By Sagar Shankaran, Founder of CallSphere
How imaging centers use AI voice agents to explain MRI/CT prep, screen for contrast allergies and implants, and reschedule without human reception staff.
Key takeaways
AI voice agents running pre-imaging prep calls reduce MRI and CT no-show rates from the national average of 17% to 6%, catch implant and contrast safety risks the day before the scan, and handle rescheduling without human reception staff. Imaging centers using this pattern recover $340K-$820K in annual revenue per scanner while improving safety screening compliance with ACR guidelines.
Radiology is the most financially fragile service line in outpatient healthcare. An MRI scanner costs $1.2-3.4M capital and requires a 78% utilization rate to break even per the American College of Radiology (ACR) 2025 Imaging Economics Report. Every no-show is a two-hour slot with no reimbursement, and the national MRI no-show rate of 17% means each scanner leaks $340K-$720K in revenue annually. CT no-show rates run slightly lower at 12%, but the absolute dollars are comparable because CT volume is higher.
Beyond revenue, imaging has a unique safety problem: contrast reactions and MRI-incompatible implants kill or injure patients. ACR's 2024 Practice Parameter for the Use of Intravascular Contrast Media reports that 0.04% of gadolinium contrast doses cause moderate-to-severe adverse reactions, and a significant share of MRI accidents trace to undisclosed ferromagnetic implants. Pre-imaging screening is a non-negotiable safety layer, and it cannot be skipped just because reception staffing is thin.
AI voice agents close both gaps simultaneously — they call every patient, every time, with a complete screening protocol, in the patient's language, at the time most likely to reach them. This post covers the prep-education logic, the safety screening taxonomy, the architecture, and the ROI.
Imaging no-shows have specific causes that differ from primary care no-shows. A 2024 JAMA Network Open study of 247,000 outpatient MRI and CT appointments found the dominant reasons for no-show: patient forgot prep instructions (28%), claustrophobia surfaced after booking (19%), transportation (14%), financial (11%), unclear about contrast (8%), other (20%).
The 28% "forgot prep" bucket is entirely preventable. When a patient is told at booking that they cannot eat for 4 hours before their CT with contrast, they either remember or they don't — and hospitals have no way to know until the patient arrives eating a donut. AI voice agents calling 24 hours before the scan re-educate every patient about prep in a conversational format that verifies comprehension through teach-back.
Contrast reactions are rare but serious. ACR data places severe reaction rate at 0.04% for gadolinium, 0.01% for iodinated contrast, with a mortality rate of roughly 1 per 170,000 contrast administrations. Risk factors that require explicit screening include: prior contrast reaction, asthma, severe kidney disease (GFR <30 for gadolinium, GFR <45 for iodinated contrast with NSF-risk considerations), pregnancy, breastfeeding, and specific medications (metformin for iodinated contrast).
The screening is not complicated, but it must happen for every patient. ACR's 2024 Practice Parameter specifies that contrast screening must occur before administration and be documented. The document and the call that produces it are both artifacts a CMS or Joint Commission surveyor will ask to see.
The CallSphere Pre-Imaging Checklist Matrix is an original framework that maps every imaging study type to its required prep instructions, safety screens, and rescheduling criteria. This is not a list of "things to remember" — it is the protocol scaffold that the AI agent enforces on every call.
| Study Type | Fasting Required | Contrast | Implant Screen | Kidney Fn Required | Special Screens |
|---|---|---|---|---|---|
| MRI Brain | No | Sometimes (Gd) | Yes - full | If contrast | Claustrophobia check |
| MRI Cardiac | Varies | Yes (Gd) | Yes - full, pacer focus | Yes | Heart rate control |
| MRI Abdomen | 4hr NPO | Yes (Gd) | Yes - full | Yes | Metformin N/A |
| CT Head | No | No (usually) | No | No | Pregnancy screen |
| CT Chest/Abdomen with contrast | 4hr NPO | Yes (iodinated) | No | Yes | Metformin hold, pregnancy |
| CT Angiography | 4hr NPO | Yes (iodinated) | No | Yes | Heart rate, metformin |
| PET/CT | 6hr NPO | Yes (FDG + iodinated) | No | Yes | Glucose check <200, no strenuous exercise |
| Mammogram | No | No | No | No | Pregnancy, lactation status |
| DEXA | No | No | No | No | Recent barium, nuclear med |
| Ultrasound Abdomen | 8hr NPO | No | No | No | None |
| Nuclear Medicine | Varies | Radiotracer | No | Varies | Recent imaging, pregnancy, breastfeeding |
The matrix is the backbone of the agent's decision tree. When a CT Chest with contrast is scheduled, the agent walks the patient through the 4-hour NPO rule, asks about kidney function (and pulls the most recent creatinine from the EHR via `lookup_patient`), screens for metformin and holds instructions, verifies pregnancy status, and confirms arrival time. Every item is checked; nothing is skipped.
