Diagnostic Image Analysis Research Articles

Preparation of abdominal computed tomography data set for patients with abdominal aortic aneurysm

BACKGROUND: Artificial intelligence (AI) technologies are actively implemented in the processing and analysis of diagnostic medical images. The accuracy and reliability of AI algorithms are determined by the amount and quality of training data sets. Currently, a need exists for increased open access data sets, particularly abdominal aortic CT angiographic studies (CTA). Limitations of existing abdominal aortic CTA data sets are binary labeling (classification of the entire study) and small number of examinations. In addition, most examinations do not contain signs of aortic pathology, which, given its variability, significantly limits their use for AI training, since the target of such algorithms is the detection of pathology.
 AIM: To prepare a CTA data set for patients with abdominal aortic aneurysm.
 METHODS: Using the CTA data set with sings of abdominal aortic aneurysm, the stages and features of data set creation for AI training in accordance with the methodological recommendations were considered. Given the basic diagnostic requirements for the selected clinical task, the terms of reference for the preparation of the data set were formed, the required sample size was calculated, and the optimal annotation scenario was determined. The next stage included the selection of initial CT data of abdominal organs in the Unified Radiology Information System, anonymization of data, semi-automatic labeling and of the area of interest (aortic wall and aortic bed) using the 3D Slicer tool and its verification by an examining radiologist, and documentation of intermediate results.
 RESULTS: The calculated sample volume included 100 scans, containing the arterial phase, with a slice thickness of up to 1.2 mm. The balance of normal vs. pathology classes was chosen to be 1:4. Partial annotation of the data (50%) was performed.
 CONCLUSIONS: A methodology for preparing CTA data sets was developed. The generated dataset, if the necessary procedures are followed, will be placed in the public domain and may be used for training and testing AI algorithms and conducting scientific research.

Artificial intelligence in dental practice: a review

The human brain is a distinctly complex structure with several interlinked neurons that transmit signals all over the body. The search for an excellent model mimicking the human mind has led to a sophisticated breakthrough in what's referred to as artificial intelligence (AI). AI methodologies have determined programs in numerous disciplines ranging from telecommunication, aerospace, robotics, medical analysis, alternate marketplace, law, science, or entertainment to name some. Medical clinical decision support system (CDSS), a factor of AI is being carried out in dentistry which includes Artificial neural networks (ANN), genetic algorithms (GA) and Fuzzy logic. Various fields of drugs such as diagnostic systems, biomedical analysis, image analysis, and drug development have utilized this complicated and tremendously advanced detail of AI. The AI systems along with virtual reality have been used now not handiest to lessen dental anxiety but also to appear as an effective tool for the non-pharmacological manipulation of pain. Ordinary, AI offers us a glimpse of the destiny tool to be able to assist dentists in an inconceivable manner.

Incorporating algorithmic uncertainty into a clinical machine deep learning algorithm for urgent head CTs.

Machine learning (ML) algorithms to detect critical findings on head CTs may expedite patient management. Most ML algorithms for diagnostic imaging analysis utilize dichotomous classifications to determine whether a specific abnormality is present. However, imaging findings may be indeterminate, and algorithmic inferences may have substantial uncertainty. We incorporated awareness of uncertainty into an ML algorithm that detects intracranial hemorrhage or other urgent intracranial abnormalities and evaluated prospectively identified, 1000 consecutive noncontrast head CTs assigned to Emergency Department Neuroradiology for interpretation. The algorithm classified the scans into high (IC+) and low (IC-) probabilities for intracranial hemorrhage or other urgent abnormalities. All other cases were designated as No Prediction (NP) by the algorithm. The positive predictive value for IC+ cases (N = 103) was 0.91 (CI: 0.84-0.96), and the negative predictive value for IC- cases (N = 729) was 0.94 (0.91-0.96). Admission, neurosurgical intervention, and 30-day mortality rates for IC+ was 75% (63-84), 35% (24-47), and 10% (4-20), compared to 43% (40-47), 4% (3-6), and 3% (2-5) for IC-. There were 168 NP cases, of which 32% had intracranial hemorrhage or other urgent abnormalities, 31% had artifacts and postoperative changes, and 29% had no abnormalities. An ML algorithm incorporating uncertainty classified most head CTs into clinically relevant groups with high predictive values and may help accelerate the management of patients with intracranial hemorrhage or other urgent intracranial abnormalities.

