging Reporting and Data System (TI-RADS) features in thyroid nodules. This retrospective study enrolled patients with thyroid nodules in Shanghai East Hospital between January 2017 and October 2021. The novel CAD software (SW_TH01/II) and three sonographers performed a qualitative analysis of the ultrasound TI-RADS features in aspect ratio, margin irregularity, margin smoothness, calcification, and echogenicity of the thyroid nodules. A total of 225 patients were enrolled. The accuracy, sensitivity, and specificity of the CAD software in "aspect ratio" were 95.6%, 96.2%, and 95.4%, in "margin irregularity" were 90.7%, 90.5%, and 90.9%, in "margin smoothness" were 85.8%, 88.5%, and 83.0%, in "calcification" were 83.6%, 81.7%, and 82.0%, in "homogeneity" were 88.9%, 90.6%, and 82.2%, in "major echo" were 85.3%, 88.0%, and 85.4%, and in "contains very hypoechoic echo" were 92.0%, 90.0%, and 92.4%. The analysis time of the CAD software was significantly shorter than for the sonographers (2.7 ± 1.6 vs. 29.7 ± 12.7 s, P < 0.001). The CAD system achieved high accuracy in describing thyroid nodule features. It might assist in clinical thyroid nodule analysis.

Zone-specific computer-aided diagnosis system aimed at characterizing ISUP ≥ 2 prostate cancers on multiparametric magnetic resonance images: evaluation in a cohort of patients on active surveillance.

To assess a region-of-interest-based computer-assisted diagnosis system (CAD) in characterizing aggressive prostate cancer on magnetic resonance imaging (MRI) from patients under active surveillance (AS). A prospective biopsy database was retrospectively searched for patients under AS who underwent MRI and subsequent biopsy at our institution. MRI lesions targeted at baseline biopsy were retrospectively delineated to calculate the CAD score that was compared to the Prostate Imaging-Reporting and Data System (PI-RADS) version 2 score assigned at baseline biopsy. 186 patients were selected. At baseline biopsy, 51 and 15 patients had International Society of Urological Pathology (ISUP) grade ≥ 2 and ≥ 3 cancer respectively. The CAD score had significantly higher specificity for ISUP ≥ 2 cancers (60% [95% confidence interval (CI): 51-68]) than the PI-RADS score (≥ 3 dichotomization: 24% [CI: 17-33], p = 0.0003; ≥ 4 dichotomization: 32% [CI: 24-40], p = 0.0003). It had significantly lower sensitivity than the PI-RADS ≥ 3 dichotomization (85% [CI: 74-92] versus 98% [CI: 91-100], p = 0.015) but not than the PI-RADS ≥ 4 dichotomization (94% [CI:85-98], p = 0.104). Combining CAD findings and PSA density could have avoided 47/184 (26%) baseline biopsies, while missing 3/51 (6%) ISUP 2 and no ISUP ≥ 3 cancers. Patients with baseline negative CAD findings and PSAd < 0.15ng/mL2 who stayed on AS after baseline biopsy had a 9% (4/44) risk of being diagnosed with ISUP ≥ 2 cancer during a median follow-up of 41months, as opposed to 24% (18/74) for the others. The CAD could help define AS patients with low risk of aggressive cancer at baseline assessment and during subsequent follow-up.

Cost-effectiveness analysis of the artificial intelligence diagnosis support system for early gastric cancers.

