Cancer in today’s world is among the various diseases leading to death worldwide, specifically breast cancer. AI ( artificial intelligence ) has proven to be a potential source of cancer detection and helps radiologists find cancerous tissues more quickly and accurately. Although we have our customary ways of detecting cancer ( CT scan, PET scan, MRI, Mammogram, etc), all these machines have possibilities of false positives. Since this project is more focused on breast cancer, the main problem of detecting the cancerous tissue in the breast is based on the density of the breast; the denser it is, the harder it is to detect it accurately. Due to some age groups ( particularly in the age range of 22-25 years old)  having denser breasts than others, it can lead to inaccurate or late detections.

Hypothesis:

 My prediction is that my AI (artificial intelligence) -powered app may scan breast cancer images accurately but using an AI (artificial intelligence) powered app to assess the breast tissue could improve the diagnosis to summarize my hypothesis, an AI( artificial intelligence) powered app will be able to detect breast cancer accurately. If my AI-powered app does accurately, then this could revolutionize the healthcare system and even the healthcare software. But the app could run into some problems, just like mammograms/MRIs ( magnetic resonance imaging), because the breast's density could affect accuracy. 

This picture illustrates the four types of breast density ( fatty, scattered fibroglandular density, heterogeneously dense, and extremely dense).

 

 

                                                                            The diagram illustrates breast cancer metastasis.

 

When a tumour is diagnosed late, the cancer is allowed to spread and enlarge. This often results in cancer metastasis ( metastasis is the movement of cancer cells to other areas of the body), and cancer metastasis is usually the main reason for death due to cancer. The most common type of cancer that is diagnosed is breast cancer. Inaccurate, late detections and false positives are all factors in not detecting the cancerous tissue in the breast. Still, there are also many other reasons, such as a radiologist being quick to overlook symptoms of breast cancer, hospitals/clinics being short-staffed, etc. All of these are problems in detecting cancerous tissues, but to solve these, I've tested AI model(s) and trained them in each category of density. Since everyone can't necessarily use these, I created an app prototype called BreastShield_AI. In this app, I integrated the most accurate AI model that I tested/trained. I put all the results/information in the charts to form my conclusion.   

 

In summation, cancer being diagnosed late or inaccurately causes several problems, and this project primarily targets whether AI ( artificial intelligence ) can be used to detect breast cancer on different types of density.

Abstract

This project investigates the possibility of AI (artificial intelligence) having a chance of being a source of breast cancer ( and cancer in general) detection. But we don’t know whether it can detect all types of cancer accurately because it has not been tested on vast amounts of cancer images. However, in my project, I will create an AI (artificial intelligence) powered app where users upload their medical images (mammograms) to the app. And the app tells them whether it is a potential risk of cancer or it is a malignant tumour. If my app accurately detects breast cancer on the test picture, then my AI-powered app has done what it intended to do: to test AI in detecting cancer in the four categories. When we are talking about AI detecting breast cancer AI has already made impressive accomplishments in detecting breast cancer but has not been put in hospitals as a first choice because the customary mammogram is their choice of cancer detection. Now having a false positive on a mammogram depends on your age because for older women chances are about 7-12% which is pretty low but for younger women particularly between 22-25 years old range false positives are at about 10-12% which is a problem for them because they have dense breasts which could mean for them to have supplementary imaging done like an MRI (magnetic resonance imaging) which are at even higher rate of false positives at about  83%. About 50% of types of cancer are detected at advanced stages which lowers the survival rate but if we detect cancer early we improve survival overall. If an AI powdered app is created, we can move on by using AI model(s) for clinical use, helping aid health practitioners, and even using them in general for cancer detection. This would especially help oncologists with early/accurate cancer detection and overall improve the survival rate of breast cancer patients. 

 

 

                                                                                          Diagram of Breast Cancer

Methodology/Materials:

To be able to test AI's accuracy in detecting breast cancer I had to use a lot of resources and data. First I had to begin with pictures of mammogram images in the four categories of density. After I picked them I gathered a lot of pre-trained AI models and tested them. I tested each pre-trained AI models ( Clarifai. IBM Watson, and Google console) with the picture and tested them to figure out the most accurate of them all. After I found out which was the most accurate I integrated it into my app ( the app was built on MIT App Inventor). After all that was done all I had to do was calculate the accuracy, sensitivity, and specificity. I used Python and building blocks to code the app, and I used a pre-trained AI model from Clarifai for the AI part of the app. And these were all the methods I used for this project. I used various materials to create the app prototype. Materials: computer ( to code/create the app), smartphone ( to test the prototype of the app), Clarifai pre-trained AI model ( to use and train the AI to detect breast cancer on different types of density), MIT App Inventor ( the platform I used to create the app) which has a drag and drop interface making it beginner friendly.