Artificial intelligence in liver cancers: Decoding the impact of machine learning models in clinical diagnosis of primary liver cancers and liver cancer metastases

Liver cancers are the fourth leading cause of cancer-related mortality worldwide. In the past decade, breakthroughs in the field of artificial intelligence (AI) have inspired development of algorithms in the cancer setting. A growing body of recent studies have evaluated machine learning (ML) and deep learning (DL) algorithms for pre-screening, diagnosis and management of liver cancer patients through diagnostic image analysis, biomarker discovery and predicting personalized clinical outcomes. Despite the promise of these early AI tools, there is a significant need to explain the ‘black box’ of AI and work towards deployment to enable ultimate clinical translatability. Certain emerging fields such as RNA nanomedicine for targeted liver cancer therapy may also benefit from application of AI, specifically in nano-formulation research and development given that they are still largely reliant on lengthy trial-and-error experiments. In this paper, we put forward the current landscape of AI in liver cancers along with the challenges of AI in liver cancer diagnosis and management. Finally, we have discussed the future perspectives of AI application in liver cancer and how a multidisciplinary approach using AI in nanomedicine could accelerate the transition of personalized liver cancer medicine from bench side to the clinic.

DMFL_Net: A Federated Learning-Based Framework for the Classification of COVID-19 from Multiple Chest Diseases Using X-rays.

Coronavirus Disease 2019 (COVID-19) is still a threat to global health and safety, and it is anticipated that deep learning (DL) will be the most effective way of detecting COVID-19 and other chest diseases such as lung cancer (LC), tuberculosis (TB), pneumothorax (PneuTh), and pneumonia (Pneu). However, data sharing across hospitals is hampered by patients' right to privacy, leading to unexpected results from deep neural network (DNN) models. Federated learning (FL) is a game-changing concept since it allows clients to train models together without sharing their source data with anybody else. Few studies, however, focus on improving the model's accuracy and stability, whereas most existing FL-based COVID-19 detection techniques aim to maximize secondary objectives such as latency, energy usage, and privacy. In this work, we design a novel model named decision-making-based federated learning network (DMFL_Net) for medical diagnostic image analysis to distinguish COVID-19 from four distinct chest disorders including LC, TB, PneuTh, and Pneu. The DMFL_Net model that has been suggested gathers data from a variety of hospitals, constructs the model using the DenseNet-169, and produces accurate predictions from information that is kept secure and only released to authorized individuals. Extensive experiments were carried out with chest X-rays (CXR), and the performance of the proposed model was compared with two transfer learning (TL) models, i.e., VGG-19 and VGG-16 in terms of accuracy (ACC), precision (PRE), recall (REC), specificity (SPF), and F1-measure. Additionally, the DMFL_Net model is also compared with the default FL configurations. The proposed DMFL_Net + DenseNet-169 model achieves an accuracy of 98.45% and outperforms other approaches in classifying COVID-19 from four chest diseases and successfully protects the privacy of the data among diverse clients.

Pancreatic Cancer Segmentation and Classification in CT Imaging using Antlion Optimization and Deep Learning Mechanism

Pancreatic cancer, a fatal type of cancer, has a very poor prognosis. To monitor, forecast, and categorise cancer presence, automated pancreatic cancer segmentation and classification utilising Computer-Aided Diagnostic (CAD) model can be used. Furthermore, deep learning algorithms can provide in-depth diagnostic knowledge and precise image analysis for therapeutic usage. In this context, our study aims to develop an Antlion Optimization-Convolutional Neural Network-Gated Recurrent Unit (ALO-CNN-GRU) model for pancreatic tumour segmentation and classification based on deep learning and CT scans. The ALO-CNN-GRU technique’s objective is to segment and categorize the presence of cancer tissues. This technique consists of pre-processing, segmentation and feature extraction and classification phases. The images go through a pre-processing stage to reduce noise from the dataset that was obtained. A hybrid Gaussian and median filter is applied for the pre-processing phase. To identify the pancreatic area that is affected, the segmentation is processed utilizing the Antlion optimization method. Then, the categorization of pancreatic cancer as benign or malignant is done by employing the classifiers of the Convolutional neural network and Gated Recurrent Unit networks. The suggested model offers improved precision and a better rate of pancreatic cancer diagnosis with an accuracy of 99.92%.

Role of Imaging and AI in the Evaluation of COVID-19 Infection: A Comprehensive Survey.

