|
A BENCHMARK OF DEEP-LEARNING BASED MODELS FOR BREAST CANCER DETECTION
|
ORTEGA FIGUEROA DAVID ESTEBAN 1, SUÁREZ BONILLA ERIKA YESENIA 2,
1 Universidad Industrial de Santander, 2 Universidad Industrial de Santander,
|
Email: erikasuarez.eysb@gmail.com
|
Abstract: Mammography is an imaging modality for the visualization of breast tissue. It is fast, noninvasive and an effective method for the early detection of breast cancer. Currently, there are multiple deep learning-based models in screening mammography with promising results in aiding detection and diagnosis tasks from mammography. These models vary widely in topology, training methods, and validation. However, these results sometimes depend on the use of test sets drawn from the same distribution as the training data, so it is common for some models to have difficulties in reproducibility, robustness and generalizability when implementation conditions change. For this reason, it is important to perform external validations to analyze the quality of the performance. In this work, AUC-ROC evaluation of five different models is performed with a private database, which has mammograms from both, breast cancer patients and healthy controls, to compare their respective performances.
|
Topic: Technologies for the diagnosis, detection or screening of cancer
|
|
|
APPLICATIONS IN BIOMEDICINE OF METALLIC NANOCOLUMNAR FILMS
|
GARCÍA MARTÍN JOSÉ MIGUEL 1,
1 Consejo Superior de Investigaciones Científicas,
|
Email: josemiguel.garcia.martin@csic.es
|
Abstract: In this talk, I will review different applications in biomedicine that we have developed using metallic nanocolumnar films (NCFs). Those films are fabricated by a physical method: glancing angle deposition using magnetron sputtering. This technique is environmentally friendly, since it is carried out at RT in a single step and does not involve chemical products (therefore, without recycling problems). Moreover, this strategy can be scaled up to large surfaces, representing a valid approach for the industrial production of nanostructured films [1]. After a brief introduction about the fabrication method, I will show several applications of these systems in biomedicine. In particular, Ti NCFs can be used as antibacterial coatings for orthopedic implants [1,2,3], Pt NCFs show improved properties as bioelectrodes for an electric stimulation platform in vitro [5], and Au NCFs are excellent substrates for the identification of biomolecules in Surface Enhanced Raman Spectroscopy, SERS [5,6].
References:
[1] R. Alvarez et al., Antibacterial Nanostructured Ti Coatings by Magnetron Sputtering: From Laboratory Scales to Industrial Reactors, Nanomaterials 9 (2019) 1217. DOI:10.3390/nano9091217
[2] I. Izquierdo-Barba et al., Nanocolumnar coatings with selective behavior towards osteoblast and Staphylococcus aureus proliferation, Acta Biomaterialia 15 (2015) 20.
DOI:10.1016/j.actbio.2014.12.023
[3] D. Medina-Cruz et al., Synergic antibacterial coatings combining titanium nanocolumns and tellurium nanorods, Nanomed.-Nanotechnol. Biol. Med. 17 (2019) 36. DOI:10.1016/j.nano.2018.12.009
[4] S. Mobini et al., Effects of nanostructuration on the electrochemical performance of metallic bioelectrodes, Nanoscale 14, (2022) 3179. DOI: 10.1039/D1NR06280H
[5] P. Díaz-Núñez et al., On the Large Near-Field Enhancement on Nanocolumnar Gold Substrates, Sci. Rep. 9 (2019) 13933. DOI: 10.1038/s41598-019-50392-w
[6] G. Barbillon et al., Gold Nanocolumnar Templates for Effective Chemical Sensing by Surface-Enhanced Raman Scattering, Nanomaterials, submitted (2022).
