Artificial Intelligent Models for Breast Cancer Early

Detection

Abstract—The increasing prevalence of cancer, particularly (AI) to help process data more quickly. AI mimics human
breast cancer, has prompted advancements in medical manage- perception, reducing the need for human effort in data analysis.
ment and diagnostic techniques. Machine Learning (ML) has As medical science advances with new discoveries in areas
emerged as a promising tool for enhancing cancer detection ac-
curacy. Studies show that, without ML intervention, malignancies like anatomy, pharmacology, and detection techniques, AI can
are often only identified post-surgery, while many unnecessary enhance these developments by efficiently analyzing informa-
surgeries for benign lesions could be avoided. This paper proposes tion. It overcomes human limitations, such as memory issues,
a comprehensive early detection model for breast cancer that fatigue, and errors, making it a valuable tool in improving
integrates findings from prior research, leveraging multiple detection methods. To accurately detect breast cancer and
detection methods. The model will utilize techniques that have
demonstrated efficacy, including DWT-based multi-resolution minimize the need for invasive procedures, we need to use
Markov Random Field (MMRF) segmentation for mammograms, artificial intelligence to develop a system that can effectively
Multi-Layer Perceptron (MLP) for histological analysis, and a identify breast cancer across various detection methods. In this
k-NN with SVM-RFE approach for gene identification. Each paper, we propose an early detection model for breast cancer
method will be trained on a dataset of at least 450 confirmed that integrates findings from previous research to work with
breast cancer cases from the Wisconsin Breast Cancer Diagnosis
(WBCD) database. The research will be conducted from October multiple detection techniques
2018 to March 2019, with the aim of furthering the application
of ML in early breast cancer detection and improving diagnostic
outcomes.
II. L ITERATURE R EVIEW
Index Terms—Breast cancer;Machine learning;Discrete
wavelet transforms;Surgery;Lesions;artificial intelligent;breast A. Methods for Diagnosing Breast Cancer
cancer;machine learning;wisconsin breast cancer diagnosis
There are various methods for screening breast cancer,
including mammography, ultrasound, CT scans, MRIs, PET
I. I NTRODUCTION
scans, microwave imaging, RF sensors, biosensors, biopsies,
Cancer is a type of tumor that can grow without limits and and DNA/gene testing. Mammography is crucial for detecting
spread to other parts of the body. It is the second leading cause breast cancer early and has been shown to reduce mortality
of death around the world. While breast cancer has a lower rates, but it can sometimes produce unclear images, so it is
death rate and a relatively high five-year survival rate com- often used alongside other imaging methods. On the other
pared to some other cancers, it still has the fifth highest disease hand, DNA and gene testing requires a thorough and sys-
burden in the world. Like many other cancers, breast cancer tematic approach to collect data, as cancer is linked to many
is treated with chemotherapy, radiotherapy, hormonal therapy, different genes, depending on the type.
Identify applicable funding agency here. If none, delete this.
and sometimes surgery, which can be expensive and painful, B. Machine Learning
making daily life difficult for patients. Therefore, preventing
late-stage breast cancer diagnosis is crucial to reduce suffering. Machine Learning (ML) is a type of computer programming
Early detection is key and can be achieved through recognizing that enables computers to learn from data. Learning, in this
early signs or through screening methods. Screening includes context, means turning experiences into knowledge. In ML,
various techniques like imaging tests, tissue examinations, and the input comes in the form of training data, which represents
genetic analysis. These methods produce a lot of data that need these experiences, and the program uses this data to develop
careful analysis to determine if someone is healthy, at high expertise with the help of additional algorithms. While ML
risk for breast cancer, or already has it. However, manual data can be seen as a branch of artificial intelligence (AI) that
processing can lead to errors or limitations, making it harder to focuses on enhancing human intelligence rather than imitating
accurately identify the issue. For example, mammograms can it, it differs from other AI methods. ML can help with routine
struggle to tell the difference between benign (non-cancerous) tasks like speech recognition and image processing, as well as
and malignant (cancerous) lesions. This uncertainty often leads complex analyses such as weather forecasting, astronomical
to more invasive procedures, like needle biopsies, to determine studies, and genomic research. Its adaptability allows it to
if a lesion is potentially dangerous, which may require surgery maintain accuracy even when faced with varying data, such
to prevent the spread of cancer. In recent years, the field as in spam detection, speech recognition, and reading different
of information systems has developed artificial intelligence handwriting styles.
