Innovative educational technologies
1Broadcasts and conferences
- Real-time broadcast of an operation to a phone, a computer or Telegram. Students, residents and colleagues from other cities ask questions in the chat or by voice right from the browser.
- Conferences and seminars: several operating rooms and a lecturer in one broadcast; the host chooses the main image.
- Case conferences and remote mentoring: a consultant joins from a browser and talks to the operating room by voice. SIP video calling (add-on module) connects the operating room with a conference hall or a department's video terminal.
- Streaming to YouTube and other RTMP services is an add-on module; it starts only with permission from the operating room.
- Below the video: the patient monitor's vital signs without personal data. In the archive the recording and the vital signs share one timeline for reviewing a clinical case.
Competency: remote mentoring, case conferences and clinical case review.
See it in the demonstration:
Broadcast
Joint broadcasts
Telegram
Patient vital signs
Archive
AI in education · neural network diagnostics · AI in research
2AI at the doctor's workplace
The neural network works right in the operating room, the endoscopy and ultrasound rooms and the intensive care unit: hints appear on the screen in real time, and patient images never leave the hospital.
Endoscopy roomPolyps, neoplasms, bleeding and ulcers are outlined during the examination.
Ultrasound roomThyroid nodules are outlined on the ultrasound scanner image.
Operating roomAbdominal organs are highlighted in colour over the laparoscope image; in the DICOM viewer the «Organs (AI)» function highlights organs and vessels on MRI in 3D.
Intensive care unitVidicap Connect assesses the risk of deterioration for every bed every minute and explains the reasons. Doctors' marks on whether a signal was relevant are used to further train the model on the hospital server.
On Vidicap Touch 2.0 the network runs on the device itself; on Touch 1.0 it runs on the device if its performance allows, otherwise via Vidicap AI Server. All models are research versions marked «not for diagnosis». The doctor always makes the decision.
How doctors train a model: 10 steps
- Direction. A department defines the task, for example «Thyroid ultrasound: nodules»; the direction is created on the platform.
- Invitations. Specialist doctors and experts receive an invitation and create an account.
- De-identification in the browser. The video never leaves the doctor's computer. Only frames cropped to the image, without the name, date or metadata, are sent to the platform. The doctor confirms the de-identification with a checkbox.
- Image annotation. The doctor outlines findings on the frames. Frames without pathology are also included in the training data.
- Expert review. The expert confirms the annotations or returns the case with a comment.
- Split by patient. The data is split into training, validation and test sets; all frames of one patient belong to the same set.
- Model training. A Faster R-CNN detector is trained on an NVIDIA 12–24 GB graphics card, at the hospital or on Vidicap equipment. The data set is deleted after training.
- Thresholds and testing. Thresholds are chosen on the validation set. The final check on the test set is performed only once.
- Delivery to Vidicap. The research model is delivered to the device together with its test results. The doctor switches it on; the previous version is kept.
- Next version. Errors saved by the doctor on Vidicap are included in training the next version after de-identification and annotation.
First result: thyroid on open data
The model is trained on the open TN5000 data set (5000 images of nodules with a biopsy-confirmed diagnosis). It was tested on 1000 images not used in training.
0,85mAP50 on the test set
85,5 %sensitivity for suspicious nodules
88 %precision for the «suspicious» class
~48 %precision for the «benign» class
Точность для доброкачественных узлов ниже, так как в наборе 72 % узлов злокачественные и доброкачественных примеров мало. Это исследовательская модель, не для диагностики. Для повышения качества необходимы данные с ваших аппаратов, аннотированные вашими специалистами.
Department participation: a department proposes a direction, and its doctors perform annotation and expert review. No operating room is required; a browser is enough.
See it in the demonstration:
AI endoscopy
AI training
DICOM and organs on MRI
Guide «AI training on the platform» · Guide «AI ultrasound»
Virtual patient · digital technologies in education
3Test environment: the platform at work with virtual equipment
A virtual hospital is deployed on the test server. Patient monitors and the hospital information system are simulated in software and send data over the same protocols as real equipment.
- 20 beds and 2 operating rooms in three units, 22 virtual Mindray monitors with vital signs, alarms and waveforms over the HL7 standard.