The introduction of artificial intelligence into the medical field has improved the diagnostic capabilities of physicians. However, few studies have analyzed the economic impact of employing artificial intelligence technologies in the clinical environment. This study evaluated the cost-effectiveness of a computer-assisted diagnosis (CADx) system designed to support clinicians in differentiating early gastric cancers from non-cancerous lesions in Japan, where the universal health insurance system was introduced. The target population to be used for the CADx was estimated as those with moderate to severe gastritis caused by Helicobacter pylori infection. Decision trees with Markov models were built to analyze the cumulative cost-effectiveness of using CADx relative to the pre-artificial intelligence status quo, a condition reconstructed from data in published reports. After conducting a base-case analysis, we performed sensitivity analyses by modifying several parameters. The primary outcome was the incremental cost-effectiveness ratio. Compared with the status quo as represented in the base-case analysis, the incremental cost-effectiveness ratio of CADx in the Japanese market was forecasted to be 11,093 USD per quality-adjusted life year. The sensitivity analyses demonstrated that the expected incremental cost-effectiveness ratios were within the willingness-to-pay threshold of 50,000 USD per quality-adjusted life year when the cost of the CAD was less than 104 USD. Using CADx for EGCs may decrease their misdiagnosis, contributing to improved cost-effectiveness in Japan.

Bayesian optimized GoogLeNet based respiratory signal prediction model from empirically decomposed gammatone visualization

Discrimination of normal and adventitious respiratory sounds from stethoscope auscultation by human ears is challenging owing to low frequency characteristics and varying frequency range for inspiration and expiration. This makes the diagnosis of pulmonary disorders a subjective one relying on the experience and hearing capability of the physician. Most computer assisted diagnosis systems formulated to address these limitations fail to capture the inherent acoustic property of respiratory sounds close to human auditory system. To circumvent this problem, this study exploits the gammatone filter banks to evaluate the distribution of all frequency components present in the signal. To categorize the respiratory signals, GoogLeNet based Convolutional Neural Networks (CNN) prediction model is developed through Time-Frequency (TF) visualization of gammatone cepstral coefficients acquired from decomposed Intrinsic Mode Functions (IMFs) in an empirical manner. As the performance of the CNN model is greatly dependent on the learning environment, this study also tends to optimize the values of the hyper parameters to enhance the classification performance of the CNN model. Accordingly, the optimal values of initial learning rate, L2 regularization and momentum are identified for both Stochastic Gradient Descent Momentum (SGDM) and Adaptive Momentum (ADAM) optimizer using Bayesian optimization technique. Experimentation is carried out for three TF visualization methods viz. spectrograms, scalograms and Constant Q spectrograms. Evaluated results show that scalogram method of classification yields high accuracy of 87.37% and 88.32% for default selection of hyperparameters using SGDM and ADAM optimizers and with optimal selection of hyperparameters based on objective function reveal an improved accuracy of 93.68 % and 95.67% for SGDM and ADAM optimizers.

Inteligencia artificial en la colonoscopia de tamizaje y la disminución del error.

Artificial Intelligence (AI) has the potential to change many aspects of healthcare practice. Image discrimination and classification has many applications within medicine. Machine learning algorithms and complicated neural networks have been developed to train a computer to differentiate between normal and abnormal areas. Machine learning is a form of AI that allows the platform to improve without being programmed. Computer Assisted Diagnosis (CAD) is based on latency, which is the time between the captured image and when it is displayed on the screen. AI-assisted endoscopy can increase the detection rate by identifying missed lesions. An AI CAD system must be responsive, specific, with easy-to-use interfaces, and provide fast results without substantially prolonging procedures. AI has the potential to help both, trained and trainee endoscopists. Rather than being a substitute for high-quality technique, it should serve as a complement to good practice. AI has been evaluated in three clinical scenarios in colonic neoplasms: the detection of polyps, their characterization (adenomatous vs. non-adenomatous) and the prediction of invasive cancer within a polypoid lesion.

