Computer-aided Image Analysis Research Articles

Survey on Adversarial Attack and Defense for Medical Image Analysis: Methods and Challenges

Deep learning techniques have achieved superior performance in computer-aided medical image analysis, yet they are still vulnerable to imperceptible adversarial attacks, resulting in potential misdiagnosis in clinical practice. Oppositely, recent years have also witnessed remarkable progress in defense against these tailored adversarial examples in deep medical diagnosis systems. In this exposition, we present a comprehensive survey on recent advances in adversarial attacks and defenses for medical image analysis with a systematic taxonomy in terms of the application scenario. We also provide a unified framework for different types of adversarial attack and defense methods in the context of medical image analysis. For a fair comparison, we establish a new benchmark for adversarially robust medical diagnosis models obtained by adversarial training under various scenarios. To the best of our knowledge, this is the first survey paper that provides a thorough evaluation of adversarially robust medical diagnosis models. By analyzing qualitative and quantitative results, we conclude this survey with a detailed discussion of current challenges for adversarial attack and defense in medical image analysis systems to shed light on future research directions. Code is available on GitHub.

Image Analysis in Histopathology and Cytopathology: From Early Days to Current Perspectives.

Both pathology and cytopathology still rely on recognizing microscopical morphologic features, and image analysis plays a crucial role, enabling the identification, categorization, and characterization of different tissue types, cell populations, and disease states within microscopic images. Historically, manual methods have been the primary approach, relying on expert knowledge and experience of pathologists to interpret microscopic tissue samples. Early image analysis methods were often constrained by computational power and the complexity of biological samples. The advent of computers and digital imaging technologies challenged the exclusivity of human eye vision and brain computational skills, transforming the diagnostic process in these fields. The increasing digitization of pathological images has led to the application of more objective and efficient computer-aided analysis techniques. Significant advancements were brought about by the integration of digital pathology, machine learning, and advanced imaging technologies. The continuous progress in machine learning and the increasing availability of digital pathology data offer exciting opportunities for the future. Furthermore, artificial intelligence has revolutionized this field, enabling predictive models that assist in diagnostic decision making. The future of pathology and cytopathology is predicted to be marked by advancements in computer-aided image analysis. The future of image analysis is promising, and the increasing availability of digital pathology data will invariably lead to enhanced diagnostic accuracy and improved prognostic predictions that shape personalized treatment strategies, ultimately leading to better patient outcomes.

Laser-assisted bioprinting of targeted cartilaginous spheroids for high density bottom-up tissue engineering

Multicellular spheroids such as microtissues and organoids have demonstrated great potential for tissue engineering applications in recent years as these 3D cellular units enable improved cell-cell and cell-matrix interactions. Current bioprinting processes that use multicellular spheroids as building blocks have demonstrated limited control on post printing distribution of cell spheroids or moderate throughput and printing efficiency. In this work, we presented a laser-assisted bioprinting approach able to transfer multicellular spheroids as building blocks for larger tissue structures. Cartilaginous multicellular spheroids formed by human periosteum derived cells (hPDCs) were successfully bioprinted possessing high viability and the capacity to undergo chondrogenic differentiation post printing. Smaller hPDC spheroids with diameters ranging from ∼100 to 150µm were successfully bioprinted through the use of laser-induced forward transfer method (LIFT) however larger spheroids constituted a challenge. For this reason a novel alternative approach was developed termed as laser induced propulsion of mesoscopic objects (LIPMO) whereby we were able to bioprint spheroids of up to 300µm. Moreover, we combined the bioprinting process with computer aided image analysis demonstrating the capacity to 'target and shoot', through automated selection, multiple large spheroids in a single sequence. By taking advantage of target and shoot system, multilayered constructs containing high density cell spheroids were fabricated.