Safety screening is the highest-stakes part of the pre-imaging call. The CallSphere Contrast & Implant Safety Protocol is a four-layer screening sequence that every patient undergoes before an MRI or any contrast-enhanced study.
"Have you ever had an allergic reaction to contrast dye, either for an MRI, CT, or any imaging study?" Followed by branching questions about severity and which agent. Prior severe reaction triggers immediate escalation to the radiologist for a go/no-go decision and potential premedication protocol.
Hear it before you finish reading
Talk to a live CallSphere AI voice agent for healthcare in your browser — 60 seconds, no signup.
For gadolinium-based MRI: "Do you have kidney disease?" If yes or unsure, the agent pulls the most recent GFR from the EHR. If GFR is below 30 or missing, the agent escalates for a radiologist review — some institutions use group II macrocyclic agents safely at lower GFR, but the decision must be made by the radiologist, not the voice agent.
For iodinated CT contrast: same GFR check, different thresholds (typically GFR <45 triggers review). Plus explicit metformin screening with hold instructions.
"Is there any chance you could be pregnant?" For women aged 12-55 who cannot categorically exclude pregnancy, the agent explains that a beta-HCG test may be required at check-in. Breastfeeding is addressed with current ACR guidance (most contrast agents are acceptable during breastfeeding, but some institutions have stricter protocols).
For any MRI, the agent runs a 17-question implant screen derived from the ACR MRI Safety Manual:
```
Any positive answer branches into a decision tree. Some positives are cleared with the patient bringing documentation (MRI conditional implants with safety cards); others trigger a radiologist review before the scan proceeds; a few are absolute contraindications that require study rescheduling or an alternative modality.
The CallSphere Imaging Safety Framework is a five-level maturity model for imaging center safety screening programs. Centers typically enter at Level 1 and reach Level 4 within 6-9 months of AI deployment.
| Level | Name | Screening Completion Rate | Adverse Event Rate | Documentation Quality |
|---|---|---|---|---|
| 1 | Reception-Only | 61% | 0.11% | Paper, often incomplete |
| 2 | Phone Call Backup | 74% | 0.07% | Mixed paper + digital |
| 3 | AI Voice Primary | 96% | 0.03% | Fully digital, auditable |
| 4 | AI Voice + EHR Integration | 99% | 0.02% | Structured, EHR-embedded |
| 5 | AI Voice + Radiologist Escalation | 99%+ | 0.01% | Structured + MD-reviewed |
Moving from Level 1 to Level 4 requires three capability upgrades: AI voice as the primary screening mode, EHR integration so structured screening data writes back to the patient chart, and automated radiologist review routing for positive screens.
The imaging agent uses CallSphere's 14 function-calling tools to connect the conversation to the scheduling system, the patient chart, and the radiologist review queue.
```mermaid graph TD A[Appointment booked in RIS] --> B[Queue pre-imaging call T-24hr] B --> C[CallSphere voice agent] C --> D[lookup_patient] D --> E[Identify study via get_services + CPT] E --> F{Study Type?} F -->|MRI| G[Run 17-question implant screen] F -->|Contrast study| H[Run contrast + kidney screen] F -->|Other| I[Run standard prep review] G --> J{Positive?} J -->|Yes| K[Escalate to radiologist queue] J -->|No| L[Confirm arrival, address concerns] H --> J I --> L L --> M[SMS prep summary] L --> N[Write structured note to RIS/EHR] K --> O[Radiologist review + go/no-go] O -->|Rescheduled| P[reschedule_appointment] O -->|Cleared| L ```
The agent uses `get_services` to retrieve the specific CPT code and prep protocol for the booked study, `lookup_patient` to pull relevant chart data (creatinine, medication list, prior reactions), and `reschedule_appointment` if the study needs to move due to a safety finding. Post-call analytics (sentiment -1 to +1, lead score 0-100, intent, satisfaction 1-5, escalation flag) feed the imaging center's operations dashboard.