CHEST X-RAY IMAGE SEGMENTATION USING MODIFIED DEEPLABV3+ METHOD

COVID-19 is a disease that affects the human respiratory system. The latest developments in September 2022 the number of confirmed cases of COVID-19 worldwide reached 608,328,548 with 6,501,469 patients who died. While in Indonesia confirmed COVID-19 reached 6,408,806 with 157,892 patients who died. Reserve Transcription Polymerase Chain Reaction (RT-PCR) is the most widely used tool. However, the latest RT-PCR test report shows that the RT-PCR test is inadequate. As an alternative, radiographic images such as chest x-rays and CT scans can help detect this. Radiographic images, especially x-rays, need processing to be able to make a diagnosis. Computer Aided Diagnosis (CAD) is a computer assisted diagnosis system that can be used as supporting information in making a diagnosis. To make it easier to make a diagnosis, we need a deep learning model that can help with this. DeepLabV3+ is a method that can carry out the segmentation process. DeepLabV3+ which is an extension of DeepLabV3 with the aim of improving the segmentation results. DeepLabV3+ uses a modified Xception as the backbone. In this study, 1,500 chest x-ray image data were used which were then divided into 80% for training data and 20% for testing data. There are 4 test scenarios in this study, namely with a learning rate of 0.01 without CLAHE, a learning rate of 0,01 and using CLAHE, a learning rate of 0,0001 without CLAHE, and a learning rate of 0,0001 using CLAHE. Of the 4 scenarios the learning rate scenario is 0,01 and using CLAHE gets the highest evaluation results using the Dice Similarity Coefficient (DSC) of 96.91%.

Smart Investigations into the Development of an Effective Computer-Assisted Diagnosis System for CT Scan Brain Depictions

Smart Investigations into the Development of an Effective Computer-Assisted Diagnosis System for CT Scan Brain Depictions

Experiences Regarding Use and Implementation of Artificial Intelligence-Supported Follow-Up of Atypical Moles at a Dermatological Outpatient Clinic: Qualitative Study.

Artificial intelligence (AI) is increasingly used in numerous medical fields. In dermatology, AI can be used in the form of computer-assisted diagnosis (CAD) systems when assessing and diagnosing skin lesions suspicious of melanoma, a potentially lethal skin cancer with rising incidence all over the world. In particular, CAD may be a valuable tool in the follow-up of patients with high risk of developing melanoma, such as patients with multiple atypical moles. One such CAD system, ATBM Master (FotoFinder), can execute total body dermoscopy (TBD). This process comprises automatically photographing a patient´s entire body and then neatly displaying moles on a computer screen, grouped according to their clinical relevance. Proprietary FotoFinder algorithms underlie this organized presentation of moles. In addition, ATBM Master's optional convoluted neural network (CNN)-based Moleanalyzer Pro software can be used to further assess moles and estimate their probability of malignancy. Few qualitative studies have been conducted on the implementation of AI-supported procedures in dermatology. Therefore, the purpose of this study was to investigate how health care providers experience the use and implementation of a CAD system like ATBM Master, in particular its TBD module. In this way, the study aimed to elucidate potential barriers to the application of such new technology. We conducted a thematic analysis based on 2 focus group interviews with 14 doctors and nurses regularly working in an outpatient pigmented lesions clinic. Surprisingly, the study revealed that only 3 participants had actual experience using the TBD module. Even so, all participants were able to provide many notions and anticipations about its use, resulting in 3 major themes emerging from the interviews. First, several organizational matters were revealed to be a barrier to consistent use of the ATBM Master's TBD module, namely lack of guidance, time pressure, and insufficient training. Second, the study found that the perceived benefits of TBD were the ability to objectively detect and monitor subtle lesion changes and unbiasedness of the procedure. Imprecise identification of moles, inability to photograph certain areas, and substandard technical aspects were the perceived weaknesses. Lastly, the study found that clinicians were open to use AI-powered technology and that the TBD module was considered a supplementary tool to aid the medical staff, rather than a replacement of the clinician. Demonstrated by how few of the participants had actual experience with the TBD module, this study showed that implementation of new technology does not occur automatically. It highlights the importance of having a strategy for implementation to ensure the optimized application of CAD tools. The study identified areas that could be improved when implementing AI-powered technology, as well as providing insight on how medical staff anticipated and experienced the use of a CAD device in dermatology.