|
Topic: Modeling, image or signal processing in medical applications
|
|
|
BREAST CANCER SCREENING AND CARCINOGENIC FIELD EFFECT: A REVIEW OF THE STATE OF THE ART
|
JEREZ JULIAN 1,
1 Universidad Industrial de Santander,
|
Email: julian2228097@correo.uis.edu.co
|
Abstract: Breast cancer is the most common cancer among women worldwide; early detection is crucial in minimizing the progression to advanced stages of breast cancer and reducing associated mortality. Breast cancer screening is defined as testing women before the onset of noticeable symptoms. New imaging modalities have recently emerged besides traditional breast cancer screening techniques, such as mammography or ultrasound. Optical imaging is a technique that has been studied for breast cancer detection, as it has excellent advantages, such as non-invasiveness and non-toxicity. The detection of the field cancerization effect or field effect has been studied to detect abnormalities related to breast cancer. This effect consists of genetic or epigenetic variations in tissues of normal histological appearance. Field Effect Detection by Spectral Analysis (FEDSA) is a novel technique developed at the Universidad Industrial de Santander, based on field effect detection to detect abnormalities in breast tissue. This technique uses near-infrared (NIR) technology to irradiate the breast and acquire signals from the light backscattered by the breast tissue. This work aims to review the literature related to breast cancer screening. In addition, field cancerization and its use for breast cancer screening are presented.
|
Topic: Technologies for the diagnosis, detection or screening of cancer
|
|
|
BREAST LESION DETECTION ALGORITHM USING IDEAL SPEED OF SOUND IMAGES
|
PERTUZ SAID 1, VARGAS MOLANO ANDRES FELIPE 2, HERNÁNDEZ DURÁN ANGIE NICOLE 3, RAMIREZ ANA 4,
1 Universidad Industrial de Santander, 2 Universidad Industrial de Santander, 3 Universidad Industrial de Santander, 4 Universidad Industrial de Santander,
|
Email: anaberam@uis.edu.co
|
Abstract: New imaging techniques have been the subject of research in recent years. In breast cancer, one of the imaging techniques that attracts attention is Ultrasound Computed Tomography, which makes it possible to obtain images that quantitatively represent the acoustic properties (speed of sound, acoustic attenuation, density) of breast tissues. In contrast to standard imaging modalities that use ionizing radiation and require breast compression, such as mammography, this imaging modality is painless, non-invasive and radiation-free. However, ultrasound computed tomography is still in the development stage, so there are no real databases of these imaging modalities. In-silico clinical trials have been very well received in the investigation of new technologies since they allow algorithms to be explored safely and economically in the early stages of technological development. Under the hypothesis that medical images contain relevant information about the development of a disease that is not perceptible through human observation, the concept of radiomic analysis has been accepted in the field of oncology. The main objective of radiomic analysis is the extraction and analysis of different quantitative features of medical images. These features are typically used to train a learning algorithm to provide a detection score. In this work, we propose an algorithm for the detection of lesions in 3D synthetic breast phantoms from SOS (speed of sound) images. To this end, a set of 120 breast phantoms was first generated, of which 60 included a lesion (spiculated mass). Then, from the phantoms, the SOS images are obtained, this is done by replacing each of the voxel labels of each tissue with the speed of sound propagation value reported in the literature. Subsequently, a methodology for the detection of lesions was implemented, which consisted of the extraction of radiomic features at the cut-off level, which feed a classifier composed of a logistic regression and a weighting algorithm. The weighting algorithm consisted of a 3-tap moving average receiving the detection score of each slice as input and giving a detection score at the breast level as output. Finally, the performance of this SOS image-based classifier is evaluated in terms of AUC. Experiments validated with 5-fold cross-validation produced an AUC of 0.73 (95% CI: 0.64-0.82), 0.89 (95% CI: 0.83-0.95), and 0.94 (95% CI: 0.89-0.98) for radiomic analysis using SOS images with a pixel size of 1.5, 2.0, and 2.5 mm, respectively. Our results suggest that radiomic analysis of SOS images could help in the breast lesion detection task. The use of radiomic analysis using reconstructions of SOS images by ultrasound computed tomography should be further investigated.