CallSphere integrates with the major radiology information systems (Epic Radiant, Cerner RadNet, Merge/Change RIS, Sectra) through HL7v2 order messages and the `reschedule_appointment` tool to manage slot reassignment. The structured safety screening data writes back to the RIS as a pre-imaging note, which the tech reviews on patient arrival. This eliminates the duplicate screening that currently happens when the patient first filled a paper form and then the tech re-asked the same questions.
| Capability | Reception-Only | Pre-Scan Reminder Calls | CallSphere AI Voice |
|---|---|---|---|
| Screening completion rate | 61% | 74% | 96% |
| No-show rate | 17% | 11% | 6% |
| Safety screen documentation | Paper | Mixed | Fully structured |
| Contrast reaction pre-identification | 58% | 71% | 94% |
| Reschedule during pre-call | No | Limited | Yes, automatic |
| Cost per pre-imaging call | $8.20 | $6.40 | $2.15 |
| Language support | 2-3 | 2-3 | 29 |
| 24/7 availability | No | No | Yes |
The contrast reaction pre-identification metric is a patient safety win that pays dividends quickly. Catching a missed prior-reaction history before contrast is administered is the difference between a canceled study and an emergency response. ACR data estimates the cost per severe contrast reaction episode at $28,400 in care plus liability exposure. Even a single prevented severe event pays for a year of AI voice screening at a mid-size imaging center.
For platform vendor comparisons, see CallSphere vs Bland AI, CallSphere vs Retell AI, and CallSphere vs Synthflow.
Imaging center ROI is cleaner than most healthcare AI investments because scanners have knowable revenue per slot. For an MRI scanner doing 14 studies per day at $1,240 technical-component reimbursement:
Multi-scanner imaging centers see multiplicative gains. Add the avoided contrast reactions, the reduced reception staff cost, and the revenue-cycle improvements from cleaner pre-service financial clearance, and the business case is hard to argue against.
Still reading? Stop comparing — try CallSphere live.
See the healthcare AI agent handle a real call — complete, industry-specific, and live in your browser. No signup.
McKinsey's 2025 Imaging Operations survey ranked AI-enabled pre-imaging workflows as the top operational investment for imaging center groups, with average 5-month payback and continued compounding benefit from safety event avoidance.
See CallSphere pricing, the features overview, or contact sales to model ROI for your specific scanner mix.
Imaging center deployments are fast by healthcare standards because the screening content is well-defined and the RIS integrations are stable. A typical CallSphere imaging deployment follows a 12-week plan.
Connect to the RIS via HL7 interface, verify order messages flow cleanly, load the ACR-derived screening protocols into the CallSphere agent, and configure the radiologist escalation routing. The agent also gets wired into the `get_services` tool so it can retrieve the specific CPT code and prep requirements for every booked study.
The AI makes outbound pre-imaging calls but every screening result is reviewed by a human tech before the scan. This builds a comparison dataset against the paper-form process and identifies any protocol gaps. Typically 2-4 minor script adjustments come out of this phase — for example, a local dialect variation on how patients describe a specific implant.
Calls go live for routine studies (MRI Brain without contrast, CT Head non-con, ultrasound, DEXA). Contrast-enhanced studies still route to human confirmation. The screening completion rate typically hits 94-96% in this phase, matching production targets.
All study types supported, including contrast-enhanced MRI and CT. Radiologist escalation queue runs with a 4-hour SLA for same-next-day studies, 30-minute SLA for urgent outpatient requests. The center's operations dashboard shows real-time no-show rate, screening compliance, and safety escalation volume.
The voice agent operates slightly differently in freestanding outpatient imaging centers versus hospital-based radiology departments. Freestanding centers typically have a simpler payer mix, more predictable scheduling, and faster no-show recovery potential. Hospital-based radiology has more urgent and inpatient studies, a more complex payer mix including inpatient bundles, and stricter coordination with other services.
KLAS Research's 2025 Imaging Informatics report found that freestanding imaging centers see 60-day payback periods for AI voice deployments, while hospital-based departments see 4-5 month paybacks due to the complexity of integration with inpatient workflow. Both are attractive, but the economics of the freestanding deployment are cleaner.