Investigation on particle size and packing tortuosity by coupling image analysis and permeability tests

This experimental study aims to enhance the understanding of the correlation among equivalent particle diameters measured using two analytical techniques: optical analysis (assisted by computer aided image analysis) and permeability tests. The presence or absence of a specific analytical method or instrument can lead to the use of an incorrect equivalent diameter. Therefore, it can be beneficial to establish conversion rules between different equivalent particle diameters obtained through various methods and instruments. The optical analysis returns an equivalent diameter value inherently independent of particle arrangement since it deals with isolated particles. In contrast, the permeability test offers an equivalent mean diameter dependent not only on the size of the particles but also on their packed arrangement. A suitable correlation between the two diameters has been proposed, shown to be a decreasing function of porosity following a power law.An unexpected outcome of the comparison between the optical method and permeametry is the possibility to isolate and characterize the effect that the packing arrangement has on pressure losses and to characterize it in terms of the tortuosity of the path that the fluid must travel through the packed bed. Our findings confirm a strong alignment between our tortuosity model, which contains the ratio between the two equivalent diameters considered here, and an empirical correlation from literature often utilized for predicting packed bed tortuosity.

KRC-APM: Key region cutting and artificial prior model for breast cancer recognition in ultrasound images

Computer-aided analysis of ultrasound images is important for the early detection of breast cancer, obtaining more treatment time and improving the likelihood of survival. Current methods are either limited by predefined Regions of Interest (ROIs), suffer from a low proportion and variable locations of tumors, or lack appropriate modeling of tumor characteristics in ultrasound images. This paper proposes a framework called Key Region Cutting and Artificial Prior Model (KRC-APM) for breast cancer recognition in ultrasound images. Firstly, to avoid the reliance on predefined ROIs, as well as to increase the proportion of tumor areas and eliminate abundant background, we propose a Key Region Cutting (KRC) method, which analyzes the confidence levels for tumor areas and tolerantly cuts high-confidence areas into key regions. Next, to model the characteristics of tumors in ultrasound images, we design three types of diagnosis-related Artificial Priors (APs) under the guidance of experienced oncologists: Shape Modeling, Margin Analysis, and Echogenicity Pattern Indication. Finally, to fuse the key region with the extracted APs, we propose an Artificial Prior Model (APM), which synthetically learns the pattern of breast cancer. The proposed KRC-APM is validated on two public datasets. The experimental results demonstrate that the key regions identified by the KRC method and the designed APs significantly enhance the performance of breast cancer recognition. Furthermore, the proposed KRC-APM framework outperforms current state-of-the-art (SOTA) methods.

Automatic segmentation of spine x-ray images based on multiscale feature enhancement network.

Automatic segmentation of vertebrae in spinal x-ray images is crucial for clinical diagnosis, case analysis, and surgical planning of spinal lesions. However, due to the inherent characteristics of x-ray images, including low contrast, high noise, and uneven grey scale, it remains a critical and challenging problem in computer-aided spine image analysis and disease diagnosis applications. In this paper, a Multiscale Feature Enhancement Network (MFENet), is proposed for segmenting whole spinal x-ray images, to aid doctors in diagnosing spinal-related diseases. To enhance feature extraction, the network incorporates a Dual-branch Feature Extraction Module (DFEM) and a Semantic Aggregation Module (SAM). The DFEM has a parallel dual-branch structure. The upper branch utilizes multiscale convolutional kernels to extract features from images. Employing convolutional kernels of different sizes helps capture details and structural information at different scales. The lower branch incorporates attention mechanisms to further optimize feature representation. By modeling the feature maps spatially and across channels, the network becomes more focused on key feature regions and suppresses task-irrelevant information. The SAM leverages contextual semantic information to compensate for details lost during pooling and convolution operations. It integrates high-level feature information from different scales to reduce segmentation result discontinuity. In addition, a hybrid loss function is employed to enhance the network's feature extraction capability. In this study, we conducted a multitude of experiments utilizing dataset provided by the Spine Surgery Department of Henan Provincial People's Hospital. The experimental results indicate that our proposed MFENet demonstrates superior segmentation performance in spinal segmentation on x-ray images compared to other advanced methods, achieving 92.61±0.431 for MIoU, 92.42±0.329 for DSC, and 99.51±0.037 for Global_accuracy. Our model is able to more effectively learn and extract global contextual semantic information, significantly improving spinal segmentation performance, further aiding doctors in analyzing patient conditions.

