RESUMO
Artificial Intelligence (AI) has revolutionized many fields, including medical imaging. This revolution has enabled the digitization of medical images, the development of algorithms allowing the use of data captured in natural language, and deep learning, enabling the development of algorithms for automatic processing of medical images from massive medical data. In Burkina Faso, early and accurate detection of breast cancer is a significant challenge due to limited resources and lack of specialized expertise. In this article, we examine the effectiveness of different artificial intelligence algorithms for breast cancer detection from pathological image.
Assuntos
Algoritmos , Inteligência Artificial , Neoplasias da Mama , Neoplasias da Mama/diagnóstico por imagem , Humanos , Feminino , Burkina Faso , Interpretação de Imagem Assistida por Computador/métodosRESUMO
The second most common type of malignant tumor worldwide is colorectal cancer. Histopathology image analysis offers crucial data for the clinical diagnosis of colorectal cancer. Currently, deep learning techniques are applied to enhance cancer classification and tumor localization in histopathological image analysis. Moreover, traditional deep learning techniques might loss integrated information in the image while evaluating thousands of patches recovered from whole slide images (WSIs). This research proposes a novel colorectal cancer detection network (CCDNet) that combines coordinate attention transformer with atrous convolution. CCDNet first denoises the input histopathological image using a Wiener based Midpoint weighted non-local means filter (WMW-NLM) for guaranteeing precise diagnoses and maintain image features. Also, a novel atrous convolution with coordinate attention transformer (AConvCAT) is introduced, which successfully combines the advantages of two networks to classify colorectal tissue at various scales by capturing local and global information. Further, coordinate attention model is integrated with a Cross-shaped window (CrSWin) transformer for capturing tiny changes in colorectal tissue from multiple angles. The proposed CCDNet achieved accuracy rates of 98.61% and 98.96%, on the colorectal histological image and NCT-CRC-HE-100 K datasets correspondingly. The comparison analysis demonstrates that the suggested framework performed better than the most advanced methods already in use. In hospitals, clinicians can use the proposed CCDNet to verify the diagnosis.
Assuntos
Neoplasias Colorretais , Aprendizado Profundo , Neoplasias Colorretais/patologia , Neoplasias Colorretais/diagnóstico , Humanos , Processamento de Imagem Assistida por Computador/métodos , Redes Neurais de Computação , Interpretação de Imagem Assistida por Computador/métodos , AlgoritmosRESUMO
Background: The COVID-19 pandemic had a substantial impact on cancer services. The aim of our study was to evaluate the recovery of endoscopic activity and cancer detection after the COVID-19 pandemic. Methods: Endoscopic data from January 2019 to December 2020 were retrospectively collected to assess the endoscopic activity and cancer detection during the COVID-19 peak period (February 2020) and the post-COVID-19 peak period (March to July 2020). Results: The COVID-19 pandemic almost brought endoscopic activity and cancer detection to a standstill. Diagnostic procedure and endoscopic resection showed the greatest reduction. With the decline in COVID-19 infections, endoscopic activity gradually returned to previous level in July. However, the detection rate of gastric cancer resumed in September, whereas colorectal cancer resumed in August. The monthly detection rates of gastric and colorectal cancers decreased from their initial peaks of 2.98 % and 6.45 %, respectively, and finally were even lower than the average in 2019. Similarly, the mean age of patients who received endoscopy also declined as the detection rates resumed. The increasing colonoscopies allowed the missing colorectal cancer patients to be caught up. In contrast, it was expected that 6.69 % of gastric cancer patients were missed and did not receive needed endoscopy. Conclusions: The recovery of cancer detection occurred later than that of endoscopic activity, especially for gastric cancer. Older people were vulnerable to the continuous impact of COVID-19 pandemic than young people for seeking medical services. Urgent efforts are required to recover and maintain cancer services before subsequent waves of the COVID-19 pandemic.