|
Topic: Technologies for the diagnosis, detection or screening of cancer
|
|
|
CHALLENGES AND OPPORTUNITIES ASSOCIATED WITH CARRYING OUT A PILOT STUDY IN HEALTHCARE ENVIRONMENT
|
ROJAS BOHÓRQUEZ LEIDY JOHANA 1,
1 Universidad Industrial de Santander,
|
Email: fisjohanarb0911@gmail.com
|
Abstract: Scientific and technological advances in health represent a challenge not only in terms of interdisciplinary research development but also in moving towards stages in which new technologies are put into operation in the healthcare environment through studies that directly involve the participation of humans. The good clinical practice guide is an international standard of ethical and scientific quality for the design, conduct, performance, monitoring, auditing, registration, analysis, and reporting of clinical studies; this model guarantees that the reported data and results of the study are credible and accurate, protecting patients’ rights, integrity, and confidentiality. Furthermore, research applied to medicine carries out a series of challenges when faced with the reality of the clinical environment, like the training of the researchers involved, the logistics in the medical center, and the communication with the health staff and the patient. In this presentation, we will share the experience of the research team involved in the project "Estudio Piloto para la Evaluación Clínica de la Tecnología FEDSA en la Detección de Cáncer de Mama” at the empresa social del estado Hospital Universitario de Santander. Field effect detection by spectral analysis is a technology developed for the early detection of breast cancer. Field effect detection by spectral analysis is being evaluated clinically at the empresa social del estado Hospital Universitario de Santander, for which it was necessary to design a data collection protocol and its approval by both the ethics committee of the Universidad Industrial de Santander and of the Hospital Universitario de Santander. It was necessary to design a protocol to perform examinations with the field effect detection by spectral analysis technology in women who attend the Hospital Universitario de Santander to undergo the mammography examination; doctors, physicists, and systems and electronics engineers of different levels of training conform project research team. The research team had the opportunity to participate in an interdisciplinary project in a clinical environment where besides the skills of each area, the development of soft skills like communication, flexibility, leadership, motivation, patience, persuasion, problem-solving, and teamwork is important. This presentation aims to socialize the challenges faced in this project, so researchers can learn about experiences that could be used to anticipate possible situations and have a better understanding of the clinical environment in which they will perform, in addition to highlighting the importance of the interaction of the researcher with health staff and with the patient to meet the needs that they present and thus make research applied to medicine an activity more human and empathetic.
|
Topic: Technologies for the diagnosis, detection or screening of cancer
|
|
|
CHALLENGES FOR TECH INNOVATION IN HEALTH CARE SECTOR IN LATIN AMERICA
|
BUENO PAULO ROBERTO 1,
1 Universidade Estadual Paulista,
|
Email: paulo-roberto.bueno@unesp.br
|
Abstract: As an academic that has been engaged with technological entrepreneurial challenges in medical diagnostics, I will demonstrate that the creation of technological-based companies involves establishing a technologic cluster dynamic that involves the cooperation between academics and private sector agents. The model is well-known in the international context but largely ignored in Latin America. I will demonstrate that nanotechnology has a huge potential for business and obviously the successful implementation of a business in this healthcare sector involves establishing scientific and innovative policies that depend on government initiatives able of helping in the integration of the academics and private financial sectors.
|
Topic: Technologies for the diagnosis, detection or screening of cancer
|
|
|
CLINICAL NOTES ON CANCER SCREENING, DIAGNOSIS AND DETECTION FROM A RADIOLOGIST'S PERSPECTIVE
|
ARPONEN OTSO 1,
1 Tampere University Hospital,
|
Email: otso.arponen@tuni.fi
|
Abstract: Screening, diagnosis, and detection of cancers have pivotal roles in improving cancer-related outcomes. Cancer screening aims at detecting malignant tumors before the onset of symptoms, whereas the characterization of detected tumors into likely benign and malignant tumors impacts how doctors manage them; indeed, the detection of tumors links to both tasks. The current multimodal approach in radiology has improved the screening practices, diagnostics, and detection of cancers. I will discuss the current practices in cancer screening, diagnostics tumor detection in radiology in hopes that the audience may put the seminar's talks into clinical context; indeed, technological development paves the way for further discoveries and hopefully to even more refined clinical practices.