Efficient Machine Learning-based Approach for Brain Tumor Detection Using the CAD System

Medical research has focused on improving diagnosis through medical imaging in recent decades. Computer Assisted Diagnosis (CAD) systems have been developed to help doctors identify suspicious areas of interest, particularly those with cancer-like characteristics. CAD systems employ various algorithms and techniques to extract important numerical measurements from medical images that clinicians can use to evaluate patient conditions. This study proposes a statistical classification-based approach to efficient brain cancer detection. The proposed approach operates in three stages: first, Gradient Vector Flow (GVF) Snake models and mathematical morphology techniques retrieve regions of interest. The second stage characterizes these regions using morphological and textural parameters. Finally, a Bayesian network uses this description as input to identify malignant and benign cancer classes. We also compared the performance of the Bayesian network with other popular classification algorithms, including SVM, MLP, KNN, Random Forest, Decision Tree, XGBoost, LGBM, Gaussian Process, and RBF SVM. The results showed the superiority of the Bayesian network for the task of brain tumor classification. The proposed approach has been experimentally validated, with a sensitivity of 100% and a classification accuracy of over 98% for tumors, demonstrating the high efficiency of cancer cell segmentation.

Brain Mapping Disease Identification System

Abstract: This paper introduces concepts and principles of how computer assisted diagnosis (CAD) systems work and addresses the challenges of processing large amounts of medical data for diagnosis. It also provides an overview of CAD methods developed in the field of diagnosing neurological diseases and which existing technologies provide what big data for analyzing which diseases. Although there is extensive work on the development of CAD systems for automated screening of diseases, experts still cannot use them regularly in their decision-making processes due to the lack of easily available online methods. This study may help professionals gain insight and understand how CAD systems can help solve diagnosis problems and may help computer designers to be inspired to create more accessible CAD systems in the future from existing medical big data rarely used in CAD at present

Analytical aspects of the antinuclear antibody test by HEp-2 indirect immunofluorescence: EFLM report on an international survey.

Detection of antinuclear antibodies (ANA) by indirect immunofluorescence assay using HEp-2cells (HEp-2 IFA) is used to screen for various autoimmune diseases. HEp-2 IFA suffers from variability, which hampers harmonization. A questionnaire was developed to collect information on HEp-2 IFA methodology, computer-assisted diagnosis (CAD) systems, training, inter-observer variability, quality assessment, reagent lot change control, and method verification. The questionnaire was distributed to laboratories by Sciensano (Belgium), national EASI groups (Italy, Croatia, Portugal, Estonia, Greece) and ICAP (worldwide). Answers were obtained by 414 laboratories. The results were analysed in the framework of the recent EFLM/EASI/ICAP ANA recommendations (companion paper). Laboratories used either HEp-2, HEp-2000, or HEp-20-10cells and most laboratories (80%) applied the same screening dilution for children and adults. The conjugate used varied between laboratories [IgG-specific (in 57% of laboratories) vs. polyvalent]. Sixty-nine percent of CAD users reviewed the automatic nuclear pattern and 53% of CAD users did not fully exploit the fluorescence intensity for quality assurance. Internal quality control was performed by 96% of the laboratories, in 52% of the laboratories only with strongly positive samples. Interobserver variation was controlled by 79% of the laboratories. Limited lot-to-lot evaluation was performed by 68% of the laboratories. Method verification was done by 80% of the respondents. Even though many laboratories embrace high-quality HEp-2 IFA, substantial differences in how HEp-2 IFA is performed and controlled remain. Acting according to the EFLM/EASI/ICAP ANA recommendations can improve the global performance and quality of HEp-2 IFA and nurture harmonization.

Detection of antinuclear antibodies: recommendations from EFLM, EASI and ICAP.

