Bios: Raymond Gao,Omolara Dawson

Program Track: Skills Development

GitHub Username:

RaymondG248 -Raymond Gao

Laradawson07 -Omolara Dawson

What was your favorite seminar? Why?

My favorite seminar was when Jason Wei because he talked about a path of Ai that I didn’t had the opportunity to learn about. When listening to his past experience it gave me a deeper insight into the field. -Raymond Gao

Louis Vaickus Even though it was earlier on in the summer and I don’t quite remember everything from it I do remember his name and how enlighted I felt after the talk. -Omolara Dawson

If you were to summarize your summer internship experience in one sentence, what would it be?

The internship experience was hard at the beginning but at the end it’s rewarding. -Raymond Gao

Made a lot of mistakes, but I’ve come so far that it was worth it, and I’m forever grateful. -Omolara Dawson

Blog Post

Team 8

Skills Development Track

Using machine learning to classify squamous cells

How can we utilize machine learning to differentiate squamous cells from cervical cancer samples, distinguishing normal from abnormal cells, and how could this enhance diagnostic efficiency and patient care?

By: Omolara Dawson and Raymond Gao

Introduction

Cervical cancer remains a primary global health concern, with squamous cell carcinoma (SCC) being the most prevalent form among cervical cancer cases. Early and accurate diagnosis is crucial for effective treatment and improving patient outcomes. Traditionally, the diagnosis of cervical cancer involves pathologists manually examining cell samples obtained from biopsies. This process is labor-intensive and highly dependent on the pathologist's experience and expertise, which can lead to variability and potential delays in diagnosis.

The limitations of manual analysis highlight the need for more efficient and reliable diagnostic tools. Advances in machine learning offer promising solutions by automating complex tasks and reducing human error. Machine learning algorithms have demonstrated their potential in various medical imaging applications, including the classification of cancerous tissues. These algorithms can learn to identify patterns and features in images that may be challenging for the human eye to detect, leading to more consistent and accurate results.

Our project aims to address these challenges by applying machine learning techniques to classify squamous cells in cervical cancer samples. We have developed a comprehensive dataset of annotated whole slide images, where we have marked the key characteristics of the cells. This annotated data serves as a training ground for our machine learning model, which is designed to differentiate between normal and abnormal squamous cells with high precision.

By training our model on this dataset, we seek to enhance the diagnostic process, providing pathologists with a tool that can quickly and accurately classify cell images. This automation is expected to reduce the time required for diagnosis, minimize variability, and support more reliable detection of cervical cancer. Ultimately, our goal is to integrate this machine learning tool into clinical practice, improving diagnostic efficiency and contributing to better patient care.

The successful implementation of this project could lead to significant advancements in the way cervical cancer is diagnosed and managed, potentially setting a new standard for accuracy and efficiency in medical diagnostics.

some examples of normal and abnormal squamous cells

{width=”1.27in” height=”1.2051093613298338in”}

Normal

{width=”0.9693252405949256in” height=”1.3251531058617674in”}

Normal

{width=”2.2in” height=”1.1838090551181102in”}

abnormal

Methods

1. Annotated Image Dataset:

- Whole Slide Images (WSIs): High-resolution images of cervical cell samples, annotated by experts, serve as the primary data for training. Ensure the dataset includes a diverse range of normal and abnormal cells for effective model learning.

- **Annotation Tools:** Use tools like QuPath to annotate images with detailed labels, including cell boundaries and classifications.

2. Machine Learning Frameworks:

- TensorFlow, PyTorch, Sciki-Learn:These popular frameworks provide libraries and tools for building, training, and evaluating machine learning models, particularly deep learning models.

3. Computational Resources:

- Graphics Processing Units (GPUs): Utilize GPUs to accelerate the training process, especially for large datasets and complex models.

Discussion

Despite these advantages, there are challenges to consider. The quality and representativeness of the annotated dataset are crucial for the model’s performance. Ensuring that the dataset includes a diverse range of samples and accurately reflects the variability in squamous cell appearances is essential for developing a robust model. Furthermore, integrating the model into clinical workflows requires careful consideration of user interface design and clinical validation to ensure that it complements existing diagnostic practices effectively. Future work will focus on expanding the dataset to include more diverse samples, refining the model to improve its accuracy and generalizability, and exploring ways to integrate the tool seamlessly into clinical practice. Continued research and development in this area promise to advance cervical cancer diagnosis further and potentially extend the benefits of machine learning to other areas of medical imaging and diagnostics.

**What We Learned:**

1. **Annotating Squamous Cells with QuPath:**

We learned how to use QuPath, an open-source software, to efficiently annotate squamous cells in whole slide images. This involved familiarizing ourselves with QuPath’s tools for segmenting and labeling cell structures, which was crucial for creating a high-quality dataset for training our machine learning model.

2. ** Machine Learning:**

We acquired practical experience in using machine learning algorithms for classifying medical images. This involved understanding the processes of data preprocessing, model training, and evaluation, which enabled us to develop a model capable of distinguishing between normal and abnormal squamous cells.

**Conclusion:**

In this project, our objective was to utilize machine learning to enhance the classification of squamous cells from cervical cancer samples. We employed QuPath for precise annotation and advanced machine learning techniques to develop a model capable of accurately distinguishing between normal and abnormal cells. This approach holds the potential to improve diagnostic accuracy and aid pathologists in providing reliable and timely diagnoses. Despite our advancements in annotating images and building the model, time constraints prevented us from fully realizing the project's potential. We faced challenges in refining the model and conducting comprehensive testing, leading to the model not achieving the expected level of performance.Our future work will center on addressing these limitations by expanding the dataset, further optimizing the model, and integrating it into clinical workflows. These steps are vital for maximizing the impact of our approach and advancing the field of medical imaging and diagnosis.

Result

We learned how to manually annotate squamous cells and were able to distinguish between normal and abnormal squamous cells. Due to a limited amount of time, we were unable to complete certain tasks involving machine learning. The future we plan to complete the remaining steps we were unable to complete.