|
Topic: Technologies for the diagnosis, detection or screening of cancer
|
|
|
DETECTION OF AUTISM USING MULTILEVEL WAVELET DECOMPOSITION AND SUPPORT VECTOR MACHINES
|
CANCINO WILLIAM 1,
1 Universidad Industrial de Santander,
|
Email: william.cancino1998@gmail.com
|
Abstract: Autism spectrum disorder (ASD) is a neurodevelopmental disorder characterized by deficits in communication and social interaction, restricted activities and interest, and repetitive behavior patterns. The current diagnosis of ASD is highly subjective and can lead to inaccuracies, as it is based on purely behavioral observations. In addition, due to the complex symptoms of the disease, a proper diagnosis can take time. However, early and accurate detection of ASD is essential to implement treatments that improve the patient's condition and quality of life. For this reason, the combination of machine learning techniques and neuroimaging has become a promising candidate to reduce subjectivity and improve the diagnostic process. Specifically, resting-state functional magnetic resonance imaging (rs-fMRI) is of special interest as it allows the study of possible abnormalities in functional brain connectivity associated with autism. Thus, in our work we propose a classification framework for ASD based on Support Vector Machines (SVM) and Multilevel Discrete Wavelet Decomposition (MDWD). For this purpose, we used 175 rs-fMRI sequences from the Autism Brain Imaging Data Exchange dataset. From these images, we extract time series of regions of interest defined by a brain atlas. In the next stage, these time series are analyzed in different frequency bands by MDWD and the resulting subseries are used to build functional connectivity matrices. Finally, the features extracted from such matrices serve as input to the SVM classifier. The results of a 5-fold cross validation show that the use of MDWD in signal analysis provides an improvement in classifier performance. Our best model achieves accuracy, precision and area under the curve of 72.5%, 81.3% and 0.788, respectively.
|
Topic: Modeling, image or signal processing in medical applications
|
|
|
DEVELOPMENT OF A BIOSENSOR FOR THE RECOGNITION OF FOLATE RECEPTORS
|
BERTEL GARAY LINDA 1, OSPINA ROGELIO 2, GARCÍA MARTÍN JOSÉ MIGUEL 3, MIRANDA DAVID ALEJANDRO 4,
1 Universidad Industrial de Santander, 2 Universidad Industrial de Santander, 3 Consejo Superior de Investigaciones Científicas, 4 Universidad Industrial de Santander,
|
Email: dalemir@uis.edu.co
|
Abstract: The overexpression of folate receptors on the cell surface is related to an abnormality associated with epithelial cancer. In this sense, a capacitive biosensor was developed that uses folic acid as bio-recognition element for the detection of folate receptors. The detector consists mainly of a titanium-tungsten oxides thin film conjugated to folic acid, which functions as a working electrode in a three-electrode electrochemical cell configuration. The interactions between the biosensor and the folate receptors were studied by means of the determination of the chemical hardness from electrochemical capacitance spectroscopy measurements. The titanium-tungsten oxides thin film was fabricated by the pulsed laser deposition method and subsequently functionalized with folic acid. The characterization of the film before and after functionalization was performed by means of atomic force microscopy, X-ray emitted photoelectron spectroscopy, Raman spectroscopy and contact angle. Folate receptor recognition assays using the fabricated biosensor showed that the detector response signal, chemical hardness (in terms of electrochemical capacitance), is selectively and directly proportional to folate receptor concentration, with a limit of detection of 0.2 nM. This result is promising in the application of this type of biosensor for the recognition of folate receptors, especially for point-of-care analysis.
|
Topic: Technologies for the diagnosis, detection or screening of cancer
|
|
|
EFFECT OF FIELD CANCERIZATION ON BREAST TEXTURE FEATURES: AN IN SILICO STUDY
|
PERTUZ SAID 1, HERNÁNDEZ DURÁN ANGIE NICOLE 2, MIRANDA DAVID ALEJANDRO 3,
1 Universidad Industrial de Santander, 2 Universidad Industrial de Santander, 3 Universidad Industrial de Santander,
|
Email: dalemir@uis.edu.co
|
Abstract: Texture features extracted from medical images have been used extensively for diagnosis and risk assessment of diseases through a practice called radiomics. Radiomics applied to breast cancer risk assessment using mammography has shown remarkable performance, however, the explanation to this remains unknown. One hypothesis points at the concept of Field Cancerization Effect as the possible explanation to it. This effect has been studied at the molecular and optical levels, and it has been found that it produces alterations of the biochemical and optical properties of the tissues. Based on the hypothesis of Field Cancerization being the working principle of radiomics for breast cancer risk assessment, we performed an in silico experiment to test the effect of modifying the breast optical properties on the texture features extracted from mammograms. Our experiment started by introducing different levels of non-localized scattering centers in a cohort of 60 voxelized breast phantoms; from these phantoms, digital mammography was simulated, and the final mammographic images were used for the extraction of texture features. We assessed the impact of the tissue properties modification on the texture features using t-test, Wilcoxon signed-rank test, Kolmogorov-Smirnov test and an equivalence test. Our experiments showed that, when the non-localized scattering centers add up to 7.9% of the total breast volume, several of the texture features studied show statistically significant differences (p<0.05). These results indicate that changes in the properties of small, non-localized volumes of breast tissue can have a non-negligible impact on the texture features of mammograms, which supports the hypothesis of the Field Cancerization Effect being the working principle of radiomics for breast cancer risk assessment.