Coronavirus disease 2019 (COVID-19) is a respiratory illness that started and rapidly became the pandemic of the century, as the number of people infected with it globally exceeded 253.4 million. Since the beginning of the pandemic of COVID-19, over two years have passed. During this hard period, several defies have been coped by the scientific society to know this novel disease, evaluate it, and treat affected patients. All these efforts are done to push back the spread of the virus. This article provides a comprehensive review to learn about the COVID-19 virus and its entry mechanism, its main repercussions on many organs and tissues of the body, identify its symptoms in the short and long terms, in addition to recognize the role of diagnosis imaging in COVID-19. Principally, the quick evolution of active vaccines act an exceptional accomplishment where leaded to decrease rate of death worldwide. However, some hurdels still have to be overcome. Many proof referrers that infection with CoV-19 causes neurological dis function in a substantial ratio of influenced patients, where these symptoms appear severely during the infection and still less is known about the potential long term consequences for the brain, where Loss of smell is a neurological sign and rudimentary symptom of COVID-19. Hence, we review the causes of olfactory bulb dysfunction and Anosmia associated with COVID-19, the latest appropriate therapeutic strategies for the COVID-19 treatment (e.g., the ACE2 strategy and the Ang II receptor), and the tests through the follow-up phases. Additionally, we discuss the long-term complications of the virus and thus the possibility of improving therapeutic strategies. Moreover, the main steps of artificial intelligence that have been used to foretell and early diagnose COVID-19 are presented, where Artificial intelligence, especially machine learning is emerging as an effective approach for diagnostic image analysis with performance in the discriminate diagnosis of injuries of COVID-19 on multiple organs, comparable to that of human practitioners. The followed methodology to prepare the current survey is to search the related work concerning the mentioned topic from different journals, such as Springer, Wiley, and Elsevier. Additionally, different studies have been compared, the results are collected and then reported as shown. The articles are selected based on the year (i.e., the last three years). Also, different keywords were checked (e.g., COVID-19, COVID-19 Treatment, COVID-19 Symptoms, and COVID-19 and Anosmia).

Non-glandular findings on breast ultrasound. Part I: a pictorial review of superficial lesions.

Ultrasound (US) represents the first-level imaging technique in the assessment of breast in young patients, whereas it is complementary to mammography in adult ones. It is not uncommon to encounter non-glandular mass during either screening or diagnostic breast US; sometimes the evaluation of palpable lump may be the reason of clinician's US request. The breast US field-of-view includes not only the glandular parenchyma, but also the tissues located anterior and posterior to it, from the skin to the ribs. In this setting, the radiologist must be familiar with the non-glandular breast diseases, which can occur in the superficial layers as well as in the chest wall. The differential diagnosis varies according to anatomic layer, so the anatomic origin is the first feature to assess and the correct localization is needed to avoid misdiagnosis and to choose, when requested, the second diagnostic step, imaging or histologic analysis. This paper is the first of two focused on non-glandular breast lesions; characterization, differential diagnosis, and pitfalls of superficial lesions are reviewed. They may be located in the dermis or hypodermis: the former are usually benign skin lesions, whereas the latter, although usually benign, may arise also from the anterior terminal lobular units, hence the papilloma, fibroadenomas, and breast cancers should be included in the differential diagnosis. US is more sensitive than CT and MRI in the assessment of superficial lesions due to higher spatial resolution.

Diagnostic Imaging Analysis and Care of Patients with Endomyocardial Fibrosis Based on Wireless Network Smart Medical Application.

The heart is one of the most important organs of the human body, but in recent years heart disease has become one of the human health killers and this paper explores endomyocardial fibrosis, which is a common cardiomyopathy, commonly seen in infants and children, and refers to a diffuse elastic fibrous disease of the endocardium. The purpose of this paper is to explore the diagnostic imaging analysis and care of patients with endocardial heart machine fibrosis using wireless network intelligent medical technology, aiming to provide a new power basis for the treatment of the disease in related patients. This paper proposes a new endocardial segmentation algorithm that aims to process image information using image features, intervene in image noise reduction and smoothing, etc., and use image grayscale values to confirm cardiac cavity grayscale values as a basis for physicians to make certain judgments for the diagnosis of patients with endocardial machine fibrosis. The experimental results show that the atrial fibrillation group is distinctly higher compared to the sinus rhythm group, with values remaining between 25 and 39, which is a significant advantage compared to other methods.

A New CT Analysis of Abdominal Wall after DIEP Flap Harvesting.