RESUMO
Guided surgery has demonstrated significant improvements in patient outcomes in some disease processes. Interest in this field has led to substantial growth in the technologies under investigation. Most likely no single technology will prove to be "best," and combinations of macro- and microscale guidance-using radiological imaging navigation, probes (activatable, perfusion, and molecular-targeted; large- and small-molecule), autofluorescence, tissue intrinsic optical properties, bioimpedance, and other characteristics-will offer patients and surgeons the greatest opportunity for high-success/low-morbidity medical interventions. Problems are arising, however, from the lack of valid testing formats; surgical training simulators suffer the same problems. Small animal models do not accurately recreate human anatomy, especially in terms of tissue volume. Large animal models are expensive and have difficulty replicating many pathological states, particularly when molecular specificity for individual cancers is required. Furthermore, the sheer number of technologies and the potential for synergistic combination leads to exponential growth of testing requirements that is unrealistic for in vivo testing. Therefore, critical need exists to expand the ex vivo/in vitro testing platforms available to investigators and, once validated, a need to increase the acceptance of these methods for funding and regulatory endpoints. Herein is a review of the available ex vivo/in vitro testing formats for guided surgery, a review of their advantages/disadvantages, and consideration for how our field may safely and more swiftly move forward through stronger adoption of these testing and validation methods.
RESUMO
Background and Aims: Survival rates for esophageal squamous cell carcinoma (ESCC) are extremely low due to the late diagnosis of most cases. An understanding of the early molecular processes that lead to ESCC may facilitate opportunities for early diagnosis; however, these remain poorly defined. Tylosis with esophageal cancer (TOC) is a rare syndrome associated with a high lifetime risk of ESCC and germline mutations in RHBDF2, encoding iRhom2. Using TOC as a model of ESCC predisposition, this study aimed to identify early-stage transcriptional changes in ESCC development. Methods: Esophageal biopsies were obtained from control and TOC individuals, the latter undergoing surveillance endoscopy, and adjacent diagnostic biopsies were graded as having no dysplasia or malignancy. Bulk RNA-Seq was performed, and findings were compared with sporadic ESCC vs normal RNA-Seq datasets. Results: Multiple transcriptional changes were identified in TOC samples, relative to controls, and many were detected in ESCC. Accordingly, pathway analyses predicted an enrichment of cancer-associated processes linked to cellular proliferation and metastasis, and several transcription factors were predicted to be associated with TOC and ESCC, including negative enrichment of GRHL2. Subsequently, a filtering strategy revealed 22 genes that were significantly dysregulated in both TOC and ESCC. Moreover, Keratin 17, which was upregulated in TOC and ESCC, was also found to be overexpressed at the protein level in 'normal' TOC esophagus tissue. Conclusion: Transcriptional changes occur in TOC esophagus prior to the onset of dysplasia, many of which are associated with ESCC. These findings support the utility of TOC to help reveal the early molecular processes that lead to sporadic ESCC.
RESUMO
The article presents, for the first time, a terahertz metamaterial absorber (TMA) designed in the shape of a cross consisting of four orthogonally positioned horn-shaped patches in succession, to detect brain cancer cells. The design exhibits the property of mu-negative material, indicating magnetic resonance. The proposed TMA has achieved an impressive absorption rate of 99.43% at 2.334 THz and a high Q-factor of 47.15. The sensing capability has been investigated by altering the refractive index of the surrounding medium in the range of 1.3 to 1.48, resulting in a sensitivity of 0.502 THz/RIU. The proposed TMA exhibits complete polarization insensitivity, highlighting this as one of its advantageous features. The adequate sensing capability of the proposed TMA in differentiating normal and cancerous brain cells makes it a viable candidate for an early and efficient brain cancer detector. This research can be the foundation for future research on using THz radiation for brain cancer detection.
