1.
Briefly explain your project:
CancerScope is a project I developed to assist doctors in analysing the progression of cancer using medical
images. Essentially, it’s a tool that helps doctors look at things like tumors and understand how they are
growing over time. I used machine learning and image processing to make this possible.
The challenge was getting the system to recognize patterns in the images accurately, so I worked a lot with
training the AI models to be precise in identifying and segmenting tumors from these images. I used
frameworks like TensorFlow and PyTorch, which helped with the deep learning part. The main goal was to
create something that could save doctors time while providing accurate insights, which could help in
making quicker treatment decisions.
It was really interesting for me because it combined my technical skills with something that could have a
real impact in healthcare.
2. Was this a team or solo project, and what role did you play in making
that project?
In CancerScope, I was part of a team where my primary focus was on both frontend development and
contributing to the AI integration. I worked on building the user interface using React, making it easy for
healthcare professionals to upload medical images and view the AI's analysis. The frontend needed to be
responsive and intuitive, which was critical since doctors would rely on it for fast and accurate results.
On the AI side, while other team members focused on the core image processing and model training, I
helped integrate the machine learning model into the platform, ensuring smooth communication between
the backend and the UI. This allowed us to display the AI’s analysis in real time, providing actionable
insights to oncologists. Collaborating with the team, we worked closely to ensure that the system was
reliable and easy to use.
3. How did you ensure the accuracy of your model?
To ensure accuracy, I used a large dataset of medical images to train the model, which allowed it to learn
more effectively. I also tested the model extensively, tweaking the parameters of the machine learning
algorithms to improve performance. For evaluation, I used metrics like precision, recall, and F1-score to
measure how well the model was identifying tumors. It took some iteration, but I was able to reach a level
where the model’s predictions were fairly accurate.
4. What challenges did you face while developing CancerScope?
One of the biggest challenges was dealing with the sheer volume of medical image data. Processing such
large amounts of data in a way that didn’t slow down the system was tricky. Another challenge was training
the deep learning models to correctly identify tumors without producing false positives or negatives. It took
a lot of trial and error, fine-tuning the model, and ensuring that the data was clean and well-prepared for
training.
5. Why did you choose TensorFlow and PyTorch for this project?
I chose TensorFlow and PyTorch because they are both widely used frameworks for deep learning, and they
have great support for the kind of image processing and machine learning tasks I needed to perform.
TensorFlow is great for deploying models in production environments, while PyTorch has a more flexible
interface for research and experimentation, which was important during the initial stages of developing
CancerScope.
6. What real-world impact do you see CancerScope having?
I believe CancerScope could make a significant impact by helping doctors make quicker, more informed
decisions. By speeding up the analysis of medical images, it allows healthcare professionals to focus more
on patient care rather than spending too much time on manual analysis. In the long run, it could improve
the accuracy of diagnoses and treatment planning, potentially leading to better patient outcomes.
7. How did you handle the labeling of medical images?
Labeling medical images was critical for training the model. I relied on a publicly available dataset that had
already been annotated by medical professionals. In a real-world setting, this would be a collaborative
effort with healthcare professionals who could provide accurate labels for the images.