Bios: Prannav Murukesh,Adharsh Laguduva

Program Track: Skills Development

GitHub Username:

imLockedIn -Prannav Murukesh

adharsh-prajith -Adharsh Laguduva

What was your favorite seminar? Why?

My favorite seminar was the new where we were taught on how to present our projects and how to answer questions and be preferred for them. This was my favorite one as it was extremely helpful for the future as I would need those methods to successful present the project and present it to the public. -Prannav Murukesh

My favorite seminar was the sminar on Open AI by Jason Wei -Adharsh Laguduva

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

It was extremely knowledgable and I learned a lot of new information about machine learning and AI. -Prannav Murukesh

An opportunity to apply the skills that I have been learning over the past few years into the medical field and into a real world project. -Adharsh Laguduva

Blog Post

Abstract:

The goal of this research project is to improve early identification of liver cancer, a disease with a high death rate and a poor prognosis because of late-stage diagnosis. To this end, machine learning and computer vision techniques are being applied. The majority of current diagnostic techniques rely on the manual interpretation of CT scans, which can be time-consuming and subject to human error.. This study intends to increase the consistency and speed of diagnosis by creating a machine learning model that can accurately identify liver cancers in CT scans. With the use of sophisticated image processing techniques, the suggested model analyzes and segments liver tissues to more accurately identify tumorous areas. This strategy could help with ongoing patient monitoring for those who are at risk of recurrence in addition to increasing the rate of early identification, providing physicians with a more dependable weapon in the fight against liver cancer.

Blog

Scientific premise:

The goal of this project is to develop a machine learning model that is able to detect tumors in the liver in CT scans. The model aims to improve early stage liver cancer detection by creating a model using deep learning algorithms.

Aims/goals:

• Develop a robust machine learning model capable of accurately detecting liver tumors in CT scans.

• Utilize deep learning techniques to train the model on a diverse set of annotated liver CT images.

• Evaluate the model's performance through rigorous testing and validation, using metrics such as Dice coefficient, precision, and recall.

• Optimize the model's accuracy by iterating on its architecture, hyperparameters, and training data.

Our approach: Image Segmentation with Preprocessing

Data Collection and Preparation

• Data Gathering: We collected liver CT scans and their corresponding tumor labels from http://medicaldecathlon.com/

• Preprocessing: We used a custom data generator to load and prepare the images. This involved normalizing the images, resizing them to 512x512 pixels, and selecting random slices to increase variety in the training data.

Model Development

• Custom Model: We created a U-Net model designed to identify and highlight tumors in 2D images.

  • Encoder : The model first compresses the image to capture important features.

  • Bottleneck: A middle layer further processes these features to understand the most crucial parts of the image.

  • Decoder: The model then rebuilds the image, focusing on identifying tumor regions with the help of the encoded features.

Training and Optimization

• Training Process: We trained the model to learn from the images using a loss function that balances both pixel accuracy and the overall ability to segment tumors..

Validation and Testing

• Testing: After training, we tested the model on separate data to ensure it works well on new images. We measured its accuracy using metrics like the Dice coefficient.

Deliverables

• A trained segmentation model that can accurately detect and localize liver tumors in CT scans.

Conclusion:

The research project developed a machine learning model for detecting liver tumors in CT scans, achieving a moderate success rate. The model was able to identify some tumorous regions but still requires significant optimization. While the results are promising, further refinement and additional training data are needed to improve accuracy and reliability.