Continuous specimen cooling during slicing and thickness measurement contributes to improved accuracy in image analysis of pathologic specimens.

Sections were prepared from a paraffin-only block and formalin-fixed paraffin-embedded (FFPE) blocks containing fish sausage and human liver specimens. The ST was compared between the sections prepared with cooling using an ice pack (IP) or a continuous cooling device (CCD) paired with a sliding microtome set at an ST of 4 µm. The sections were stained with eosin or aniline blue, and the association between the percent area of positive staining and ST was determined using computer-aided analysis of images captured with a whole slide scanner. The average STs of the paraffin-only block sections measured by four practitioners were 5.01-5.41 and 4.09-4.33 µm in samples prepared using an IP and a CCD, respectively. Therefore, subsequent analyses included sections prepared using the CCD. The ST of the tissue surface was significantly thinner than that of the paraffin surrounding the tissue section. Furthermore, the percent areas of positive staining for eosin and aniline blue were significantly correlated with ST in both the fish sausage and liver sections. The analysis of the ST and percent area of positive staining in 60 sections of the same block, which were categorized into quantiles based on ST, revealed a significant difference in the percent area of positive staining between the thicker and thinner sections. Specimen sectioning should be performed with a CCD, ST should be measured before the staining of pathologic specimens prepared for quantitative analysis, and histologic examination should be performed using specimens with uniform ST.

Early diagnosis of skin oncologic diseases using artificial intelligence technologies

The last decade has seen significant progress in computer-aided image analysis and recognition, with modern computer-aided diagnostic algorithms not only catching up with, but in many aspects surpassing human abilities. At the heart of this breakthrough is the development of deep convolutional neural networks, which have given a new impetus to medical diagnosis, particularly of skin cancers. In this paper, we analyzed photo-based skin disease classification systems developed using algorithms based on deep learning convolutional neural networks. Such methods have been variously reported to enable automated diagnosis of skin neoplasms with high sensitivity and specificity. A disease that requires more detailed analysis of graphic images — skin melanoma — was chosen as the main object of study. Early diagnosis of melanoma is of great socio-economic importance, as in this case the prognosis of patients is significantly improved. The aim of this work is to analyze the results of artificial intelligence (AI) applications in dermatology, especially in the context of early detection of skin melanoma. Scientific articles were searched in PubMed, Scopus and eLIBRARY databases using the keywords “convolutional neural networks”, “skin cancer” and “artificial intelligence”. The depth of the search was 10 years. The final analysis included 38 sources where the results of our own research were presented. The advantages of artificial intelligence methods for dermatologists were analyzed. Artificial intelligence can significantly assist dermatologists in developing visual neoplasm diagnosis skills and improve diagnostic accuracy. The use of AI to process dermatoscopic data in conjunction with the analysis of anamnestic and clinical information from medical records will reduce the burden on the healthcare system through correctly diagnosed benign skin tumors. All of this promises to have a significant impact on the future development of dermatovenerology.

Standardization of honey as a tissue fixative for histopathology: A morphometric study

Background: Tissue fixation is a crucial step to preserve the tissues in a life-like state with minimal disruption to its cellular and chemical composition for histopathological examination. The search for an effective alternate tissue fixative to the routinely used formaldehyde has gained interest as constant exposure to formaldehyde has proven to be toxic. Honey, an organic substance with high acidity and hygroscopic nature, exhibits tissue fixative properties and has been used in the present study. The present study aimed to standardize honey as a tissue fixative for histopathology by comparing it with formalin. Materials and Methods: In vitro study Oral tissue samples of goat were fixed in 10% honey and 10% formalin solution, respectively, for 24-48 h, followed by routine tissue processing and microscopic examination of 37 slides per group. 2200 epithelial cells (1100 per group) were selected for the computer-aided morphometric image analysis (Fiji-Image J) by three observers. Cell area (CA), cell perimeter (CP), nuclear area (NA), nuclear perimeter (NP), cytoplasmic area (Cyt A), and nuclear-cytoplasmic ratio were the parameters studied. Mann-Whitney U-test (STATA/IC version 16) for inter-group comparison was done and P < 0.05 was considered statistically significant. Results: The probability of epithelial cells in the honey-fixed group to have greater NA, NP, and N/C ratio was about 50%-60%. The probability of epithelial cells in formalin-fixed tissues to have greater CA, CP, and Cyt A was about 70%. Conclusion: Honey is a better nuclear fixative than formalin. Cytoplasmic shrinkage of epithelial cells should be taken into consideration while fixing tissues with honey.