The abdominal microsurgical flap based on the deep inferior epigastric artery perforator (DIEP) flap has become the most popular option worldwide for autologous breast reconstruction. Several authors have investigated the results of reconstructed breasts, but the literature lacks systematic reviews exploring the donor site of the abdominal wall. To fulfil our aims, a new diagnostic muscle imaging analysis was designed and implemented. This study focused on rectus abdominal muscle morphology and function in a single series of 12 consecutive patients analysed before and after breast reconstruction with a microsurgical DIEP flap. Patients were divided into two groups, namely, “ipsilateral reconstruction” and “contralateral reconstruction”, depending on the side of the flap harvest and breast reconstruction, then evaluated by computed tomography (CT) scans scheduled for tumor staging, and clinically examined by a physiatrist. Numerous alterations in muscle physiology were observed due to surgical dissection of perforator vessels, and rectus muscle distress without functional impairment was a common result. Postoperatively, patients undergoing “contralateral reconstruction” appeared to exhibit fewer rectus muscle alterations. Overall, only three patients were impacted by a long-term deterioration in their quality of life. On the basis of the newly developed and implemented diagnostic approach, we concluded that DIEP microsurgical breast reconstruction is a safe procedure without major complications at the donor site, even if long-term alterations of the rectus muscle are a common finding.

Results of clinical application of the color contrasting method of digitalx-rays

Introduction. Color processing of X-ray images has a long history and initially was directed for improvement of analysis of medical diagnostic images. Purpose — to evaluate the results of clinical application of the method for color contrast enhancement of digital radiographs.Material and methods. The study was carried out in the X-ray department of the city Mariinsky hospital in St. Petersburg, having installed a computer program on the workstations of the radiologists for carrying out color contrasting of digital radiographs (CCDR). The CCDR program allowed the radiologist to select one of 63 trajectories in 8 colors. Choosing options for coloring, we settled on a warm, cold, full scale, as well as 4 combinations of one color from cold and warm scales with a saturation from 0 to 100%. The CCDR performed 100 digital radiographs of various anatomical areas.We analyzed a variety of colors and their percentage of saturation in terms of optimal transmission of pathological signs of anatomical areas. 27 radiologists assessed the possibilities of CCDR in X-ray diagnostics.Results. Clinical application of CCDR showed that thanks to this method, tissues of different densities were distinguished in more detail on the X-ray image, since their contours were more expressively emphasized. Pathological symptoms, indistinctly expressed on a black-and-white radiograph, were convincingly reflected in the colorized image, which increased the sensitivity and specificity of the diagnosis. Thanks to color post-processing, it was possible to optimize subtle radiological signs of structural bone changes, traumatic injuries of the ribs, impaired pneumatization of the lungs (infiltration, hypoventilation), pneumo-, hydrothorax, and others. Of the radiologists, 77% considered it important to use CCDR in X-ray diagnostics.Conclusion. A digital radiograph, contrasted with a color of optimal saturation, has distinct advantages over a traditional blackand-white X-ray image, since it reveals hidden or subtle diagnostic information. The diagnostic efficiency of the method of color contrasting of radiographs is higher than the analysis of black-and-white images up to 13%.To increase the diagnostic capabilities of X-ray diagnostics, it is advisable to include the method of color contrasting in the package of computer post-processing of images, using at least three gamuts in the standard, the saturation of which is 25–50%.The colorized image does not replace black and white, but complements it, resolving the diagnostic doubts of the radiologist.

Esquemas de diferencias finitas en el procesamiento de imágenes

Digital Image processing has been a research area of interest in the last decades, standing out for its applications in the analysis of diagnostic images and astronomical images. In this paper, we perform an overview of edge detection methods through finite-difference to present edge detection as a problem-based learning strategy for numerical differentiation, in order to improve the students’ skills in modeling and algorithmic thinking in numerical analysis courses. In addition, we present image restoration through finite-difference as a problem involving partial differential equations and software tools.

Socio-Economic Disparities in Access to Diagnostic Neuroimaging Services in the United Kingdom: A Systematic Review.