RESUMO
BACKGROUND AND OBJECTIVE: Papanicolaou staining has been successfully used to assist early detection of cervix cancer for several decades. We postulate that this staining technique can also be used for assisting early detection of oral cancer, which is responsible for about 300,000 deaths every year. The rational for such claim includes two key observations: (i) nuclear atypia, i.e., changes in volume, shape, and staining properties of the cell nuclei can be linked to rapid cell proliferation and genetic instability; and (ii) Papanicolaou staining allows one to reliably segment cells' nuclei and cytoplasms. While Papanicolaou staining is an attractive tool due to its low cost, its interpretation requires a trained pathologist. Our goal is to automate the segmentation and classification of morphological features needed to evaluate the use of Papanicolaou staining for early detection of mouth cancer. METHODS: We built a convolutional neural network (CNN) for automatic segmentation and classification of cells in Papanicolaou-stained images. Our CNN was trained and evaluated on a new image dataset of cells from oral mucosa consisting of 1,563 Full HD images from 52 patients, annotated by specialists. The effectiveness of our model was evaluated against a group of experts. Its robustness was also demonstrated on five public datasets of cervical images captured with different microscopes and cameras, and having different resolutions, colors, background intensities, and noise levels. RESULTS: Our CNN model achieved expert-level performance in a comparison with a group of three human experts on a set of 400 Papanicolaou-stained images of the oral mucosa from 20 patients. The results of this experiment exhibited high Interclass Correlation Coefficient (ICC) values. Despite being trained on images from the oral mucosa, it produced high-quality segmentation and plausible classification for five public datasets of cervical cells. Our Papanicolaou-stained image dataset is the most diverse publicly available image dataset for the oral mucosa in terms of number of patients. CONCLUSION: Our solution provides the means for exploring the potential of Papanicolaou-staining as a powerful and inexpensive tool for early detection of oral cancer. We are currently using our system to detect suspicious cells and cell clusters in oral mucosa slide images. Our trained model, code, and dataset are available and can help practitioners and stimulate research in early oral cancer detection.
RESUMO
Lung and colon cancers are leading contributors to cancer-related fatalities globally, distinguished by unique histopathological traits discernible through medical imaging. Effective classification of these cancers is critical for accurate diagnosis and treatment. This study addresses critical challenges in the diagnostic imaging of lung and colon cancers, which are among the leading causes of cancer-related deaths worldwide. Recognizing the limitations of existing diagnostic methods, which often suffer from overfitting and poor generalizability, our research introduces a novel deep learning framework that synergistically combines the Xception and MobileNet architectures. This innovative ensemble model aims to enhance feature extraction, improve model robustness, and reduce overfitting.Our methodology involves training the hybrid model on a comprehensive dataset of histopathological images, followed by validation against a balanced test set. The results demonstrate an impressive classification accuracy of 99.44%, with perfect precision and recall in identifying certain cancerous and non-cancerous tissues, marking a significant improvement over traditional approach.The practical implications of these findings are profound. By integrating Gradient-weighted Class Activation Mapping (Grad-CAM), the model offers enhanced interpretability, allowing clinicians to visualize the diagnostic reasoning process. This transparency is vital for clinical acceptance and enables more personalized, accurate treatment planning. Our study not only pushes the boundaries of medical imaging technology but also sets the stage for future research aimed at expanding these techniques to other types of cancer diagnostics.
Assuntos
Neoplasias do Colo , Aprendizado Profundo , Neoplasias Pulmonares , Humanos , Neoplasias Pulmonares/diagnóstico por imagem , Neoplasias Pulmonares/classificação , Neoplasias do Colo/diagnóstico por imagem , Neoplasias do Colo/classificação , Inteligência ArtificialRESUMO
Introduction: Our research addresses the critical need for accurate segmentation in medical healthcare applications, particularly in lung nodule detection using Computed Tomography (CT). Our investigation focuses on determining the particle composition of lung nodules, a vital aspect of diagnosis and treatment planning. Methods: Our model was trained and evaluated using several deep learning classifiers on the LUNA-16 dataset, achieving superior performance in terms of the Probabilistic Rand Index (PRI), Variation of Information (VOI), Region of Interest (ROI), Dice Coecient, and Global Consistency Error (GCE). Results: The evaluation demonstrated a high accuracy of 91.76% for parameter estimation, confirming the effectiveness of the proposed approach. Discussion: Our investigation focuses on determining the particle composition of lung nodules, a vital aspect of diagnosis and treatment planning. We proposed a novel segmentation model to identify lung disease from CT scans to achieve this. We proposed a learning architecture that combines U-Net with a Two-parameter logistic distribution for accurate image segmentation; this hybrid model is called U-Net++, leveraging Contrast Limited Adaptive Histogram Equalization (CLAHE) on a 5,000 set of CT scan images.