 C. DWT and MRF in Image Processing
Zheng and Chan (2001) used Discrete Wavelet Transform
(DWT) to create four sub-band images (HH, HL, LH, and LL)
by extending it into two dimensions along the x and y axes.
Each sub-band contains different information and is processed
with various algorithms. Meanwhile, Markov Random Field
III. M ETHODS (MRF) helps reduce noise in the images and recover the
original image for better viewing. By combining DWT and
Proposed Methods: For this research, we plan to use three MRF into the DWT-based multi- resolution MRF (MMRF)
methods in parallel: DWT-based multi-resolution MRF seg- segmentation algorithm, they were able to produce a clearer
mentation for mammography, MLP for histology analysis, image that merges the original with a refined version, enhanc-
and k-NN combined with SVMRFE for gene identification. ing the clarity and supporting better diagnostic decisions.
We recommend using data from the WBDC and at least 450
breast cancer positive samples for each method. The research D. Application of ML in Gene Expression Data
is scheduled to take place from October 2018 to March 2019. With advances in DNA and gene data, it’s now possible to
We will analyze each hypothesis using Sequence Equation identify specific genes that show abnormal mutations linked
Modeling (SEM) with LISREL to assess its significance. to cancer, such as P53, BRCA1, and BRCA2. Sharma and
Additionally, the research will qualitatively explore legal and Paliwal (2012) proposed a Bayesian classification approach
social issues related to breast cancer. to select and classify crucial cancer-related genes. They sug-
gested organizing these genes into smaller subsets, which can
A. Role of Machine Learning for Cancer Diagnosis be updated by comparing their relationships with one another.
Eventually, these related gene subsets are combined into a
To reach a complete diagnosis of cancer, accuracy and single informative subset that accurately identifies cancer. In a
speed is fundamental. Beth Israel Deaconess Medical Center study by Bhola and Tiwari (2015), they used a model featuring
(BIDMC) and Harvard Medical School research found that 12,625 genes and achieved an accuracy of 73.33 percent with
using AI in combination with pathologist work may serve 99.5 just 24 samples. By employing the k-Nearest Neighbor (k-NN)
percent accurate diagnosis compared to using AI alone (92and as a classifier and Support Vector Machine Recursive Feature
pathologist work alone (96 percent). This combination will Elimination (SVMRFE) for feature selection, they reached
give advantage as artificial intelligence usage reduces health a maximum accuracy of 97.5 percent, outperforming other
costs and human resources need, Machine Learning (ML) is classifiers and methods.
an AI technology that has evolved over the past few decades,
Fig. 1. Fig. 1. Bayesian Method Classification of WBCD based DISCUSSION
on Aloraini (2012) allowing for in-depth analysis of data, Reviewed literature suggests that combining various proven
whether it’s text or images, by identifying patterns from past algorithms and methods can enhance accuracy in specific
inputs. Initially, ML techniques for diagnosing breast cancer fields. For instance, MLP is effective for histologic exami-
were tested on open source databases, but the Wisconsin nation, DWT-based multi-resolution MRF (MMRF) segmen-
Breast Cancer Diagnosis (WBCD) dataset has become a key tation works well for mammography, and the k-NN – SVM-
benchmark. The Bayesian Network is a graphical tool that RFE method is suitable for gene identification. When paired
combines probability and graph theory to represent uncertain with skilled human resources, these techniques can generate
information. It uses conditional independence and probabilities computerized data and information that support breast cancer
to describe how different variables relate to each other. In diagnosis more efficientl
her research, Aloraini (2012) applied the Bayesian Network E. Hypothesis
to further analyze the WBCD.