Socio-economic factors affecting health care can lead to delays in diagnosis of neurological conditions, consequentially affecting treatment and morbidity rates. This inequality in health care can leave patients from lower socio-economic backgrounds more vulnerable to a poorer quality of care from health care providers in the United Kingdom (U.K.). Aims: In this systematic review, we assess the impact of socio-economic status on the use of diagnostic neuroimaging in the U.K., measured by the timeliness, accessibility and appropriate use of computed tomography (CT), magnetic resonance imaging (MRI), ultrasonography, electroencephalography (EEG) and single-photon emission computed tomography (SPECT). We specifically evaluate the non-surgical use of neuroimaging techniques as this relies on the judgment of primary care-givers (e.g., doctors and radiologists), where health disparities are most common. This study includes the analysis of diagnostic imaging used for dementia, minor head injury, stroke, cancer, epilepsy, chronic inflammatory demyelinating polyneuropathy and Parkinson’s disease. With this study, we aim to assess the health inequalities at disease diagnosis. Methods: Using Medline (via Ovid), PubMed and Web of Science databases as sources of information, we critically appraise existing studies on neuroimaging use in the U.K. health care system, published between January 2010 and February 2021. Findings: A total of 18 studies were included in this research, revealing that there was an increase in patients of Black and Asian communities diagnosed with dementia and at an earlier age. There was little evidence to suggest that a lack of access to diagnostic imaging is associated with socio-economic status. However, there are data to suggest that people of a lower socio-economic background require more specialist services with diagnostic neuroimaging tools. In addition, there is evidence to suggest that diagnostic neuroimaging techniques could be utilised more effectively by health care workers to prevent unnecessary delays in diagnosis for patients in lower socio-economic areas.

Therapeutic effectiveness and safety of sintilimab-dominated triple therapy in unresectable hepatocellular carcinoma

Immune checkpoint inhibitor therapy has shown promising results in patients with unresectable hepatocellular carcinoma. This study aimed to evaluate the effectiveness and safety of sintilimab, a programmed cell death protein-1 (PD-1) blockade, combined with sorafenib and transhepatic arterial chemotherapy and embolization in this patient population, compared with sintilimab monotherapy and sintilimab-sorafenib duotherapy. This was a 22 months single center retrospective cohort study in China. 80 patients with unresectable hepatocellular carcinoma were included, with diagnosis confirmed by either histologic, cytologic or diagnostic imaging analysis. The patients were divided into three groups based on therapeutic regimen: sintilimab monotherapy (sintilimab group, n = 22), sintilimab-sorafenib duotherapy (duplex group, n = 23), sintilimab-sorafenib and transcatheter arterial chemoembolization combined therapy (triple group, n = 35). The principal evaluation criteria were overall survival and progression free survival in the population, assessed according to response evaluation criteria in solid tumors, version 1.1 (RECIST 1.1). Secondary evaluation criteria were safety, objective response rate and disease control rate. From March 1st, 2019 to December 31, 2020, 80 patients with unresectable hepatocellular carcinoma were included and divided into three treatment groups (22 received sintilimab monotherapy, 23 received sintilimab-sorafenib duotherapy, and 35 received sintilimab-sorafenib combined with transcatheter arterial chemoembolization). The median overall survival of all patients was 11.0 months (95% CI 7.7–14.3). Median overall survival was 13.0 months (95% CI NE–NE), 9.0 months(95% CI 6.3–11.7)and 3.0 months (95% CI 1.9–4.1, p < 0.0001) in the triple therapy, duplex and sintilimab groups respectively, while the corresponding median progression-free survival were 5.0 months (95% CI 2.9–7.1, p < 0.001), 4.0 months (95% CI 2.8–5.2) and 2.0 months (95% CI 1.7–2.3). Disease control and clinical benefits rates were higher in the triple therapy group (80%, 95% CI 63.1–91.6, p < 0.001; 54.3%, 95% CI 36.6–71.2, p < 0.01) compared to the sintilimab group. Median duration of disease control was 4.0 months (95% CI NE–NE, p < 0.01) in the triple therapy group, longer than that of the duplex group (2.0 months, 95% CI 0.9–3.1) and sintilimab group (2.0 months, 95% CI 0.8–3.2). Grade 3 or 4 treatment-related adverse events occurred in 26.3% of 80 patients with hypertension was the most common event observed (38, 47.5%), however, other severe toxic effects were infrequent. Sintilimab combined with sorafenib and transcatheter arterial chemoembolization might have more beneficial effects on overall and progression-free survival and on the duration of disease control outcomes than both sintilimab monotherapy and sintilimab-sorafenib duotherapy in patients with unresectable hepatocellular carcinoma. This triple therapy model might represent an innovative and effective option for inoperable liver cancer.