RESUMO
Epithelial ovarian cancer (EOC) is the deadliest women's cancer and has a poor prognosis. Early detection is the key for improving survival (a 5-year survival rate in stage I/II is over 70% compared to that of 25% in stage III/IV) and can be achieved through methylation markers from circulating cell-free DNA (cfDNA) using a liquid biopsy. In this study, we first identify top 500 EOC markers differentiating EOC from healthy female controls from 3.3 million methylome-wide CpG sites and validated them in 1,800 independent cfDNA samples. We then utilize a pretrained AI transformer system called MethylBERT to develop an EOC diagnostic model which achieves 80% sensitivity and 95% specificity in early-stage EOC diagnosis. We next develop a simple digital droplet PCR (ddPCR) assay which archives good performance, facilitating early EOC detection.
RESUMO
Imaging flow cytometry, which combines the advantages of flow cytometry and microscopy, has emerged as a powerful tool for cell analysis in various biomedical fields such as cancer detection. In this study, we develop multiplex imaging flow cytometry (mIFC) by employing a spatial wavelength division multiplexing technique. Our mIFC can simultaneously obtain brightfield and multi-color fluorescence images of individual cells in flow, which are excited by a metal halide lamp and measured by a single detector. Statistical analysis results of multiplex imaging experiments with resolution test lens, magnification test lens, and fluorescent microspheres validate the operation of the mIFC with good imaging channel consistency and micron-scale differentiation capabilities. A deep learning method is designed for multiplex image processing that consists of three deep learning networks (U-net, very deep super resolution, and visual geometry group 19). It is demonstrated that the cluster of differentiation 24 (CD24) imaging channel is more sensitive than the brightfield, nucleus, or cancer antigen 125 (CA125) imaging channel in classifying the three types of ovarian cell lines (IOSE80 normal cell, A2780, and OVCAR3 cancer cells). An average accuracy rate of 97.1% is achieved for the classification of these three types of cells by deep learning analysis when all four imaging channels are considered. Our single-detector mIFC is promising for the development of future imaging flow cytometers and for the automatic single-cell analysis with deep learning in various biomedical fields.
RESUMO
BACKGROUND: Early screening using low-dose computed tomography (LDCT) can reduce mortality caused by non-small-cell lung cancer. However, â¼25% of the 'suspicious' pulmonary nodules identified by LDCT are later confirmed benign through resection surgery, adding to patients' discomfort and the burden on the healthcare system. In this study, we aim to develop a noninvasive liquid biopsy assay for distinguishing pulmonary malignancy from benign yet 'suspicious' lung nodules using cell-free DNA (cfDNA) fragmentomics profiling. METHODS: An independent training cohort consisting of 193 patients with malignant nodules and 44 patients with benign nodules was used to construct a machine learning model. Base models using four different fragmentomics profiles were optimized using an automated machine learning approach before being stacked into the final predictive model. An independent validation cohort, including 96 malignant nodules and 22 benign nodules, and an external test cohort, including 58 malignant nodules and 41 benign nodules, were used to assess the performance of the stacked ensemble model. RESULTS: Our machine learning models demonstrated excellent performance in detecting patients with malignant nodules. The area under the curves reached 0.857 and 0.860 in the independent validation cohort and the external test cohort, respectively. The validation cohort achieved an excellent specificity (68.2%) at the targeted 90% sensitivity (89.6%). An equivalently good performance was observed while applying the cut-off to the external cohort, which reached a specificity of 63.4% at 89.7% sensitivity. A subgroup analysis for the independent validation cohort showed that the sensitivities for detecting various subgroups of nodule size (<1 cm: 91.7%; 1-3 cm: 88.1%; >3 cm: 100%; unknown: 100%) and smoking history (yes: 88.2%; no: 89.9%) all remained high among the lung cancer group. CONCLUSIONS: Our cfDNA fragmentomics assay can provide a noninvasive approach to distinguishing malignant nodules from radiographically suspicious but pathologically benign ones, amending LDCT false positives.
