Network Interface Preprocessing DICOM Data for Medical Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current medical inference systems face inefficiencies due to the speed limitations of central processing units (CPUs) in preprocessing data and the time consumed in copying data between memory components, which affects the overall inference speed and accuracy.
Innovation Solution
The system employs a network interface and parallel processing units, including a Data Processing Unit (DPU) and Graphics Processing Unit (GPU), to preprocess data from a Digital Imaging and Communications in Medicine (DICOM) format into a tensor format, optimizing data processing and inference operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If CPUs perform preprocessing operations on input data, then data can be prepared for inference engines, but the speed of CPUs limits the overall inference speed
Solution Approach 1:
The patent segments the preprocessing task from the CPU by introducing a Network Interface Card (NIC) with dedicated preprocessing capabilities. The NIC handles data reception, format conversion from DICOM to tensor format, and preprocessing operations independently, while the CPU focuses on running inference engines. This segmentation allows parallel execution of preprocessing and inference operations, eliminating the CPU speed bottleneck.
Solution Approach 2:
The NIC acts as an intermediary device between the data source and the CPU-based inference engines. It performs format conversion and preprocessing operations that would otherwise burden the CPU, thereby improving overall system throughput and inference speed without sacrificing preprocessing quality.
2Speed
If data is copied from network interface to CPU memory and then to GPU memory, then data can be accessed by processing units, but time is consumed in these copying operations
Solution Approach 1:
The NIC performs format conversion and preprocessing operations before data reaches the CPU, preparing the data in the required tensor format upfront. This preliminary action eliminates the need for subsequent format conversion operations and reduces the number of memory copy operations required, as data is prepared in its final form closer to the processing units.
Solution Approach 2:
The patent merges multiple operations (data reception, format conversion, preprocessing) into a single device (NIC), eliminating the need for separate processing steps and reducing the number of data transfers between components. This consolidation reduces overall data copying time and improves data access speed.
3Reliability
If multiple preprocessing operations are performed to prepare data for inference engines, then reliable inference results are achieved, but the complexity of the system increases
Solution Approach 1:
The NIC serves as an intermediary that handles all format conversion and preprocessing operations, ensuring data is properly prepared for inference engines without requiring complex CPU-based preprocessing pipelines. This maintains inference accuracy while simplifying the overall system architecture by centralizing preprocessing functionality.
Solution Approach 2:
The NIC performs self-service by automatically converting data from DICOM format to tensor format and executing preprocessing operations without requiring extensive CPU intervention. This autonomous preprocessing capability maintains data quality and inference reliability while reducing system complexity.
Data Source
AI summary
Methods and systems for obtaining data having a first format, converting the data to a second format, storing the converted data in memory accessible by at least one parallel processing unit, and processing the converted data stored in the memory using the at least one parallel processing unit.