 

Broad Review:

Even though the idea of AI ( artificial intelligence ) being used in medical fields/screening isn't new it hasn't been implanted into hospitals/clinics to help with medical screening and cancer detection. Mammograms/MRIs have been implemented and the inaccurate/late detections of these machineries have affected the survival rate of a cancer patient.  Although we don't know whether AI can detect breast cancer in different categories of breast density we can train it and test it. As I have mentioned previously I'll calculate the accuracy, sensitivity, and specificity and compare it to those of an MRI ( magnetic resonance imaging )  and a Mammogram. I'll conduct about three trials to pick the pre-trained AI model and twenty trials to test each breast density with the pre-trained AI model. Since there are four categories of breast density ( fatty, scattered fibroglandular density, heterogeneously dense, and extremely dense ) the twenty trials will be divided into four which means four trials for each breast density category. I'll use three different formulas to calculate the accuracy, sensitivity, and specificity of the AI model.

                                                                                Formula:

Accuracy =  True positives + True negatives                  Sensitivity =    True positives             Specificity =  True Negatives

                   ______________________                      _______________________                    ________________________                                   

                       Total predictions                             True positives + false positives                 True negatives + False positives 

 

 

Breast cancer has taken a toll on many women's lives in various ways, such as changes to body shape, hair loss, early menopause, loss of Fertility, and even emotional tolls with strong feelings of anger, anxiety, sadness, and so on. Learning this empowered me to find something that would help the process of breast cancer diagnosis, something that would be quick and accessible. At first, it was hard to find the exact type of cancer that I wanted to base my app off of, but I had to base my app and train the AI model on a type of cancer that was common. That was when I got the idea of creating an AI-powered model that could diagnose breast cancer. My objective/goals are to create an app that can detect breast cancer accurately and revolutionize breast cancer diagnosis. The significance of AI ( artificial intelligence) in cancer detection overall is very important to the healthcare industry for Medical Imaging Analysis. Medical imaging is one of AI’s successful/accelerating AI healthcare applications ( artificial intelligence). One of the top reasons I chose this project is because I wanted to do something that could make a difference in the healthcare industry, and when I saw an article talking about the improvements in AI blueprints, especially in the healthcare industry

 

After I finish calculating the accuracy, sensitivity, and specificity of the pre-trained AI ( artificial intelligence ) model, I'll conclude whether I can use AI ( artificial intelligence ) to detect the tumours in every category of breast density or not. Then, I can make my conclusion based on the results and analysis.

Breast cancer is one of the many types of cancer that women worldwide struggle with daily. Early cancer detection can significantly enhance survival rates, as it allows for treatment at an initial stage instead of when the cancer has progressed. Mammograms/MRIs can often overlook cancerous tissues, particularly in women aged around 22-25 years, due to the density of their breast tissue, which makes it more difficult for machines to identify cancerous areas. Recent studies indicate that AI has potential in breast cancer detection and in identifying other types of cancer. My primary goal for this project is to develop an AI-powered app capable of accurately scanning mammograms/ultrasounds. Breast cancer has profoundly impacted many women's lives in various ways, including changes to body shape, hair loss, early menopause, infertility, and significant emotional challenges, such as feelings of anger, anxiety, and sadness. empowered/encouraged me to make something that would help women all over the world, something that would revolutionize the process of a breast cancer diagnosis. Make it quicker and easier ( easier in a way of easy access). Some methods and tools I used to make the app were  Clarifai, a platform that uses AI (artificial intelligence), and MIT App Inventor, a beginner-friendly platform(s) for people who want to create a mobile app.

First, I began by testing the AI ( artificial intelligence ) models to determine which was the most accurate so that I could move on to the next phase. I used it to scan mammograms/ultrasounds to determine the main point of the project: Can AI ( artificial intelligence ) detect breast cancer accurately? I put the results in the chart below.

 

  These results are on accuracy, and this is how accurate the AI model is after using the formula I stated in the method section.