RESUMO
BACKGROUND: Despite its excellent screening effectiveness and sensitivity for breast cancer (BC), digital breast tomosynthesis (DBT) is controversial due to its high radiation exposure and long reading time. This study examines the diagnostic accuracy of DBT and digital mammography (DM) for BC screening and diagnosis in women with dense or non-dense breast tissue. MATERIAL AND METHODS: PRISMA-compliant searches were performed on Medline, Embase, PubMed, Web of Science, and the Cochrane databases for articles comparing DBT and DM for BC screening until March 2023. Meta-analysis was performed using RevMan sofware, and the Cochrane Risk of Bias Assessment Tool was employed to assess study quality. RESULTS: This meta-analysis included 11 trials with a total of 2,124,018 individuals. Screening with DBT resulted in a greater cancer detection rate, as demonstrated by a risk ratio (RR) of 1.27 (95% confidence interval (95% CI): 1.14-1.41). Digital breast tomosynthesis also had a reduced recall rate, with a RR of 0.88 (95% CI: 0.78-0.99), higher sensitivity and specificity values (pooled sensitivity of 0.91 (95% CI: 0.59-0.99)) and pooled specificity of 0.90 (95% CI: 0.42-1.0)) than DM (pooled sensitivity of 0.86 (95% CI: 0.52-1.0) and pooled specificity of 0.81 (95% CI: 0.12-1.0)). All acquired data exhibited reliability, lack of bias and statistical significance (p < 0.05). CONCLUSION: Digital breast tomosynthesis is a more effective screening and diagnostic assessment tool for women with dense or non-dense breasts than DM in terms of incremental cancer detection, sensitivity and recall rate.
RESUMO
Reports of mammary Paget's disease (MPD) as a manifestation of breast cancer recurrence are rare. MPD presents a particular challenge when emerging more than two decades after a breast cancer treated with evidence-based therapy. There is a broad spectrum of non-malignant causes for dermatitis of the nipple during the initial presentation that may delay cancer work-up. This case highlights the MPD work-up and management in the context of a personal history of breast cancer. This unique clinical presentation emphasizes the importance of vigilant cancer surveillance for timely intervention, especially for a presumed cured cancer.
RESUMO
In breast cancer (BC) pathogenesis models, normal cells acquire somatic mutations and there is a stepwise progression from high-risk lesions and ductal carcinoma in situ to invasive cancer. The precancer biology of mammary tissue warrants better characterization to understand how different BC subtypes emerge. Primary methods for BC prevention or risk reduction include lifestyle changes, surgery, and chemoprevention. Surgical intervention for BC prevention involves risk-reducing prophylactic mastectomy, typically performed either synchronously with the treatment of a primary tumor or as a bilateral procedure in high-risk women. Chemoprevention with endocrine therapy carries adherence-limiting toxicity.