H1: The use of Multilayer Perceptron (MLP) in histopatho-
logical examination has a positive impact on the accuracy
B. ML Algorithms and MLP Performance
of breast cancer detection results (BCDR). H2: The appli-
In Agarap’s 2018 study on ML algorithms using the WBCD cation of Discrete Wavelet Transform (DWT)- based multi-
dataset, the Multilayer Perceptron (MLP) outperformed six resolution Markov Random Field (MMRF) segmentation in
other algorithms (GRU-SVM, linear regression, L1-NN, L2- mammography has a positive impact on the accuracy of breast
NN, Softmax Regression, and SVM) with an accuracy of 99.03 cancer detection results (BCDR). H3: The utilization of k-
percent. It handled the largest dataset with 512,896 points, Nearest Neighbors (k-NN) - Support Vector Machine Recur-
achieving the lowest false positive rate at 1.27 percent, and sive Feature Elimination (SVMRFE) for gene identification
high true positive and true negative rates of 99.21 percent and has a positive impact on the accuracy of breast cancer detection
98.73 percent, respectively, in distinguishing between benign results (BCDR). H4: The combination of MLP histopatholog-
and malignant tumors. The MLP, developed by Rosenblatt ical examination, DWT-MMRF segmentation mammography,
in 1958, consists of hidden layers of perceptrons that use and k-NN - SVMRFE gene identification has a positive impact
activation functions for function approximation. on the accuracy of breast cancer detection results (BCDR)
F. Legal and Social Issues
AI can boost patient confidence by reducing knowledge
gaps, but it may also be unsettling as tasks once managed by
humans are now handled by machines, potentially neglecting
the need for human interaction. Additionally, in case of errors,
machines might be perceived as a mask, obscuring account-
ability. New technologies like self-driving cars or autopilot
planes, while designed to assist, often create anxiety, especially
regarding accountability in accidents—should blame fall on
the technology provider, the company, or the human opera-
tor? Similarly, in healthcare, doctors must remain the final
decision-makers, interpreting AI results and managing patient
care. Unlike machines, doctors consider the emotional impact
of information, especially in sensitive bioethical cases. There-
fore, even with AI, the role of the doctor as a decisionmaker
and caregiver is indispensable.
CONCLUSION
MLP histopathologic examination, DWT-MMRF segmenta-
tion mammography, and k-NN – SVMRFE gene identification
are all expected to contribute positively to breast cancer
detection. The integration of these methods is likely to enhance
the accuracy of early detection. Our research outcomes may
provide a basis for further studies on the use of machine
learning in the early detection of breast cancer.
IV. R EFERENCES
REFERENCES [1] Merriam-Webster Dictionary, Cancer,
Springfield: Merriam-Webster Dictionary, 2018, forthcoming.
[2] Roser M, Ritchie H, ”Cancer,” Oxford: Our World in
Data, 2018, forthcoming. [3] Breast Cancer Organization,
“Breast Cancer Treatment Causes Severe Side Effects for
Many Women,” Ardmore: Breast Cancer Organization, Febru-
ary 2017, forthcoming. [4] Anderson BO et al., “Guideline
Implementation for Breast Healthcare in Low-Income and
Middle-Income Countries: Overview of the Breast Health
Global Initiative Global Summit 2007,” Cancer, vol. 113, pp.
2221–2243, 2008. [5] Wang L, “Early Diagnosis of Breast
Cancer,” J. Sensors, vol. 7, pp. 1572, July 2017. [6] Bahl
M, Barzila R, Yedidia AB, Locascio NJ, Yu L, Lehman CD,
”High-Risk Breast Lesions: A Machine Learning Model to
Predict Pathologic Upgrade and Reduce Unnecessary Surgi-
cal Excision,” J. Rad. RSNA, vol. 000, pp. 1-9, 2017. [7]
Al-Shamasneh ARM, Obaidellah UH, ”Artificial Intelligence
Techniques for Cancer Detection and Classification: A Re-
view,” European Scientific Journal, vol. 13, January 2017.
[8] Miller DD, Brown EW, ”Artificial Intelligence in Medical
Practice: The Questions to Answer?” The American Journal of
Medicine, vol. 131, pp. 129-134, 2018. [9] Shalev-Shwartz S,
Ben-David S, Understanding Machine Learning: From Theory
to Algorithms, Cambridge: Cambridge University Press, 2014.
[10] Kritz J, ”Artificial Intelligence Achieves Near-Human
Performance in Diagnosing Breast Cancer: Human and Com-
puter Analyses Together Identify Cancer with 99.
 Abstract—Cancer, particularly breast cancer, is increasingly detection of breast cancer, particularly in places where hospital
prevalent worldwide and a major cause of death. Early detection access may be limited.
significantly improves survival rates and reduces treatment costs.
