A regionally recognized Radiology Practice in Florida, with 7 MRI machines, provides imaging services, mostly to out-patient referrals from Healthcare providers in the region. The center specializes in MR reading and is committed to improving the health and well-being of its local community. Their focus on building a reputation among Healthcare Providers by providing high-quality diagnostic services has driven performance improvement efforts at the institution. As one of the initiatives towards this, the use of AI-assisted reading to improve the report quality for MR spine exams, was undertaken.
Background and Challenges:
Back pain and neck pain have a high incidence (~90%) especially in the population above 40 years of age, with an upward trend due to changing population demographics and occupation. The MR Spine exams are the preferred choice of investigation for evaluating back pain and these spine exams constitute about 35% – 40% of the overall MR workload. Results of the Bi-annual Internal quality audit for the second half of 2020 showed that 39% of MR Spine reports showed disagreements between radiologists and 34 reports out of 1000 had reported at least one pathology at the wrong spinal level, making it a potential legal suite. From radiologists’ point of view, Reading of spine exams was very repetitive and subjective and could be a possible factor for high inter-observer variations.
Synapsica presented its solution Spindle, an AI tool to assist MR Spine reading to help improve the quality and objectivity of outgoing reports. Spindle AI works on MR images of the spine (both cervical and lumbar) and assists in reading by:
- Automatically labels vertebral levels
- Evaluates stenosis with central canal and thecal sac diameters
- Grades Listhesis and measures vertebral displacement
- Evaluates disc degeneration and disc heights
- Detects compression fractures
Implementation and Security:
- Integration Ready
- HIPAA compliant
- Encrypted data transfer
The practice was already using a RIS-PACS solution to manage its Radiology workflow. Spindle AI was integrated with the existing workflow system, which received anonymized DICOM images and sent back information in GSPS and JSON format. The receiving RIS-PACS system showed images annotated with GSPS overlay and JSON results with objective evaluation of pathologies could be included in the report with a click of a button. The spindle was employed as a “Clinical Research” tool and aided in creating MR Spine readings, which were evaluated again in a Bi-Annual Quality audit after 6 months.
Overall, AI-assisted reading with Spindle was able to reduce errors by 28%. The radiologists were able to use Spindle as part of the existing workflow and found pre-labeled vertebrae in images that made the reading convenient. The reading team positively perceived AI as a second pair of eyes to pick up pathologies that could have been missed during eyeballing the images and objective measurements by AI helped as a decision support tool in borderline cases.
|Use Cases||Without Spindle||Spindle AI assisted|
|Wrong level errors||34 (out of 1000)||5 (out of 1000)|
|Missed Transitional vertebrae||27 (out of 1000)||15 (out of 1000)|
|Disagreements in Stenosis||19%||11%|
|Missed Compression Fractures||21%||6%|
Chief Medical Officer:
“The initial results from Spindle-assisted reading are very promising. The team was very responsive with the integration of AI in our existing workflow, training was easy and staff could pick it up quickly for regular use. The providers we work with have also noticed the change in our reporting pattern and received it positively with increased business, which was one of our major concerns. We are excited about the launch of upcoming features that will cover disc herniation and neural foramina assessments, I believe it will help improve the quality of our reports further and reduce our turnaround times.”