|
Topic: Mammographic image analysis
|
|
|
IMPROVING THE BREAST ULTRASOUND IMAGE RESOLUTION USING GENERATIVE ADVERSARIAL NETWORK
|
RAMIREZ ANA 1, REYES OSCAR M. 2, ABREO SERGIO A. 3, SOLANO JUAN CARLOS 4, VEGA EDWARD 5,
1 Universidad Industrial de Santander, 2 Universidad Industrial de Santander, 3 Universidad Industrial de Santander, 4 Universidad Industrial de Santander, 5 Universidad Industrial de Santander,
|
Email: edward.vega@correo.uis.edu.co
|
Abstract: For women with high risk of breast cancer, the American Cancer Society recommends a yearly breast screening using mammography or magnetic resonance, usually starting at 30 years old. However, breast cancer detection of young women using mammography is difficult for women with dense breast. In general, magnetic resonance can be used to confirm a diagnostic in young women, but this technique is highly costly. Breast ultrasound imaging is also a technique of low cost that is especially helpful in women with dense breast tissue. It is also a very safe technique because it does not expose a person to ionizing radiation. The major issue of the breast ultrasound imaging is the resolution of the images compared to the mammograms or magnetic resonance images. In this work, we propose the use of generative adversarial network-neural networks to improve the resolution of breast ultrasound. The generative adversarial network will be trained with a database having more than 3000 images of low-resolution ultrasound and its equivalent high-resolution ultrasound. The high-resolution ultrasound images are obtained from magnetic resonance images, where the tissues that are present in the images are identified and then converted to the ultrasound equivalent. The generative adversarial network-neural networks will be validated with around 1000 low-resolution ultrasound images.
|
Topic: Modeling, image or signal processing in medical applications
|
|
|
PREDICTIVE & PROGNOSTIC BIOMARKERS IN BREAST CANCER USING COMPUTATIONAL IMAGING PHENOTYPES AND ARTIFICIAL INTELLIGENCE
|
GASTOUNIOTI AIMILIA 1,
1 Washington University,
|
Email: a.gastounioti@wustl.edu
|
Abstract: Radiomics and artificial intelligence have undoubtedly expanded the utility of medical imaging in predictive and prognostics models in cancer research, and as a result they have also pervaded breast cancer screening as one of the most promising computerized breast imaging tools. In this presentation, I will share with you example research projects ranging from (a) imaging phenotypes extracted from mammography and their role in advancing breast cancer risk assessment; to (b) magnetic resonance imaging phenotypes of tumor heterogeneity predictive of breast cancer recurrence; and (c) prognostic models based on dynamic positron emission tomography imaging biomarkers.
|
Topic: Radiology, nuclear medicine and imaging
|
|
|
REGION OF INTEREST CLASSIFIER IN MAMMOGRAPHIC IMAGES USING LOGISTIC REGRESSION
|
CABEZA NATALIA 1, BRAVO MARIA ANGELICA 2,
1 Universidad Industrial de Santander, 2 Universidad Industrial de Santander,
|
Email: maria2182344@correo.uis.edu.co
|
Abstract: The early detection of breast cancer is important for providing opportune treatment and improve the possibilities of survival of the patient. For this reason, screening techniques are implemented, such as full-field digital mammography, allowing the radiologist to give a more precise diagnosis. Recently, machine learning has shown great potential for computer-aided diagnosis. It has been shown that these methodologies reduce the incidence of false positives since they are more accurate in the classification of the regions of interest. In this work, we propose a model for the classification of lesions in mammograms based on logistic regression where the predictor variables are features extracted by the software Openbreast.