However, current diagnostic methods have limitations, including
high resource demands and accessibility issues. There is a need III. METHODOLOGY
for user-friendly technologies that provide reliable results for This study uses the Wisconsin Breast Dataset to propose
early breast cancer diagnosis. This study explores the use of
Artificial Intelligence, specifically Artificial Neural Networks, classification methods for the early diagnosis of breast cancer.
Support Vector Machines, and Random Forest algorithms, to The data was classified using the Artificial Neural Network,
classify breast cancer image features using the Wisconsin Breast Support Vector Machine, and Random Forest techniques. The
Cancer dataset. The results indicate that the Artificial Neural flow chart in Figure 1 describes the steps followed for the
Network achieved a 99 percent success rate in identifying classification process. Fig. 1. Flowchart of Study A. Data
cancerous and benign images, highlighting its potential for early
detection. These findings may pave the way for further research Description There are several public datasets available that
into improving breast cancer diagnosis. contain images of breast cancer. In this study, we used
Wisconsin Breast Cancer Dataset the Wisconsin from UCI
I. I NTRODUCTION ML Repository. This dataset consists of 699 instances with
Cancer, particularly breast cancer, is a leading global health 11 attributes. Table II contains the features of the data set.
crisis, responsible for one in six deaths worldwide, with a
significant impact in less developed countries. Recent data A. Data Description
shows that breast, lung, and colorectal cancers are on the rise, Define abbreviations and acronyms the first time they are
with breast cancer accounting for 30 percent of new cases used in the text, even after they have been defined in the
in women. Early detection is crucial for improving survival abstract. Abbreviations such as IEEE, SI, MKS, CGS, ac, dc,
rates, but traditional diagnostic methods like mammography and rms do not have to be defined. Do not use abbreviations
and MRI can be costly when the disease progresses. in the title or heads unless they are unavoidable.
This study focuses on using classification algorithms to
enhance early breast cancer prediction. We analyze the Wis- B. Data Preprocessing
consin Breast Cancer Dataset, applying various classification
Datasets often have different types of issues, like missing
techniques, including Artificial Neural Networks, Support Vec-
values, duplicates, or errors. To address these problems, we
tor Machines, and Random Forest methods, to identify the
need to preprocess the data, which involves cleaning it up.
most effective approach for early disease detection.
This includes reducing missing values and noise.
II. RELATED WORK We also standardized the data, which means we converted it
into a consistent format across different datasets. Before using
Various models and strategies using different imaging tech- the data for our research, we transformed it to make it suitable.
niques are available for early breast cancer detection. Mam-
To standardize the data, we used the StandardScaler tool
mography is the most common and effective method for
from the sklearn library. By applying the fit transform()
identifying breast cancer at its earliest stages. This technique
method, we scaled the data so that very large and very small
involves compressing the breast between an imaging device
values fit within a specific range.
and a plate to capture images from multiple angles—typically
front and side views—allowing for early diagnosis of the
C. Machine Learning Algorithms
disease. If mammography results are uncertain, additional
imaging may be needed for better visualization, such as We built the model using several machine learning algo-
ultrasonography, which uses sound waves to create cross- rithms, including Artificial Neural Networks (ANN), Support
sectional images. This method is especially useful for women Vector Machines (SVM), and Random Forest.
with dense breast tissue, as it can reveal small cancerous ANNs are mathematical models made up of interconnected
areas that mammography might miss. Magnetic Resonance units called neurons. They process information by receiving
Imaging (MRI) is the most sensitive method for diagnosing signals, combining them, transforming the data, and producing
breast cancer and helps assess the extent of the disease when a numerical result. Unlike traditional processors that handle
abnormalities are found in other imaging tests. Additionally, tasks in a linear way, each neuron in an ANN focuses on a
genetic analysis can identify genes related to breast cancer, different part of a complex problem.
focusing on those that regulate breast cell development and Support Vector Machines (SVM) are techniques that classify
inhibit cancer cell growth. To effectively use the tools and data by finding the best way to separate different classes. They
facilities for breast cancer detection, healthcare providers need mainly work for binary classification, identifying which class
specific knowledge and access to equipment. However, it can is further away from the other in the training data.
be challenging to treat all breast cancer patients quickly, espe- Random Forest is a supervised learning algorithm used for
cially in healthcare settings, while also continuing research on both classification and regression tasks. It works by creating
high-risk groups. This situation highlights the need for tools multiple decision trees to enhance classification accuracy
that can assist dermatologists and patients in accessing early during calculations.