Antinuclear antibodies (ANA) are important for the diagnosis of various autoimmune diseases. ANA are usually detected by indirect immunofluorescence assay (IFA) using HEp-2 cells (HEp-2 IFA). There are many variables influencing HEp-2 IFA results, such as subjective visual reading, serum screening dilution, substrate manufacturing, microscope components and conjugate. Newer developments on ANA testing that offer novel features adopted by some clinical laboratories include automated computer-assisted diagnosis (CAD) systems and solid phase assays (SPA). A group of experts reviewed current literature and established recommendations on methodological aspects of ANA testing. This process was supported by a two round Delphi exercise. International expert groups that participated in this initiative included (i) the European Federation of Clinical Chemistry and Laboratory Medicine (EFLM) Working Group "Autoimmunity Testing"; (ii) the European Autoimmune Standardization Initiative (EASI); and (iii) the International Consensus on ANA Patterns (ICAP). In total, 35 recommendations/statements related to (i) ANA testing and reporting by HEp-2 IFA; (ii) HEp-2 IFA methodological aspects including substrate/conjugate selection and the application of CAD systems; (iii) quality assurance; (iv) HEp-2 IFA validation/verification approaches and (v) SPA were formulated. Globally, 95% of all submitted scores in the final Delphi round were above 6 (moderately agree, agree or strongly agree) and 85% above 7 (agree and strongly agree), indicating strong international support for the proposed recommendations. These recommendations are an important step to achieve high quality ANA testing.

Development of artificial intelligence for the detection and staging of esophageal cancer

: Computer-assisted diagnosis (CAD) using deep learning, based on convolutional neural networks (CNNs), is rapidly developing in modern society and clinical practice. In the medical field of endoscopy, a CAD system can be used in the detection and staging of superficial esophageal cancer. In the detection of superficial esophageal squamous cell carcinoma (ESCC), several studies were reported as using still images of conventional white-light endoscopy and narrow-band imaging (NBI). The sensitivity of CAD systems is over 90%, and in some reports, higher than that of expert endoscopists. In addition, there is a report using video for a validation set that showed good performance. In diagnosing invasion depth, there are reports using conventional white-light imaging, NBI, and magnifying endoscopy with NBI using still images; there are also reports of intrapapillary capillary loop pattern identification using still images. Using white-light endoscopy and NBI, CAD systems showed sensitivity of 84.1–95.4%, specificity of 73.3–79.2%, and accuracy of 80.9–92.9% for differentiating SM1 cancers from SM2 or SM3 cancers in pathology. Additional systems could accurately classify intrapapillary capillary loop patterns as normal or abnormal, with comparable performance to experienced endoscopists. As for esophageal adenocarcinoma (EAC) that arises in Barrett’s esophagus (BE), there are detection reports that show favorable performance. Furthermore, the CAD system from one report identified the location suitable for biopsy; this is helpful for endoscopists because of the difficulty in determining the location of early EAC in BE. Although these studies are still only at research level, excellent performance has been achieved for detecting and staging of superficial esophageal carcinoma by CAD systems. In the near future, CAD systems will support us in detecting and staging esophageal cancers in daily clinical practice, leading to a better prognosis.

딥러닝 기반 위 질환 컴퓨터 보조 진단 시스템 개발 및 Grad-CAM을 활용한 시각화

Gastric cancer ranked first in the incidence of cancer in Korea from 2015 to 2018, and it is a trend that many people suffer from regardless of gender. Because the cause of gastric cancer is not clear and early gastric cancer has no specific symptoms, patients often miss the appropriate treatment time. Gastric endoscopy is the most effective method for diagnosing gastric cancer, but early gastric cancer accompany only subtle changes that are difficult to distinguish with the naked eye, and the diagnosis accuracy may vary depending on the proficiency of the specialist. Therefore, we proposed an EfficientNetV2-L-based Computer-assisted Diagnosis (CADx) system for gastric diseases in this paper. EfficientNetV2 is a network that can have high classification performance for gastric diseases even with a small number of parameters as a model designed using the Training-NAS method. In general, the diversity of data affects the performance of deep learning models. So, we augmented data using Cifar10 augmented policy. Through this method, sensitivity was improved from 0.878 to 0.966 in early gastric cancer and normal classification, and from 0.690 to 0.817 in abnormal and normal classification. In addition, we tried to visualize the region of interest of the model for the location of the lesion by applying Grad-CAM.

