Learn how we helped a customer in Medical Imaging ensure higher detection accuracy of cancer screening applications than the conventional image processing methods.
Deep Learning - Early Detection of Lung Cancer with CNN
Lung Cancer remains the leading cause of cancer-related death in the world. If detected earlier, lung cancer patients have much higher survival rate (60-80%). The current CADe/CADx systems have sensitivity of 80-85% on average with a recent study reporting 94% with a higher false positive rate of 7 per scan.
The scope of this project is to use Deep Learning techniques for early detection of lung cancer from CT scans and use the results in clinical diagnostics and cancer screening applications to support radiologist's diagnosis.
To start with, the QuEST Global team working on Medical Imaging prepared CT DICOM images from LIDC data set for Deep Neural Network Training. The LIDC dataset contains annotated scans with nodule location. The architecture for Convolutional Neural Network (CNN) using Caffe for nodule detection was created and the network was trained using the Dataset. This was followed by analyzing and Fine tuning the network performance and accuracy. After validation and testing, the model generated was deployed and used to detect nodules in new CT images given.
Tools & Technologies used includes:
The engineers at QuEST Global was able to ensure higher detection accuracy than conventional image processing methods. The team is currently working on the initial prototype with an aim to achieve sensitivity above 95% and low false positives per scan.