 

IBM Watson	Google Cloud Console	Clarifai General Image     Recognition
Trial 1: ( scattered fibroglandular density) Scattered fibroglandular density is listed as not dense in the picture above. Results: 81%	Page unresponsive; therefore, not applicable ( then when page was responsive )	Trial 1: ( scattered fibroglandular density) Scattered fibroglandular density is listed as not dense in the picture above. Results: 94%
Trial 2: ( fatty ) Fatty density is listed as not dense in the picture above. Results: 79-82% ( results range because I tested twice accidentally and I decided to put both results)	Page unresponsive; therefore, not applicable ( then when page was responsive )	Trial 2: ( fatty )  Fatty density is listed as not dense in the picture above. Results: 89%
Trial 3 : ( heterogeneously dense) Heterogeneously dense is listed as dense in the picture above. Results: 91%	Page unresponsive; therefore, not applicable ( then when page was responsive )	Trial 3 : ( heterogeneously dense) Heterogeneously dense is listed as dense in the picture above. Results: 93%
Trial 4: ( extremely dense ) Extremely dense is listed as thick in the picture above. Results: 84%	Page unresponsive; therefore, not applicable ( then when page was responsive )	Trial 4: ( extremely dense ) Extremely dense is listed as dense in the picture above. Results:  87%

 

All these results were listed after the pre-trained AI models were trained with the same pictures and then tested with different ( different as in not the same as the ones I trained it with but the same for each model ) pictures so that both pre-trained AI ( artificial intelligence ) can be given the same chance. After looking through the results, I decided to choose Clarifai's General Image Recognition pre-trained AI ( artificial intelligence ) to conduct my experiments.

The next phase is to test the pre-trained AI ( artificial intelligence ) I chose with pictures of mammograms and test if AI ( artificial intelligence ) can be used to detect breast cancer on different types of density. After I have charted this phase, I move on to form my conclusion based on the results I get from testing the AI ( artificial intelligence ) model on the mammogram screenings.

Looking at the results from the chart, I've concluded that Clarifai General Image Recognition is the most accurate with being trained and will be the AI ( artificial intelligence ) model I will use for the experiment.

 

      These are the results of Clarifai General Image Recognition after concluding it's the most accurate after being trained (results may vary due to each trial; I didn't train it, or the training got less. 

 

 

                         ( training got less each trial or didn't get trained)

                                                Ethical Considerations and Arguments :

 

Although AI ( artificial intelligence ) has groundbreaking results in the medical field, there are ethical issues and some people's opinions on why AI ( artificial intelligence ) hasn't been embedded into hospitals/clinics. Some say that AI ( artificial intelligence )  has no "ethical morals," and some say that by using deep learning machines, we should embed ethical and moral principles into the algorithm. What is more concerning is that the article doesn't fail to state valid points as to why AI ( artificial intelligence )  shouldn't be used because of the confidentiality of patients but also states the fact that AI ( artificial intelligence ) will cause huge global employment for medical professions. The article says " If, finally, AI algorithms replace doctors’ evaluations, there is a rationale to wonder whether all medical doctors are going to be replaced by robots using AI algorithms." which bothers me to wonder if the person cared to think about when a breast cancer patient is lying in a hospital bed diagnosed with stage four breast cancer all because a radiologist failed to see the symptoms earlier or overlooked them. AI has and will make more groundbreaking achievements. I do agree with the article that AI ( artificial intelligence ) can't completely replace doctors; it can make it more accessible and faster.

 

                            This is the anatomy of the breast 

 

 

                                                                   Conclusion of analysis

After looking at the results of my analysis, I can tell that  Clarifai General Image Recognition, after being well-trained, it was more than capable of detecting breast cancer and excelled in scattered fibroglandular density, fatty, and heterogeneous dense. It didn't necessarily excel in extremely dense; it did well for a pre-trained AI ( artificial intelligence ) model. Even though answers are remained inconclusive, the question is answered. AI ( artificial intelligence ) can detect breast cancer in different types of density but not to the point it could replace doctors.

In conclusion, I finished the app prototype where you can upload your image ( mammogram) and send it to the pre-trained AI model for analysis. and it can reply ( only the fully published app can do that). The prototype can only show the coding, the buttons, the web, and the UI (user interface). I proved my hypothesis right when I said I think that AI is significantly improving in detecting not just breast cancer but all types of cancer. 

Independent variables: the form of AI algorithm used ( in my case, Clarifai's main general image recognition) and quality of the medical images.

Dependent variables: accuracy of the cancer detection false positive/negative rate. 

Controlled variables: test conditions, number of images.  

 

Hypothesis (which has been proven to be partially right)

My prediction is that my AI (artificial intelligence) powered app may scan breast cancer images accurately, but using an AI (artificial intelligence) powered app to assess the breast tissue could improve the diagnosis to summarize my hypothesis, an AI( artificial intelligence) powered app will be able to detect breast cancer accurately. If my AI-powered app does accurately, then this could revolutionize the healthcare system and even the healthcare software. But the app could run into some problems, just like the mammograms/MRIs ( magnetic resonance imaging), because the breast's density could affect accuracy. 