Assuntos
Neoplasias da Mama , Carcinoma Intraductal não Infiltrante , Humanos , Feminino , Neoplasias da Mama/terapia , Neoplasias da Mama/patologia , Neoplasias da Mama/genética , Carcinoma Intraductal não Infiltrante/terapia , Carcinoma Intraductal não Infiltrante/patologiaRESUMO
Background: Our previous studies have demonstrated that Raman spectroscopy could be used for skin cancer detection with good sensitivity and specificity. The objective of this study is to determine if skin cancer detection can be further improved by combining deep neural networks and Raman spectroscopy. Patients and methods: Raman spectra of 731 skin lesions were included in this study, containing 340 cancerous and precancerous lesions (melanoma, basal cell carcinoma, squamous cell carcinoma and actinic keratosis) and 391 benign lesions (melanocytic nevus and seborrheic keratosis). One-dimensional convolutional neural networks (1D-CNN) were developed for Raman spectral classification. The stratified samples were divided randomly into training (70%), validation (10%) and test set (20%), and were repeated 56 times using parallel computing. Different data augmentation strategies were implemented for the training dataset, including added random noise, spectral shift, spectral combination and artificially synthesized Raman spectra using one-dimensional generative adversarial networks (1D-GAN). The area under the receiver operating characteristic curve (ROC AUC) was used as a measure of the diagnostic performance. Conventional machine learning approaches, including partial least squares for discriminant analysis (PLS-DA), principal component and linear discriminant analysis (PC-LDA), support vector machine (SVM), and logistic regression (LR) were evaluated for comparison with the same data splitting scheme as the 1D-CNN. Results: The ROC AUC of the test dataset based on the original training spectra were 0.886±0.022 (1D-CNN), 0.870±0.028 (PLS-DA), 0.875±0.033 (PC-LDA), 0.864±0.027 (SVM), and 0.525±0.045 (LR), which were improved to 0.909±0.021 (1D-CNN), 0.899±0.022 (PLS-DA), 0.895±0.022 (PC-LDA), 0.901±0.020 (SVM), and 0.897±0.021 (LR) respectively after augmentation of the training dataset (p<0.0001, Wilcoxon test). Paired analyses of 1D-CNN with conventional machine learning approaches showed that 1D-CNN had a 1-3% improvement (p<0.001, Wilcoxon test). Conclusions: Data augmentation not only improved the performance of both deep neural networks and conventional machine learning techniques by 2-4%, but also improved the performance of the models on spectra with higher noise or spectral shifting. Convolutional neural networks slightly outperformed conventional machine learning approaches for skin cancer detection by Raman spectroscopy.
RESUMO
Invasive ductal carcinoma (IDC) in breast specimens has been detected in the quadrant breast area: (I) upper outer, (II) upper inner, (III) lower inner, and (IV) lower outer areas by electrical impedance tomography implemented with Gaussian relaxation-time distribution (EIT-GRTD). The EIT-GRTD consists of two steps which are (1) the optimum frequencyfoptselection and (2) the time constant enhancement of breast imaging reconstruction.foptis characterized by a peak in the majority measurement pair of the relaxation-time distribution functionγ,which indicates the presence of IDC.γrepresents the inverse of conductivity and indicates the response of breast tissues to electrical currents across varying frequencies based on the Voigt circuit model. The EIT-GRTD is quantitatively evaluated by multi-physics simulations using a hemisphere container of mimic breast, consisting of IDC and adipose tissues as normal breast tissue under one condition with known IDC in quadrant breast area II. The simulation results show that EIT-GRTD is able to detect the IDC in four layers atfopt= 30, 170 Hz. EIT-GRTD is applied in the real breast by employed six mastectomy specimens from IDC patients. The placement of the mastectomy specimens in a hemisphere container is an important factor in the success of quadrant breast area reconstruction. In order to perform the evaluation, EIT-GRTD reconstruction images are compared to the CT scan images. The experimental results demonstrate that EIS-GRTD exhibits proficiency in the detection of the IDC in quadrant breast areas while compared qualitatively to CT scan images.
Assuntos
Neoplasias da Mama , Carcinoma Ductal de Mama , Impedância Elétrica , Tomografia , Humanos , Feminino , Neoplasias da Mama/diagnóstico por imagem , Tomografia/métodos , Carcinoma Ductal de Mama/diagnóstico por imagem , Distribuição Normal , Mama/diagnóstico por imagem , Simulação por Computador , Algoritmos , Processamento de Imagem Assistida por Computador/métodosRESUMO
In accordance with the World Health Organization data, cancer remains at the forefront of fatal diseases. An upward trend in cancer incidence and mortality has been observed globally, emphasizing that efforts in developing detection and treatment methods should continue. The diagnostic path typically begins with learning the medical history of a patient; this is followed by basic blood tests and imaging tests to indicate where cancer may be located to schedule a needle biopsy. Prompt initiation of diagnosis is crucial since delayed cancer detection entails higher costs of treatment and hospitalization. Thus, there is a need for novel cancer detection methods such as liquid biopsy, elastography, synthetic biosensors, fluorescence imaging, and reflectance confocal microscopy. Conventional therapeutic methods, although still common in clinical practice, pose many limitations and are unsatisfactory. Nowadays, there is a dynamic advancement of clinical research and the development of more precise and effective methods such as oncolytic virotherapy, exosome-based therapy, nanotechnology, dendritic cells, chimeric antigen receptors, immune checkpoint inhibitors, natural product-based therapy, tumor-treating fields, and photodynamic therapy. The present paper compares available data on conventional and modern methods of cancer detection and therapy to facilitate an understanding of this rapidly advancing field and its future directions. As evidenced, modern methods are not without drawbacks; there is still a need to develop new detection strategies and therapeutic approaches to improve sensitivity, specificity, safety, and efficacy. Nevertheless, an appropriate route has been taken, as confirmed by the approval of some modern methods by the Food and Drug Administration.