 [width=0.5]image.png
Fig. 1. Enter Caption

D. Evaluation Metrics
We used precision, recall, F-measure, and accuracy values
presented in Table III. Precision is the proportion of estimated
positive values. Recall value is a metric that shows how many
transactions need to predict as positive. The harmonic mean of
recall and accuracy is known as the Fmeasure. The most used
metric for accuracy is the proportion of properly identified
samples to the total number of samples.
E. Experimental Evaluation
This section discusses the ”Wisconsin Breast Cancer”
dataset, experimental results, and evaluation metrics. In this
research, we use Python 3.10 to implement our proposed
method. We tested our classification technique using the k -
fold cross-validation method. In addition, we have utilized the
10- fold technique: the data set is separated into ten chunks.
Nine of the dataset’s folds are utilized for training, while
the last set is used for analysis and testing. Out of the 569
examples, we used 398 for the training set and 171 for testing
F. RESULTS
This section illustrates the results of the experimental eval-
uation. Table IV shows the results of the performance criteria.

G. CONCLUSION
Breast cancer is one of the most common cancers in women,
and early detection is vital for helping doctors and improving
patient survival. This study used the Wisconsin Breast Cancer
Dataset to compare different machine learning methods for
identifying breast cancer. Overall, the results showed that each
algorithm achieved over 90 percent accuracy in distinguishing
between benign and malignant tumors. Among the three algo-
rithms tested, the Artificial Neural Network (ANN) performed
the best.
Recent studies have also used the Wisconsin dataset with
different features and algorithms. For example, one study
achieved 98.83 percent accuracy using Random Forest, while
another got 97 percent accuracy with Support Vector Machine
(SVM). Other studies reported 98 percent accuracy with ANN
and 96.5 percent with Random Forest and SVM.
When comparing our results, the performance of SVM and
Random Forest in this study was better than what previous
research reported using the same dataset. Additionally, our
results were strong when compared to related studies using
different datasets, indicating consistency and reliability in our
findings.
 A New Computerized Method for Missed Cancer
Detection in Screening Mammography

Abstract—Current computer-aided diagnosis (CAD) methods detected cancer mammograms, and normal mammograms.
for mammography face a bottleneck due to their independent ”Missed- cancer” refers to cases where cancer was either
”reader” approach, rather than complementing radiologists as a wrongly diagnosed (false-negative) or correctly diagnosed but
”second reader”. This study proposes a novel CAD method to
enhance early breast cancer detection in screening mammograms visible on an earlier mammogram. ”Detected-cancer” includes
by focusing on computerized analysis of cancers overlooked by mammograms where cancer was identified at the time of
radiologists. The approach employs multi-mode detection, breast screening.
area partitioning, and weighted classification using distinguishing
features from missed cancer analysis. Results show significant
improvement in detection performance, particularly for missed B. Characteristics of the database
cases, with earlier detection and fewer false positives.
keywords=CAD, mammography, de- The database characteristics were described by distributions
tection, breast density, missed cancer, of: (a) exam numbers, (b) reasons for missed cancer, (c) mass
doi=10.1109/ICITECHNOLOGY.2007.4290460 shape, (d) mass margin, and (e) mass density, shown in Fig.
1. Most cases had two exams (detected and missed), and the
I. I NTRODUCTION main reasons for false negatives were interpretation errors,
Mammography is important for women over 50 as a lack of significant evidence, or no visible signs. The majority
screening tool, but concerns exist about its accuracy. False of missed masses had irregular shapes, equal density, and
positives can lead to unnecessary follow-up tests and biopsies, spiculated margins.
which are costly, physically harmful, and emotionally stressful.