PILN: A posterior information learning network for blind reconstruction of lung CT images

Background and objective:Computer tomography (CT) imaging technology has played significant roles in the diagnosis and treatment of various lung diseases, but the degradations in CT images usually cause the loss of detailed structural information and interrupt the judgement from clinicians. Therefore, reconstructing noise-free, high-resolution CT images with sharp details from degraded ones is of great importance for the computer-assisted diagnosis (CAD) system. However, current image reconstruction methods suffer from unknown parameters of multiple degradations in actual clinical images.Methods:To solve these problems, we propose a unified framework, so called Posterior Information Learning Network (PILN), for blind reconstruction of lung CT images. The framework consists of two stages: Firstly, a noise level learning (NLL) network is proposed to quantify the Gaussian and artifact noise degradations into different levels. Inception-residual modules are designed to extract multi-scale deep features from the noisy image, and residual self-attention structures are proposed to refine deep features to essential representations of noise. Secondly, by taking the estimated noise levels as prior information, a cyclic collaborative super-resolution (CyCoSR) network is proposed to iteratively reconstruct the high-resolution CT image and estimate the blur kernel. Two convolutional modules are designed based on cross-attention transformer structure, named as Reconstructor and Parser. The high-resolution image is restored from the degraded image by the Reconstructor under the guidance of the predicted blur kernel, while the blur kernel is estimated by the Parser according to the reconstructed image and the degraded one. The NLL and CyCoSR networks are formulated as an end-to-end framework to handle multiple degradations simultaneously.Results:The proposed PILN is applied to the Cancer Imaging Archive (TCIA) dataset and the Lung Nodule Analysis 2016 Challenge (LUNA16) dataset to evaluate its ability in reconstructing lung CT images. Compared with the state-of-the-art image reconstruction algorithms, it can provide high-resolution images with less noise and sharper details with respect to quantitative benchmarks.Conclusions:Extensive experimental results demonstrate that our proposed PILN can achieve better performance on blind reconstruction of lung CT images, providing noise-free, detail-sharp and high-resolution images without knowing the parameters of multiple degradation sources.

Commentary: Machine learning for autism spectrum disorder diagnosis - challenges and opportunities - a commentary on Schulte-Rüther et al. (2022).

The commentary cites a study by Schulte-Rüther et al. (Journal of Child Psychology and Psychiatry, 2022) that proposed a machine learning model to predict a clinical best-estimate diagnosis of ASD when existing other co-occurring diagnoses. We discuss the valuable contribution of this work to developing a reliable computer-assisted diagnosis (CAD) system for ASD and point out that related research can be integrated with other multimodal machine learning methods. For future studies on developing the CAD system for ASD, we propose problems that need to be solved and potential research directions.

Image synthesis with disentangled attributes for chest X-ray nodule augmentation and detection

Lung nodule detection in chest X-ray (CXR) images is common to early screening of lung cancers. Deep-learning-based Computer-Assisted Diagnosis (CAD) systems can support radiologists for nodule screening in CXR images. However, it requires large-scale and diverse medical data with high-quality annotations to train such robust and accurate CADs. To alleviate the limited availability of such datasets, lung nodule synthesis methods are proposed for the sake of data augmentation. Nevertheless, previous methods lack the ability to generate nodules that are realistic with the shape/size attributes desired by the detector. To address this issue, we introduce a novel lung nodule synthesis framework in this paper, which decomposes nodule attributes into three main aspects including the shape, the size, and the texture, respectively. A GAN-based Shape Generator firstly models nodule shapes by generating diverse shape masks. The following Size Modulation then enables quantitative control on the diameters of the generated nodule shapes in pixel-level granularity. A coarse-to-fine gated convolutional Texture Generator finally synthesizes visually plausible nodule textures conditioned on the modulated shape masks. Moreover, we propose to synthesize nodule CXR images by controlling the disentangled nodule attributes for data augmentation, in order to better compensate for the nodules that are easily missed in the detection task. Our experiments demonstrate the enhanced image quality, diversity, and controllability of the proposed lung nodule synthesis framework. We also validate the effectiveness of our data augmentation strategy on greatly improving nodule detection performance.