Phygital's Deep Learning Solution Helped a Leading Diagnostic Center Reduce Turnaround Time by 54%
Our client is a reputed diagnostic center serving hospitals and small clinics across California. They have a presence across 3 different locations in the state. Their clients rely on them for fast processing of X-rays, MRI scanning, and several pathology reports. Driven by a formidable team of radiologists, Radiologic technologists, Medical Imaging technicians, Image Processing specialists, and Medical physicists, the client had made a name for high-quality imaging services and accurate inference of conditions.
The client had to deal with a continuous inflow of X-ray and MRI scanning needs from its customers. This created a long pipeline of urgent imaging requirements, and our client needed help to maintain the right balance. The problem was further compounded by the interpretation method, which was primarily manual and depended heavily on the availability of a radiologist. On average, it took 50 minutes to generate a report and approximately 7 hours for the final report.
Additionally, high attrition rates slowed down the image review process and increased the turnaround time. It also led to an increased workload on in-house radiologists, leading to reporting errors.
Yet another problem faced by the center was the lack of a centralized system for data storage. This added to their disadvantage as it led to the splitting of professional communities and reduced contact between radiologists and clinicians.
These three major problems greatly impacted other aspects of the center's operations. This included:
- Improper implementation of the quality maintenance program
- Compromise with standard imaging protocols
- Poor communication because of increasing workload leads to errors
- Frequent X-ray equipment breakdowns lead to increased expenses
The client wanted to overcome these challenges for more streamlined operations and growth. So, they approached Phygital for a data-driven solution.
Phygital built a team of data scientists and analysts to study the problem. The team worked hand-in-hand with the center's radiologists and image-processing experts to look into the crux of the problem.
After a thorough understanding, our team concluded that the solution to the problem lay in leveraging a learning and classification technology that can process and differentiate images swiftly and accurately and understand how to assign importance to objects in the image (from reasonable and abnormal conditions).
Our experts chose Convolution Neural Network to develop a customized solution for the client. The reason for choosing this deep learning algorithm was that this technology requires less preprocessing than other classification algorithms and can easily learn filters and characteristics.
The primary goal of the intended solution was to serve as a real-time pre-screening mechanism to analyze images and automatically determine and assign the right image to the right radiologist based on expertise and availability. The other objective of the solution was to enable a centrally located expert to assist at multiple locations in real time while the patient was on the scanner table.
To develop the solution, our team collected approximately 1,30,000 X-ray images to train the Machine Learning model. The X-ray images were passed through an image process to enhance the diagnostic information and suppress the non-diagnostic information. This input image data was driven through multiple convolutions and pooling layers before the flattened data was distributed to the ANN classifier. Finally, we utilized a high-level API called Keras with a Tensorflow backend to build a neural network that served as a solution.
Benefits of Our Solutions
Our custom solution delivered the following benefits to the client:
- A centralized data management system to receive images from other centers
- Provide expert advice in real-time from a central location
- A standardized image analysis process across centers with up to 92% accuracy in image recognition and classification
- Elimination of human supervision for identifying important image features
- Proper workload balancing among radiologists eliminated errors and resulted in significant dollar savings
- A decrease in the turnaround time (TAT) by 54%
Get in touch with us to learn more about our services and expertise.