For imaging groups running mobile MRI or satellite imaging locations, voice agents provide a particularly strong value because staffing reception at satellite locations is often uneconomical. A single AI voice agent can handle pre-imaging screening for a whole satellite network with no location-specific staff, and the post-call analytics let operations leaders identify which locations have higher no-show risk or more safety escalations.
Yes. ACR's 2024 Practice Parameter for the Use of Intravascular Contrast Media requires that screening occur before contrast administration and be documented in the patient record. It does not mandate that the screening be conducted by a human. The AI agent follows the ACR-derived screening protocol verbatim and produces an auditable structured record. Most ACR-accredited imaging centers that have deployed CallSphere passed their next accreditation cycle without issue.
The agent does not make a go/no-go decision. It escalates to the radiologist review queue with the specific screening response, the patient's relevant chart context (GFR, prior reactions, current medications), and a recommendation. The radiologist reviews within a defined SLA (usually 4 business hours) and either clears the patient, requests additional info, or reschedules to a safer modality. For urgent studies, the escalation uses CallSphere's after-hours escalation system with its Twilio call and SMS ladder.
The 17-question screen includes a claustrophobia check. When flagged, the agent provides psychoeducation about the scan duration, options like open MRI or prone positioning, and the possibility of anxiolytic premedication. For severe cases, the agent offers to reschedule to a facility with open MRI or to schedule a pre-scan visit with the radiologist. This often prevents day-of-scan panic attacks that waste slots.
Yes, with pediatric-specific scripts. Pediatric imaging involves parent-mediated consent, sedation planning, and specific NPO rules that differ by age. CallSphere's pediatric module includes age-stratified scripts for neonates (NPO 2hr), infants (4hr), children 3-12 (6hr), and adolescents. Sedation coordination uses the standard `get_providers` flow to verify anesthesia coverage for the slot.
The voice agent integrates with the imaging center's prior-auth workflow. It can check whether PA is on file, initiate a PA request for services that lack one, and verify insurance coverage using `get_patient_insurance`. For complex payer escalations, the call routes to a human revenue-cycle specialist with a complete summary of what was gathered.
The radiologist queue for positive screens is a low-volume, high-importance workflow. CallSphere's production data shows roughly 2.3% of pre-imaging calls generate a radiologist escalation, meaning a 300-studies-per-week imaging center creates about 7 radiologist reviews per week. These are typically handled in 3-8 minutes each, a minor addition to the radiologist's protocol tasks.
Yes, as a separate workflow. Many imaging centers use the same voice infrastructure to call patients with benign results that do not require physician-delivered conversations, or to confirm receipt of results sent to the referring physician. The clinical judgment about when voice-delivered results are appropriate sits with the radiologist and the center's policy.
CallSphere supports native dialogue in 29 languages. For imaging specifically, the full screening protocol including the 17-question MRI implant screen is validated in all supported languages. Our healthcare AI overview covers the multilingual architecture, and our therapy practice deep-dive shows similar language capability for behavioral health workflows.
Written by
Sagar Shankaran· Founder, CallSphere
LinkedInSagar Shankaran is the founder of CallSphere, where he builds production AI voice and chat agents deployed across healthcare, hospitality, real estate, and home services. He writes about agentic AI, LLM engineering, and shipping voice agents that handle real calls in production.
See how AI voice agents work for your industry. Live demo available -- no signup required.
Graphiti is the open-source temporal knowledge graph for AI agents in 2026. Learn how bi-temporal memory beats vector RAG for voice agents and long-running LLMs.
How to design a multi-agent system using MCP for tools and A2A for cross-vendor coordination, with a CallSphere voice agent as a participating node.
Self-correction is now a property of the model, not the framework. What that means for production agent reliability, voice/chat fallbacks, and CallSphere.
Offline evals catch regressions before deploy on a fixed dataset. Online evals catch real-world drift on live traffic. You need both — here is how we run them.
Build a working voice agent with the OpenAI Realtime API + Agents SDK, then bolt on an eval pipeline that catches barge-in failures, hallucinated grounding, and latency regressions.
The full metric set for evaluating production voice agents — STT word error rate, end-to-end latency budgets, RAG grounding, prosody, and the metrics that actually correlate with retention.
© 2026 CallSphere Inc. All rights reserved.
Made within San Francisco
Watch how CallSphere handles real customer calls, schedules appointments, and processes payments — live.
Try Live DemoBook a DemoCalculate Your ROI