Investigating the Spatial Distribution of Proliferating Cells in Primary Ovarian Cancers.

Ovarian cancer (OC) accounts for about 4% of female cancers globally. While Ki67-immunopositive (Ki67+) cell density is commonly used to assess proliferation in OC, the two-dimensional (2D) distribution pattern of these cells is poorly understood. This study explores the 2D distribution pattern of Ki67+ cells in primary OC tissues and models the proliferation process to improve our understanding of this hallmark of cancer. A total of 100 tissue cores, included in a tissue microarray (TMA) representing 5 clear cell carcinomas, 62 serous carcinomas, 10 mucinous adenocarcinomas, 3 endometrioid adenocarcinomas, 10 lymph node metastases from OC, and 10 samples of adjacent normal ovary tissue, were stained using a standardized immunohistochemical protocol. The computer-aided image analysis system assessed the 2D distribution pattern of Ki67+ proliferating cells, providing the cell number and density, patterns of randomness, and cell-to-cell closeness. Three computer models were created to simulate behavior and responses, aiming to gain insights into the variations in the proliferation process. Significant differences in Ki67+ cell density were found between low- and high-grade serous carcinoma/mucinous adenocarcinomas (p = 0.003 and p = 0.01, respectively). The Nearest Neighbor Index of Ki67+ cells differed significantly between high-grade serous carcinomas and endometrioid adenocarcinomas (p = 0.01), indicating distinct 2D Ki67+ distribution patterns. Proxemics analysis revealed significant differences in Ki67+ cell-to-cell closeness between low- and high-grade serous carcinomas (p = 0.002). Computer models showed varied effects on the overall organization of Ki67+ cells and the ability to preserve the original 2D distribution pattern when altering the location and/or density of Ki67+ cells. Cell proliferation is a hallmark of OCs. This study provides new evidence that investigating the Ki67+ cell density and 2D distribution pattern can assist in understanding the proliferation status of OCs. Moreover, our computer models suggest that changes in Ki67+ cell density and their location are critical for maintaining the 2D distribution pattern.

Processing-microstructure correlations in material extrusion additive manufacturing of carbon fiber reinforced ceramic matrix composites

Liquid silicon infiltration is a cost-effective manufacturing route for carbon fiber reinforced silicon carbide (C/C-SiC)11C/C-SiC: carbon fiber reinforced silicon carbide, which is a ceramic matrix composite with high specific stiffness associated with damage tolerance. To reduce the cost-intensive machining while improving the geometric freedom of design, a processing route for C/C-SiC based on material extrusion additive manufacturing was investigated. Filament-based material extrusion was applied to manufacture green bodies made of short carbon fiber reinforced polyetheretherketone (PEEK)22PEEK: polyetheretherketone. Additionally, a thermo-oxidative crosslinking step, which is assumed to be a function of the specific surface area, was introduced prior to pyrolysis to prevent the re-melting of PEEK. Via computer-aided image analysis, the influence of process parameters during material extrusion on the green bodies’ specific surface area were investigated using statistical design of experiments. It was shown that it is possible to selectively adjust the specific surface area by modifying the microstructure of the carbon fiber reinforced green bodies and hence the shrinkage during pyrolysis, the liquid silicon infiltration kinetics as well as the phase composition of the obtained C/C-SiC. The understanding of the processing-microstructure correlations enabled an improvement of the mechanical properties. With a layer height of 0.1 mm, an infill density of 100% and a printing direction of ± 45°, a flexural strength of 80.3 ± 6.7 MPa was measured.

