intel-oneAPI

Team Name : TEAM MARINES

Problem Statement : MEDICAL IMAGE PROCESSING

Team Leader Email : rahultalari737@gmail.com

Team members email :

Ramganeshlankada@gmail.com

mayankgla21@gmail.com

srinub@gvpce.ac.in ( Professional )

Prototype Video

link : https://www.linkedin.com/posts/ram-ganesh-1b2189208_activity-7073002717509537792-rBJg/?utm_source=share&utm_medium=member_desktop

A Brief of the Prototype:

Benchmarking

General Un-optimized code

The Medical image processing application prototype consists of 3 main categories they are :

• Brain Tumour Segmentation
• RSNA Pneumonia Detection
• Kidney Tumour & Stone Detection

A well documented code for each category is present in medical_img_processing folder with all the

• Data Visualizations
• IPYNB files
• .py files
• Serialized models
• Images

Intel Optimized SYCL / DPC++ code

Once the general optimized code was done we started crafting the SYCL/DPC++ code using following path : Python --> C++ conversion --> DPC++ conversion --> SYCL conversion --> Intel Advisor

All the intel optimized SYCL/DPC++ codes are present in INTEL_ONE_API_medical_img_processing folder.

Intel Advisor Analysis

Detailed analysis of Intel Advisor for SYCL/DPC++ Pneumonia Detection is categoried into :

• Offload Modelling
• Threading
• Vector optimization
• Roofline Analysis

Images :

Offload Modelling

Vectorization

Threading

Acceleration

Tech Stack:

Machine learning Stack

• Tensorflow
• Keras
• PyTorch
• TensorBoard

Intel libraries

• Intel SYCL / DPC++ library
• Intel Base Toolkit
• Intel Advisor
• Intel VTune Profiler

Application Deployment Stack

• Streamlit
• HTML
• DICOM Viewer Embedding

Execution Instructions:

git clone https://github.com/Rahul-Talari/Intel-Api-hackathon/
cd Application 
streamlit run "path/to/multiple disease pred.py"

Remember that the file contains statically set paths to some dependent files, scrutinize the multiple disease pred.py files and the change the required.