RESUMO
Artificial Intelligence (AI)-based image analysis has immense potential to support diagnostic histopathology, including cancer diagnostics. However, developing supervised AI methods requires large-scale annotated datasets. A potentially powerful solution is to augment training data with synthetic data. Latent diffusion models, which can generate high-quality, diverse synthetic images, are promising. However, the most common implementations rely on detailed textual descriptions, which are not generally available in this domain. This work proposes a method that constructs structured textual prompts from automatically extracted image features. We experiment with the PCam dataset, composed of tissue patches only loosely annotated as healthy or cancerous. We show that including image-derived features in the prompt, as opposed to only healthy and cancerous labels, improves the Fréchet Inception Distance (FID) by 88.6. We also show that pathologists find it challenging to detect synthetic images, with a median sensitivity/specificity of 0.55/0.55. Finally, we show that synthetic data effectively train AI models.
RESUMO
OBJECTIVES: The randomized TOmosynthesis plus SYnthesized MAmmography (TOSYMA) screening trial has shown that digital breast tomosynthesis plus synthesized mammography (DBT + SM) is superior to digital mammography (DM) in invasive breast cancer detection varying with breast density. On the other hand, the overall average glandular dose (AGD) of DBT is higher than that of DM. Comparing the DBT + SM and DM trial arm, we analyzed here the mean AGD and their determinants per breast density category and related them to the respective invasive cancer detection rates (iCDR). METHODS: TOSYMA screened 99,689 women aged 50 to 69 years. Compression force, resulting breast thickness, the calculated AGD obtained from each mammography device, and previously published iCDR were used for comparisons across breast density categories in the two trial arms. RESULTS: There were 196,622 exposures of 49,227 women (DBT + SM) and 197,037 exposures of 49,132 women (DM) available for analyses. Mean breast thicknesses declined from breast density category A (fatty) to D (extremely dense) in both trial arms. However, while the mean AGD in the DBT + SM arm declined concomitantly from category A (2.41 mGy) to D (1.89 mGy), it remained almost unchanged in the DM arm (1.46 and 1.51 mGy, respectively). In relative terms, the AGD elevation in the DBT + SM arm (64.4% (A), by 44.5% (B), 27.8% (C), and 26.0% (D)) was lowest in dense breasts where, however, the highest iCDR were observed. CONCLUSION: Women with dense breasts may specifically benefit from DBT + SM screening as high cancer detection is achieved with only moderate AGD elevations. CLINICAL RELEVANCE STATEMENT: TOSYMA suggests a favorable constellation for screening with digital breast tomosynthesis plus synthesized mammography (DBT + SM) in dense breasts when weighing average glandular dose elevation against raised invasive breast cancer detection rates. There is potential for density-, i.e., risk-adapted population-wide breast cancer screening with DBT + SM. KEY POINTS: Breast thickness declines with visually increasing density in digital mammography (DM) and digital breast tomosynthesis (DBT). Average glandular doses of DBT decrease with increasing density; digital mammography shows lower and more constant values. With the smallest average glandular dose difference in dense breasts, DBT plus SM had the highest difference in invasive breast cancer detection rates.