Machine learning in the detection of dental cyst, tumor, and abscess lesions

Background and ObjectiveDental panoramic radiographs are utilized in computer-aided image analysis, which detects abnormal tissue masses by analyzing the produced image capacity to recognize patterns of intensity fluctuations. This is done to reduce the need for invasive biopsies for arriving to a diagnosis. The aim of the current study was to examine and compare the accuracy of several texture analysis techniques, such as Grey Level Run Length Matrix (GLRLM), Grey Level Co-occurrence Matrix (GLCM), and wavelet analysis in recognizing dental cyst, tumor, and abscess lesions.Materials & MethodsThe current retrospective study retrieved a total of 172 dental panoramic radiographs with lesion including dental cysts, tumors, or abscess. Radiographs that failed to meet technical criteria for diagnostic quality (such as significant overlap of teeth, a diffuse image, or distortion) were excluded from the sample. The methodology adopted in the study comprised of five stages. At first, the radiographs are improved, and the area of interest was segmented manually. A variety of feature extraction techniques, such GLCM, GLRLM, and the wavelet analysis were used to gather information from the area of interest. Later, the lesions were classified as a cyst, tumor, abscess, or using a support vector machine (SVM) classifier. Eventually, the data was transferred into a Microsoft Excel spreadsheet and statistical package for social sciences (SPSS) (version 21) was used to conduct the statistical analysis. Initially descriptive statistics were computed. For inferential analysis, statistical significance was determined by a p value < 0.05. The sensitivity, specificity, and accuracy were used to find the significant difference between assessed and actual diagnosis.ResultsThe findings demonstrate that 98% accuracy was achieved using GLCM, 91% accuracy using Wavelet analysis & 95% accuracy using GLRLM in distinguishing between dental cyst, tumor, and abscess lesions. The area under curve (AUC) number indicates that GLCM achieves a high degree of accuracy. The results achieved excellent accuracy (98%) using GLCM.ConclusionThe GLCM features can be used for further research. After improving the performance and training, it can support routine histological diagnosis and can assist the clinicians in arriving at accurate and spontaneous treatment plans.

A Comprehensive Review and Analysis of Deep Learning-Based Medical Image Adversarial Attack and Defense

Deep learning approaches have demonstrated great achievements in the field of computer-aided medical image analysis, improving the precision of diagnosis across a range of medical disorders. These developments have not, however, been immune to the appearance of adversarial attacks, creating the possibility of incorrect diagnosis with substantial clinical implications. Concurrently, the field has seen notable advancements in defending against such targeted adversary intrusions in deep medical diagnostic systems. In the context of medical image analysis, this article provides a comprehensive survey of current advancements in adversarial attacks and their accompanying defensive strategies. In addition, a comprehensive conceptual analysis is presented, including several adversarial attacks and defensive strategies designed for the interpretation of medical images. This survey, which draws on qualitative and quantitative findings, concludes with a thorough discussion of the problems with adversarial attack and defensive mechanisms that are unique to medical image analysis systems, opening up new directions for future research. We identified that the main problems with adversarial attack and defense in medical imaging include dataset and labeling, computational resources, robustness against target attacks, evaluation of transferability and adaptability, interpretability and explainability, real-time detection and response, and adversarial attacks in multi-modal fusion. The area of medical imaging adversarial attack and defensive mechanisms might move toward more secure, dependable, and therapeutically useful deep learning systems by filling in these research gaps and following these future objectives.