 

 Background Research 

 Breast cancer has taken a toll on many women's lives in various ways, such as changes to body shape, hair loss, early menopause, loss of Fertility and even emotional tolls with strong feelings of anger, anxiety, sadness, and so on. Learning this empowered me to make something that would help women all over the world, something that would revolutionize the process of breast cancer diagnosis. At first, it was hard to find the exact type of cancer that I wanted to base my app off of, but then I remembered an article that was being given out at a local store near my house. It was about breast cancer, and that was when I got the idea of creating an app that could diagnose breast cancer. My objective/goals are to create an app that can detect breast cancer accurately and revolutionize breast cancer diagnosis. The significance of AI ( artificial intelligence) in cancer detection overall is very important to the healthcare industry, for example, in Medical Imaging Analysis. Medical imaging is one of AI’s successful/accelerating AI healthcare applications ( artificial intelligence). One of the top reasons I chose this project is because I wanted to do something that could make a difference in the healthcare industry, and when I saw an article talking about the improvements in AI’s blueprints especially in the healthcare industry I decided that I wanted to do something like this for the science fair. 

                                                               Sources of Error

1. I didn't publish the app.

2. I used a pre-trained AI model.

3. I should've done more trials for more accurate results.

4. Again, I could've coded an AI model that was specifically made to detect breast cancer ( I didn't have enough computing power or data ).

5. I  should've tested more AI models.

                                                            Summation of project

In summation, the target of this project was to see if AI (artificial intelligence) integrated into an app can detect breast cancer on different types of breast density. From this project, I've learned a lot about cancer and its biological background and even how to code ( I coded using builder blocks on MIT app Inventor). Now, the main question is whether AI (artificial intelligence) integrated into an app can detect breast cancer on different types of breast density, and the answer is yes it can. AI (artificial intelligence) has and will continue to make enhancements in the healthcare industry, but in my case, I have not coded a pre-trained model; I have simply used one, and while the answer to my project is clear, the results remain partially inconclusive. So, AI  (artificial intelligence) is a potential source of detection of cancer, but in my case, I had various sources of error, which has led me to this conclusion that AI ( artificial intelligence )  can help enhance our healthcare industry, but we would need various improvements in its blueprints and we also need to take consideration for its morals and ethical views for me I have used a pretrained-model so I couldn't give accurate results for a pre-trained model ( what I mean by this is that I didn't code a AI model specifically for detecting breast cancer but if I did I may have had better results and we could've seen the full potential of AI) since it is not the full potential of AI (artificial intelligence). However, all together, this project was informal and a partial success.

 Cysf sources

https://www.breastcancer.org/screening-testing/artificial-intelligence

 

https://www.cancer.gov/research/infrastructure/artificial-intelligence

 

https://www.aamc.org/news/it-cancer-artificial-intelligence-helps-doctors-get-clearer-picture

 

https://www.cancer.gov/news-events/cancer-currents-blog/2022/artificial-intelligence-cancer-ima ng

 

https://www.bcrf.org/blog/ai-breast-cancer-detection-screening/

 

https://cancer.ca/en/living-with-cancer/coping-with-changes/your-emotions-and-

 

cancer#:~:text=Loneliness%20and%20isolation&text=You%20may%20feel%20too%20sick,often%20as%20they%20did%20before

 

https://jamanetwork.com/journals/jamaoncology/fullarticle/2811409 

 

https://www.science.org/doi/10.1126/science.aay9040

 

https://buildfire.com/learn-to-code-mobile-app-fast/

 

https://pmc.ncbi.nlm.nih.gov/articles/PMC6403009/#:~:text=AI%20affords%20high%20objectivity%20through,visual%20assessment%20by%20human%20experts.

 

https://appinventor.mit.edu/

 

https://www.clarifai.com/

 

https://www.cancer.gov/types/common-cancers#:~:text=The%20most%20common%20type%20of,are%20combined%20for%20the%20list. 

 

https://www.researchgate.net/figure/Formulae-for-a-accuracy-b-sensitivity-and-c-specificity_fig8_319208171

 

https://www.pneumon.org/Risks-of-Artificial-Intelligence-AI-in-Medicine,191736,0,2.html#:~:text=Currently%2C%20the%20major%20risk%20concerns,are%20also%20not%20fully%20convinced. ( I used some used points from this article but only to back up my arguments)

https://www.ncbi.nlm.nih.gov/books/NBK9963/

First I would like to thank God for giving me the motivation. Second I want to thank my family for providing me with the materials and listening to me over and over as I presented my project to them and for helping me acquire all the materials needed for my project. And I would also like to thank the science fair coordinator (Ms.Kale) for making this possible. Again hank you to all those who helped me finish this project.