- Clinical scenarios on demand: hypotension, desaturation, arrhythmia, bradycardia, admission, transfer and discharge of a patient.
- The ward screen and the NEWS2 score, notifications to the doctor on duty, protocols under order No. 254, AI-based deterioration risk assessment, integration with a model of the HIS and a virtual analyzer.
- in development Virtual Vidicap operating rooms: broadcast, recording and the examination archive on the test server without physical devices.
See it in the demonstration:
Vidicap Connect
Vidicap emulator
All patients and data on the test server are fictitious. No real equipment or data of healthcare organizations is connected to it.
Digital security and personal data protection
4Patient data stays in the hospital
- Names, medical records, vital signs and recordings are stored on the hospital's servers. The vidicap.cloud cloud receives only equipment status data, relays broadcasts to invited viewers and accepts de-identified frames that a doctor has personally submitted for AI training.
- Access levels: guest, nurse, doctor, head of department, technician. Each staff member signs in to Vidicap with a personal PIN or badge and gets access only to the permitted functions.
- Only invited viewers can see a broadcast; the viewer's name is shown over the video. Backups of the hospital server are encrypted with the hospital's own key.
- The development takes into account the requirements of Belarus Ministry of Health order No. 953 on digital operating rooms.
Competency: handling digital patient data in line with the requirements.
See it in the demonstration:
Privacy
Badge sign-in
Server tree
Education management and AI
5Management, control and documents
- «Vidicap check» and «Fleet check»: you can see in advance which functions each device supports and what equipment will be needed.
- Journals: who watched a broadcast, what events occurred on a device, which notifications the doctors on duty received.
- Built-in training introduces a new staff member to the main functions of Vidicap. The WHO checklist (add-on module): the three pauses of safe surgery, confirmed with a badge.
- Protocols under order No. 254 in Vidicap Connect: monitor data goes into the anaesthesia record and the daily intensive care sheet automatically, with no manual transcription.
- Buildings, the hospital and the cloud are joined in a tree of servers: updates and licences are distributed automatically.
See it in the demonstration:
Vidicap check
Hospital server
Vidicap Connect
7Opportunities for participants
Medical universities and departments
What you will test: broadcasts and conferences for students and residents, an AI direction on your subject, data annotation by the department's doctors and experts.
What is required: a subject, the department's experts, an organization account on vidicap.cloud.
Medical colleges
What you will test: broadcasts from partner hospitals' operating rooms, the Vidicap emulator and the test environment with a virtual unit for classes.
What is required: computers with an up-to-date browser and accounts on vidicap.cloud.
Hospitals and national research and practical centres
What you will test: broadcasts from the operating room, AI at doctors' workplaces, Vidicap Connect in the test environment; then Vidicap Touch in the operating room or an already installed Touch 1.0, the hospital server and the examination archive.
What is required: an operating room or unit and the clinic's permission to use de-identified material for AI training.
Specialist doctors
What you will test: annotation and expert review of images in the browser, by invitation.
What is required: your own videos or images that your clinic has approved for use.
IT departments
What you will test: the test environment: exchange with the HIS (HL7, API, webhooks) using a model of the HIS, the tree of servers, the training computer software.
What is required: a server in the hospital network and, for AI training, a computer with an NVIDIA 12–24 GB graphics card (Vidicap equipment can be used instead).
Platform
- An organization account on vidicap.cloud or your own hospital server
- An AI direction and invitations for doctors
- A training computer with an NVIDIA 12–24 GB card, your own or provided by Vidicap
Equipment
- Vidicap Touch in the operating room or an already installed Touch 1.0
- Up to 4 video inputs: endoscope, ultrasound, room camera
- A patient monitor with HL7 support
- AI on Touch 1.0 runs on the device itself if its performance allows, otherwise via Vidicap AI Server
AI training requires no operating room; a browser is enough.
All features described are available, except those marked «in development». The AI models are research versions, not for diagnosis.
📝 Apply for testing
Tell us whom you represent and what you plan to test. We will review the request and contact you. No registration is required.
You can also write to us: info@vidicap.cloud