Resiliency and Risk Assessment of Smart Vision-Based Skin Screening Applications with Dynamics Modeling

The prevalence of skin diseases remains a concern, leading to a rising demand for the advancement of smart, portable, and non-invasive automated systems and applications. These sought-after technologies allow for the screening of skin lesions through captured images, offering improved and accessible healthcare solutions. Clinical methods include visual inspection by dermatologists; computer-aided vision-based image analysis at healthcare settings; and, lastly, biopsy tests, which are often costly and painful. Given the rise of artificial intelligence-based techniques for image segmentation, analysis, and classification, there remains a need to investigate the resiliency of personalized smartphone (hand-held) skin screening systems with respect to identified risks. This study represents a unique integration of distinct fields pertaining to smart vision-based skin lesion screening, resiliency, risk assessment, and system dynamics. The main focus is to explore the dynamics within the supply chain network of smart skin-lesion-screening systems. With the overarching aim of enhancing health, well-being, and sustainability, this research introduces a new framework designed to evaluate the resiliency of smart skin-lesion-screening applications. The proposed framework incorporates system dynamics modeling within a novel subset of a causal model. It considers the interactions and activities among key factors with unique mapping of capability and vulnerability attributes for effective risk assessment and management. The model has been rigorously tested under various case scenarios and settings. The simulation results offer insights into the model’s dynamics, demonstrating the fact that enhancing the skin screening device/app factors directly improves the resiliency level. Overall, this proposed framework marks an essential step toward comprehending and enhancing the overall resiliency of smart skin-lesion-screening systems.

An Objective Computer-Assisted Measurement of Sonographic Renal Cortical Echogenicity: The Splenorenal Index.

Renal cortical echogenicity represents a marker of renal function. However, evaluation of the renal echotexture is subjective and thus disposed to error and interrater variability. Computer-aided image analysis may be used to objectively assess renal cortical echogenicity by comparing the echogenicity of the left kidney to that of the spleen; the resultant ratio is referred to as the splenorenal index (SRI). We performed a retrospective review of all adult patients who received a renal ultrasound over a 45-day period at our institution. Demographic data and kidney function laboratory values were documented for each patient. Regions of interest (ROIs) were selected in the left renal cortex and spleen using ImageJ software. The SRI was calculated as a ratio of the mean pixel brightness of the left kidney cortex ROI to the mean pixel brightness of the spleen ROI. The SRI was then correlated with serum creatinine, blood urea nitrogen, and estimated glomerular filtration rate. We found that among the 94 patients included in the study, the SRI had a significant positive correlation with serum creatinine ( r = 0.43, P < 0.001) and serum blood urea nitrogen ( r = 0.45, P < 0.001) and negative correlation with estimated glomerular filtration rate ( r = -0.47, P < 0.001). Our data indicate that SRI may serve as a valuable tool for sonographic evaluation of renal parenchymal disease.

Histopathological Images Analysis and Predictive Modeling Implemented in Digital Pathology-Current Affairs and Perspectives.

In modern clinical practice, digital pathology has an essential role, being a technological necessity for the activity in the pathological anatomy laboratories. The development of information technology has majorly facilitated the management of digital images and their sharing for clinical use; the methods to analyze digital histopathological images, based on artificial intelligence techniques and specific models, quantify the required information with significantly higher consistency and precision compared to that provided by optical microscopy. In parallel, the unprecedented advances in machine learning facilitate, through the synergy of artificial intelligence and digital pathology, the possibility of diagnosis based on image analysis, previously limited only to certain specialties. Therefore, the integration of digital images into the study of pathology, combined with advanced algorithms and computer-assisted diagnostic techniques, extends the boundaries of the pathologist's vision beyond the microscopic image and allows the specialist to use and integrate his knowledge and experience adequately. We conducted a search in PubMed on the topic of digital pathology and its applications, to quantify the current state of knowledge. We found that computer-aided image analysis has a superior potential to identify, extract and quantify features in more detail compared to the human pathologist's evaluating possibilities; it performs tasks that exceed its manual capacity, and can produce new diagnostic algorithms and prediction models applicable in translational research that are able to identify new characteristics of diseases based on changes at the cellular and molecular level.

Exploring the Power of Deep Learning: Fine-Tuned Vision Transformer for Accurate and Efficient Brain Tumor Detection in MRI Scans.