For the implementation of the classifier, we use a public dataset named INbreast which contains 410 images and includes cases of women with both breasts affected and different lesions such as masses, calcifications, asymmetries, and distortions.
It also contains labels and annotations made by radiologists where regions of interest were previously selected. For training and testing the performance of the classifier, different regions of interest were selected and cropped for obtaining two subsets of 106 images each one, the first containing normal images and the other abnormal images. After this, texture and shape features of the regions are extracted using the Openbreast software. Texture features include different methods such as Gray-Level Co-Occurrence Matrix (GLCM), Gray-Level Run-Length Matrix (GLRL), Gray-Level Histogram Analysis (GLHA), and Gray-Level Sharpness Measurement. Additionally, the shape features were extracted based on mathematical descriptors of the region such as area, concavity, circularity, and perimeter. In total, 44 features were extracted, corresponding to 34 and 10 texture and shape features, respectively.
The proposed method was validated using 5-folds cross-validation. Performance was measured in terms of classification accuracy and are under the ROC curve (AUC), obtaining 73.1% and 0.78, respectively. In order to improve the results, further work on the exploration of extraction and selection of features is warranted.
|
Topic: Mammographic image analysis
|
|
|
TRANSFER LEARNING IN DATA-LIMITED SCENARIOS FOR BREAST CANCER RISK ASSESSMENT
|
AFRICANO GERSON 1,
1 Universidad industrial de santander,
|
Email: gersonf098@gmail.com
|
Abstract: Worldwide, breast cancer is the leading cause of cancer dead among women. Mammography is the leading screening image modality for early breast cancer detection, which has shown to reduce mortality rates. However, there is still plenty room for improvement where accurate risk assessment has the potential to improve early detection by allowing the creation of personalized screening recommendations.
In the last few years, AI systems have shown promise in identifying women with a high risk of developing breast cancer based on screening mammography exams. However, the development of AI systems requires the collection of large annotated datasets, which are costly and time-consuming to obtain. The access to a suitable dataset for AI development is one of the main limitation in the validation of the promise of the systems. In order to speed up the validation, it is imperative to evaluate alternative ways to develop AI systems with smaller datasets. In the literature, the transfer learning technique has been used to reduce the amount of data and improve the performance. Transfer learning broadly comprises two parts: first, pretraining the system in a large readily available dataset (source). Second, the pre-trained system is adjusted and re-trained in the target dataset. There is evidence suggesting that the impact of transfer learning depends on the similarity of the source and target datasets. However, transfer learning has been mainly evaluated in conditions where the source and the target dataset are entirely different, which hampers the impact of reducing the size of the target dataset.
In this work, we developed an AI system for breast cancer risk assessment in a data-limited scenarios. For this purpose, we use transfer learning from a source dataset similar to the target dataset. The target dataset corresponds to a case-control study with 286 women. 143 screening exams of women diagnosed with breast cancer (cases) and 143 screenings from healthy women (controls). We matched cases and controls by: mammographic system, screening year, and age. For transfer learning, we select a baseline system pre-trained over mammography images for breast cancer detection. The system was adjusted and retrained using our dataset for risk assessment. To assess the performance of the developed system, we use the area under the ROC curve (AUC) with 95% confidence interval (CI). For validation, we compare the developed system with state-of-the-art automatic systems for breast cancer risk assessment, such as breast density, OpenBreast, and Mirai. We used Delong's test to assess statistically significant differences.
The developed system yield to an AUC of 0.55 (95% CI 0.48-0.62). The evaluated state-of-the-art systems AUCs of 0.48 (0.41-0.55), 0.59 (0.52-0.65), and 0.60 (0.54-0.67) for breast density, OpenBreast, and Mirai, respectively. There was no statistically significant difference between the developed system and the state-of-the-art systems. Notice that, despite of the small dataset used for developing the AI system, we obtain similar performance to state-of-the-art systems, validating the potential of the considered strategy in scenarios where the collection/access to large datasets is challenging.
|
Topic: Mammographic image analysis
|
|
- PDF Download