Case report: Buschke-Löwenstein tumor in a nonimmunosuppressed patient

The present work aims to report a case of Buschke-Lowenstein tumor (BLT), which is a mass of genital warts that usually affects immunosuppressed people. The reported case was diagnosed in a young patient with no known immunosuppression. Several tests were performed to confirm the diagnostic hypothesis, including immunohistochemistry, histological, molecular and imaging analysis. The results obtained were confirmatory in all analyses, except in the molecular one. Because BLT is a rare condition, there is still great literary heterogeneity regarding the ideal treatment, but some options can be considered, such as excision and radiotherapy.

Fundamental insights on ignition and combustion of natural gas in an active fueled pre-chamber spark-ignition system

Pre-chamber spark-ignition (PCSI), either fueled or non-fueled, is a leading concept with the potential to enable diesel-like efficiency in medium-duty (MD) and heavy-duty (HD) natural gas (NG) engines. However, the inadequate scientific base and simulation tools to describe/predict the underlying processes governing PCSI systems is one of the key barriers to market penetration of PCSI for MD/HD NG engines. To this end, experiments were performed in a heavy-duty, optical, single-cylinder engine fitted with an active fueled PCSI module. The spatial and temporal progress of ignition and subsequent combustion of lean-burn natural gas using PCSI system were studied using optical diagnostic imaging and heat release analysis based on main-chamber and pre-chamber pressure measurements.Optical diagnostics involving simultaneous infrared (IR) and high-speed (30 kfps) broadband and filtered OH* chemiluminescence imaging are used to probe the combustion process. Following the early pressure rise in the pre-chamber, IR imaging reveals initial ejection of unburnt fuel-air mixture from the pre-chamber into the main-chamber. Following this, the pre-chamber gas jets exhibit chemical activity in the vicinity of the pre-chamber region followed by a delayed spread in OH* chemiluminescence, as they continue to penetrate further into the main-chamber. The OH* signal progress radially until the pre-chamber jets merge, which sets up the limit to a first stage, jet-momentum driven, mixing-controlled (temperature field) premixed combustion. This is then followed by the subsequent deceleration of the pre-chamber jets, caused by the decrease in the driving pressure difference (ΔP) as well as charge entrainment, resulting in a flame front evolution, where mixing is not the only driver. Chemical-kinetic calculations probe the possibility of flame propagation or sequential auto-ignition in the second stage of combustion. Finally, key phenomenological features are then summarized so as to provide fundamental insights on the complex underlying fluid-mechanical and chemical-kinetic processes that govern the ignition and subsequent combustion of natural gas near lean-limits in high-efficiency lean-burn natural gas engines employing PCSI system.

Technology for High-Sensitivity Analysis of Medical Diagnostic Images.

Control and analysis of small, inaccessible to human vision changes in medical images make it possible to focus on diagnostically important radiological signs important for the correct diagnosis.The aim of the study was to develop information technology facilitating the early diagnosis of diseases using medical images.Materials and MethodsTo control changes in the image, we used its transformation based on solving a particular case of the knapsack problem. The proposed transformation is highly sensitive to any changes in the image and provides the possibility to record deviations visually with high accuracy. Medical images were obtained using cone beam computed tomography.ResultsPractical evaluation of the information technology on tomograms showed the following: the transformed images of healthy bone tissue fragments from different parts of the jaw have similar shapes and nearly the same range of brightness. The transformed image of bone tissue after treatment has a shape close to that of the transformed image of healthy bone tissue. The transformed image of the affected bone tissue has a shape and brightness range differing from the shape and color of the transformed images of healthy bone tissue and bone tissue after treatment. However, transformation of medical images obtained with the Planmeca ProMax 3D Classic device (Finland) allows recording changes that account for less than 0.0001% of the entire image.ConclusionThe proposed method allows human vision to capture changes as small as nearly one pixel in the transformed image, which is impossible with the original medical image. Increasing the color contrast of the transformed medical image makes it possible to reveal the structure of the analyzed medical image fragment. The proposed image transformation method can be used for early diagnosis of diseases and in other fields of medicine.

Analysis of Diagnostic Images of Artworks and Feature Extraction: Design of a Methodology.