A brain tumor is a significant health concern that directly or indirectly affects thousands of people worldwide. The early and accurate detection of brain tumors is vital to the successful treatment of brain tumors and the improved quality of life of the patient. There are several imaging techniques used for brain tumor detection. Among these techniques, the most common are MRI and CT scans. To overcome the limitations associated with these traditional techniques, computer-aided analysis of brain images has gained attention in recent years as a promising approach for accurate and reliable brain tumor detection. In this study, we proposed a fine-tuned vision transformer model that uses advanced image processing and deep learning techniques to accurately identify the presence of brain tumors in the input data images. The proposed model FT-ViT involves several stages, including the processing of data, patch processing, concatenation, feature selection and learning, and fine tuning. Upon training the model on the CE-MRI dataset containing 5712 brain tumor images, the model could accurately identify the tumors. The FT-Vit model achieved an accuracy of 98.13%. The proposed method offers high accuracy and can significantly reduce the workload of radiologists, making it a practical approach in medical science. However, further research can be conducted to diagnose more complex and rare types of tumors with more accuracy and reliability.

Morphometric Features of Cells of Cervical Intraepithelial Neoplasia Using Computer Aided Image Analysis

Aim: Our goal is to evaluate and compare the different cell morphometric parameters in the various categories of cervical intraepithelial neoplasia (CIN).&#x0D; Study Design: We conducted a cross-sectional study of all cervical intraepithelial neoplasia identified in some areas of biopsies of cervical carcinomas.&#x0D; Place and Duration of Study: The study was in the department of histopathology of LAUTECH Teaching Hospital, Osogbo from January 2006 to December 2015.&#x0D; Materials and Methods: The histologic slides produced were scanned using CS2 APERIO digital slide scanner with standard settings. The scanned slides were then viewed using Qu Path-0.3.2 software. All areas of cervical intraepithelial neoplasia were selected. The perimeter, area, maximum caliper, circularity, eccentricity, and the various measures of hematoxylin optic densities (OD) of the nucleus of the cells were measured. The cytoplasmic morphometric measurements and the nuclear-cell ratio were then estimated.&#x0D; Results: Two cases of CIN 1, one case of CIN 2 and one case of CIN 3 were seen in the background of the eighty-seven cases of cervical carcinomas seen during the study period. A total of 5,935 cells were in the annotated regions with cervical intraepithelial neoplasia. The cells in CIN 1 area are 3,828, CIN 2 area has 1183 cells and CIN 3 areas has 924 cells. Cells of CIN 3 had higher average nuclei hematoxylin optic density (OD), nuclei hematoxylin OD standard deviation and maximum hematoxylin OD than cells of CIN 2 and CIN 1. Cells of CIN 1 have higher average cell area, cell perimeter, cell circularity and mean cytoplasm eosin OD compared to cells of CIN 2 and 3.&#x0D; Conclusion: Computer-aided image analysis using Qu Path software can be very instrumental in differentiating the various categories of cervical intraepithelial neoplasia especially in fragmented biopsies.

Abstract P3-03-25: An Automated Risk Stratification System for Breast Cancer Screening using Thermalytix