False negatives, on the other hand, can delay the detection III. NEW CAD SYSTEM
and treatment of breast cancer, allowing it to progress to a
more serious stage, which affects survival rates and increases The new CAD system builds on two generations of mass
healthcare costs. Reading mammograms is challenging due to detection algorithms for digitized mammograms and includes
subtle signs in complex images and a low number of abnormal insights from missed cancer analysis. The strategies used in
cases, leading to missed cancers. False-negative rates range this study are:
from 4 percent to 34 percent. For instance, about 9 percent of
cancers found during screening were visible in mammograms A. Multi-mode detection by breast density classification
taken two years earlier but were missed, classifying them as
screening errors. To improve mammography accuracy without Baseline testing showed that lesions in dense breasts are
the challenges of double reading, computer-aided diagnosis more likely to be missed by CAD. To improve detection, a
(CAD) is used as a second reader to assist radiologists, multi-mode approach was used, where mammograms were
who make the final decision. CAD methods fall into two classified by breast density before applying the appropriate
categories: one compares asymmetry between the left and detection mode. Due to the small dataset, mammograms were
right breast, with irregularities indicating suspicious areas. categorized into two groups: density ¡25 percent and ¿25
The other focuses on identifying features within a single percent, with each mode trained on its respective group.
image to differentiate cancer from normal tissue, involving
feature extraction and classification. This paper introduces a B. Breast area partition and region based adaptive detection
new CAD system aimed at improving the detection of missed
cancers in screening mammograms, building on previous CAD Cancer locations in mammograms vary greatly in their
algorithms for mass detection and research on breast density’s likelihood of being missed, so partitioning the breast area helps
impact on CAD accuracy. with further processing. The partitioning process includes four
steps: (i) identifying the breast boundary and nipple, (ii)
II. MATERIALS identifying the pectoral muscle and view (CC or MLO), and
A. Data Collection (iii) area partitioning, as shown in Fig. 3. Due to the varying
The study collected 100 cases of missed cancers to cre- density of breast tissue, the central area is defined as a triangle
ate three datasets: missed cancer mammograms, screening- covering 1/4 of the total breast area, with each side parallel to
the chest wall or pectoral muscle boundary. Fig. 4 shows the
Identify applicable funding agency here. If none, delete this. likelihood of missed cancers in each region.
 The results show that the new system significantly improves
detection performance for both detected and missed cases.
There was a particularly notable improvement in identifying
missed cancers, although overall detection performance is still
lower compared to detected cases. Additionally, the new CAD
strategy allows for earlier detection with fewer false positive
signals.
R EFERENCES
REFERENCES Rutqvist et al. (1990) assessed data showing
that mammography screening reduces breast cancer mortality.
Dershaw (1997) discussed the importance of mammographic
screening for women aged 50 and older. Huynh et al. (1998)
examined the issue of false-negative mammograms. Van Di-
jck et al. (1993) reviewed the detectability of breast cancer
in mammography screening programs by analyzing previous
mammograms. Bird et al. (1992) analyzed cancers missed dur-
ing screening mammography. Martin et al. (1979) highlighted
instances of breast cancer missed by mammography. Baines et
al. (1990) discussed the role of reference radiologists and the
impact of interobserver agreement on cancer detection delays.
Fig. 1. Enter Caption Adler and Wahl (1995) explored new imaging methods for
the breast and their potential findings. Giger (1993) focused
on computer-aided diagnosis in breast imaging. Semmlow et
C. Weighted classification using the distinguishing features al. presented a fully automated system for screening mammo-
identified in missed cancer analysis grams.
The classification uses a modified hybrid structure that
includes: (i) a combined ”hard” and ”soft” decision-making
strategy; (ii) adjusted decision thresholds based on the analysis
of missed cancer features. For instance, since there was a
notable difference in ”mass size” between detected and missed
cases, the threshold for this feature was lowered to increase
the chances of detecting missed cancers; (iii) candidate com-
petition is weighted using region likelihood values.
IV. D ETECTION P ERFORMANCE
Two types of evaluations were conducted: standalone de-
tection sensitivity/specificity and early detection. Figures 5-6
display the FROC curves for detecting missed and detected
cancer cases. The new CAD system showed better detection
performance at both stages, with a more significant improve-
ment in missed cancer detection, as it was specifically designed
for that purpose.
Another evaluation focused on early detection using the
new CAD system, measuring how many months earlier CAD
detected cancer compared to a radiologist without CAD as-
sistance. It is assumed that all true positive detections were
accepted by the radiologist, with no negative impact from
CAD’s false positives on diagnosis decisions.
C ONCLUSION
This study explores a new CAD strategy aimed at im-
proving early breast cancer detection in screening mammo-
grams. Unlike existing CAD methods, this system acts as
a supportive ”second reader” for radiologists, focusing on
analyzing and detecting cancers that may have been missed.