Digital images represent the primary tool for diagnostics and documentation of the state of preservation of artifacts. Today the interpretive filters that allow one to characterize information and communicate it are extremely subjective. Our research goal is to study a quantitative analysis methodology to facilitate and semi-automate the recognition and polygonization of areas corresponding to the characteristics searched. To this end, several algorithms have been tested that allow for separating the characteristics and creating binary masks to be statistically analyzed and polygonized. Since our methodology aims to offer a conservator-restorer model to obtain useful graphic documentation in a short time that is usable for design and statistical purposes, this process has been implemented in a single Geographic Information Systems (GIS) application.

Abstract PO-074: The impact of phenotypic bias in the generalizability of deep learning models in non-small cell lung cancer

Abstract Although deep learning analysis of diagnostic imaging has shown increasing effectiveness in modeling non-small cell lung cancer (NSCLC) outcomes, a minority of proposed deep learning algorithms have been externally validated. Given a majority of these models are built on single institutional datasets, their generalizability across the entire population remains understudied. Moreover, the effect of biases that exist among institutional training dataset on overall generalizability of deep learning prognostic models is unclear. We attempted to identify demographic and clinical characteristics which if over-represented within training data could affect the generalizability of deep learning models aimed at predicting survival in patients with non-small cell lung cancer (NSCLC). Using a dataset of pre-treatment CT images of 422 patients diagnosed with non-small cell lung cancer (NSCLC), we examined deep learning model performance across demographic and tumor specific factors. Demographic factors of interest included age and gender. Clinical factors of interest included tumor histology, overall stage, T-Stage, and N-Stage. The effect of bias among training data was examined by varying the representation of demographic and clinical populations within the training and validation datasets. Model generalizability was measured by comparing AUC values among validation datasets (biased versus unbiased). AUC was estimated using 1,000 bootstrapped samples of 400 patients from validation cohorts. We found training datasets with biased representation of NSCLC histologist to be associated with greatest decrease in generalizability. Specifically, we found over-representation of adenocarcinoma within training datasets to be associated with an AUC reduction of 0.320 (0.296 - 0.344 CI, p&amp;lt;.001). Similarly over-representation of squamous cell carcinoma was associated with an AUC reduction of 0.177 (0.156 - 0.201 CI, p&amp;lt;.001). Biases in age (AUC 0.103, p&amp;lt;0.001), T stage (0.170, p=0.01 ), and N stage (0.120, p= 0.01) were also associated with reduced generalizability among deep learning models. Gender bias within training data was not associated with decreases in generalizability. Deep learning models of non-small cell lung cancer outcomes fail to generalize if trained on bias datasets. Specifically, overrepresentation of histologic subtypes may decrease the generalizability of deep learning models for NSCLC. Efforts to assure training data is representative of population demographics may lead to improved generalizability across more diverse patient populations. Citation Format: Aidan Gilson, Justin Du, Guneet Janda, Sachin Umrao, Marina Joel, Rachel Choi, Roy Herbst, Harlan Krumholz, Sanjay Aneja. The impact of phenotypic bias in the generalizability of deep learning models in non-small cell lung cancer [abstract]. In: Proceedings of the AACR Virtual Special Conference on Artificial Intelligence, Diagnosis, and Imaging; 2021 Jan 13-14. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(5_Suppl):Abstract nr PO-074.

Dynamic-Fusion-Based Federated Learning for COVID-19 Detection.

Medical diagnostic image analysis (e.g., CT scan or X-Ray) using machine learning is an efficient and accurate way to detect COVID-19 infections. However, the sharing of diagnostic images across medical institutions is usually prohibited due to patients’ privacy concerns. This causes the issue of insufficient data sets for training the image classification model. Federated learning is an emerging privacy-preserving machine learning paradigm that produces an unbiased global model based on the received local model updates trained by clients without exchanging clients’ local data. Nevertheless, the default setting of federated learning introduces a huge communication cost of transferring model updates and can hardly ensure model performance when severe data heterogeneity of clients exists. To improve communication efficiency and model performance, in this article, we propose a novel dynamic fusion-based federated learning approach for medical diagnostic image analysis to detect COVID-19 infections. First, we design an architecture for dynamic fusion-based federated learning systems to analyze medical diagnostic images. Furthermore, we present a dynamic fusion method to dynamically decide the participating clients according to their local model performance and schedule the model fusion based on participating clients’ training time. In addition, we summarize a category of medical diagnostic image data sets for COVID-19 detection, which can be used by the machine learning community for image analysis. The evaluation results show that the proposed approach is feasible and performs better than the default setting of federated learning in terms of model performance, communication efficiency, and fault tolerance.