Abstract Background: As opposed to conventional age-based population-level breast screening strategies, risk-stratified breast screening programs are emerging as a new approach to balanced population screening methodology where the screening frequency and choice of modality (mammography/tomosynthesis or magnetic resonance imaging) is determined based on accurate personalized estimation of an individual’s risk score. A woman identified with low risk can now be screened less frequently, avoiding repeated mammography screening where radiation risk outweighs the benefits in that particular individual. The standard questionnaire based risk stratification is found to be less reliable and imaging based risk stratification mechanisms are being explored in recent years. In this study, we evaluate the performance of a new computer-aided image analysis technique called Thermalytix that automatically generates a personalized risk score using a combination of imaging and questionnaire information for risk stratification of women. Methodology: Thermalytix is an artificial intelligence system that uses thermal imaging and questionnaire data for predicting the risk of breast cancer. Thermalytix analyzes spatio-thermal signatures and vascular patterns in the breast region along with patients’ complaints and age to generate a score called B-Score (or BHARATI Score) which ranges from 1 to 5. B-Score of 1 indicates low risk of malignancy and a B-Score of 5 indicates the highest risk of malignancy. To evaluate the effectiveness of risk stratification using B-Scores, we performed retrospective analysis of thermal and participants’ data acquired from two registered clinical studies. One study (CTRI/2017/10/0 10 115) is a multi-site study conducted in Bangalore, India, and the other study (NCT04688086) is a single site study conducted in Delhi, India. Both these study sites are geographically distant with 2000 KM apart from each other and comprise a diversified population from India. Results: In total, 717 eligible women were considered in this study with age varying from 18 years to 80 years. Reports from standard of care procedures involving mammography, ultrasound and biopsy (as needed), were collectively considered by a radiologist to determine the ground truth for malignancy. Out of 717 women, 85 women were thus concluded as malignant. When used in a blinded fashion, Thermalytix graded 275 women as B-Score 1 (lowest risk), 225 women as B-Score 2, 44 women as B-Score 3, 137 women as B-Score 4 and 36 women as B-Score 5 (highest risk). The fraction of malignancies in the cohorts corresponding to B-Score categories from 1 to 5 were found to be progressively higher (0.36%, 1.33%, 29.55%, 33.58% and 61.11%, respectively) - showing the correctness of the proposed personalized risk scoring methodology. Conclusion: Thermalytix test, a low-cost, radiation-free, contactless and privacy aware test was used as a technique to determine the breast cancer risk of a woman.. The results obtained in the study show that a high B-score of 5 indicates a high risk for malignancy with 61.11% chance of breast cancer. Likewise the lowest B-Score of 1 indicates low risk for malignancy with just 0.36% percentage of women in the cohort found with malignancy. These results combined with other experiential benefits of Thermalytix test makes it a promising risk stratification mechanism enabling differential frequency of screening while balancing the cost and risk versus benefit. Large scale studies, however, need to be conducted to see the ground benefits of the proposed approach in a screening program implementation. Distribution of study population in different risk cohorts Higher risk correlates with higher malignancy rate Citation Format: Siva Teja Kakileti, Himanshu Madhu, Richa Bansal, Akshita Singh, Sudhakar Sampangi, Bharat Aggarwal, Geetha Manjunath. An Automated Risk Stratification System for Breast Cancer Screening using Thermalytix [abstract]. In: Proceedings of the 2022 San Antonio Breast Cancer Symposium; 2022 Dec 6-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2023;83(5 Suppl):Abstract nr P3-03-25.

Unsupervised anomaly detection in brain MRI: Learning abstract distribution from massive healthy brains

PurposeTo develop a general unsupervised anomaly detection method based only on MR images of normal brains to automatically detect various brain abnormalities. Materials and methodsIn this study, a novel method based on three-dimensional deep autoencoder network is proposed to automatically detect and segment various brain abnormalities without being trained on any abnormal samples. A total of 578 normal T2w MR volumes without obvious abnormalities were used for model training and validation. The proposed 3D autoencoder was evaluated on two different datasets (BraTs dataset and in-house dataset) containing T2w volumes from patients with glioblastoma, multiple sclerosis and cerebral infarction. Lesions detection and segmentation performance were reported as AUC, precision-recall curve, sensitivity, and Dice score. ResultsIn anomaly detection, AUCs for three typical lesions were as follows: glioblastoma, 0.844; multiple sclerosis, 0.858; cerebral infarction, 0.807. In anomaly segmentation, the mean Dice for glioblastomas was 0.462. The proposed network also has the ability to generate an anomaly heatmap for visualization purpose. ConclusionOur proposed method was able to automatically detect various brain anomalies such as glioblastoma, multiple sclerosis, and cerebral infarction. This work suggests that unsupervised anomaly detection is a powerful approach to detect arbitrary brain abnormalities without labeled samples. It has the potential to support diagnostic workflow in radiology as an automated tool for computer-aided image analysis.