IVF-DM
Scroll Company

How a lifebegins, andwho decides.

IVF-DM · decision support for IVF

Scroll slowly. Everything you will see is alive in code, one cell at a time.

Before.

Days −12 to 0 · follicles

For about two weeks, stimulation grows follicles, each holding one egg.

Decision: how much stimulation. IVF-DM estimates the ovarian response to a starting dose.

One egg.

Day 0 · retrieval

The eggs are collected. Only the mature ones can be fertilized.

Decision: when to trigger final maturation.

Millionstry.

Day 0 · fertilization

In the lab, sperm meet the egg. Or one sperm is injected directly.

Decision: IVF or ICSI, with the couple's history in view.

One cell.

Day 1 · zygote

Two pronuclei, one from each parent, drift toward each other.

Two.Four.Eight.

Days 2 to 3 · cleavage

The embryo divides without growing. The embryologist watches its rhythm.

Every division feeds back into the estimate.

Together.

Day 4 · morula

The cells press close until their borders almost disappear.

A choice.

Day 5 · blastocyst

A hollow ball: cells for the baby on one side, cells for the placenta around.

Decision: which embryo, and how many. IVF-DM lays out the estimated odds of each option.

Life.

Day 6 · hatching

It breaks out of its shell, ready to implant. From here, no model decides.

The model shows the odds. The doctor decides.

IVF-DM is decision support for reproductive specialists. At each step of the cycle it estimates the odds of every option, shows the factors behind the estimate, and shows what changes if the plan changes.

01 · Why

One in four.

Only one in four fresh embryos transferred ends in a birth. Most attempts fail, and each one costs the patient months.

Each cell is one fresh embryo transferred; a living cell ends in a birth. UK, 2023, patients using their own eggs. Source: HFEA, Fertility Treatment 2023: Trends and Figures (preliminary).

1

Most transfers fail.

75% of fresh embryos transferred in the UK in 2023 did not end in a birth. At ages 43 to 44, 95% did not.

2

Every attempt is costly.

52,400 patients went through more than 77,500 cycles in one year, about one and a half each. A cycle means weeks of hormone injections, an egg retrieval and a two-week wait, and 73% of cycles were not funded by the NHS.

3

Plans rest on a few factors.

Specialists usually plan a cycle around six or seven main factors. The clinic's own records hold far more, and nobody has time to analyse them at the moment of a decision.

IVF-DM puts the clinic's own evidence in front of the doctor, at the moment of each decision.

Sources: HFEA, Fertility Treatment 2023 (patients, cycles, birth rates, NHS funding). Six or seven factors: clinical experience of Dr. Kseniia Tikhaeva.

02 · Where it helps

Four moments.

A cycle turns on a few decisions. IVF-DM gives each one its own model.

Stimulation · days −12 to −2

How to stimulate

When to start, which drugs and in what combination, the starting dose, dose adjustments, add-on therapy.

Maturation · days −2 to 0

When to collect

The type of trigger for final egg maturation, and the day and time of retrieval.

Laboratory · days 0 to 5

How to fertilize

IVF or ICSI, the culture medium, special culture conditions.

Transfer · days 3 to 6

What to transfer

Transfer now or freeze, which embryo, and how many.

For each option: the estimated chance of success, of failure and of complications, with the uncertainty around each estimate.

03 · What the doctor sees

The screen.

Synthetic patient · cycle 2
age36AMH1.4AFC9BMI23

Cycle 1: 5 eggs at 150 IU

Decision now
  1. Starting dose
  2. Trigger and retrieval
  3. IVF or ICSI
  4. Which embryo
Expected mature eggs ≈ 6.8

middle 50%: 6 to 8

OHSS risk low
Why this estimate?
  • cycle 1 response+1.6
  • antral follicles+0.8
  • AMH−0.9
  • age−1.1

mature eggs vs. a typical patient

225 IU

The doctor picks the dose, not the model.

Concept illustration. The patient is synthetic and the values are illustrative, not output of a trained model. Drag the dose, or the pink cell on the rim of the lens, to see what changes.
1

Every option, side by side

Each dose with its expected result drawn as a range, so the overlap between options is visible.

2

The risks of each option

Complications such as OHSS are shown next to the expected benefit, not in a separate report.

3

Why the model thinks so

The factors that raise or lower the estimate for this patient, so the doctor can check the reasoning.

04 · How it works

Record to decision.

STEP 1

The record

The cycle record as the clinic already keeps it: age, ovarian reserve, history, protocol, laboratory data.

STEP 2

The models

Multi-class gradient boosting and Bayesian models, one set per decision point. The Bayesian part gives an honest range, not just one number.

STEP 3

The odds

Odds for each option, the factors that drive them, and how they shift when the plan changes.

STEP 4

The doctor decides.

Reads the odds, weighs them against the patient in front of them, and decides.

IVF-DM connects to the clinic's records system through an API, or works on its own with manual entry. The models are retrained as new de-identified outcomes arrive.

05 · Data

Only what's safe.

Recordsas the clinic keeps them
De-identificationidentifiers removed
Modelslearn from what passes
Step 1 · public register

HFEA anonymised register

Treatment cycles in the UK from 1991 to 2018, with patient identifiers removed, available to researchers on application. First models are trained and validated here.

Step 2 · partner clinic

A clinic in Israel

De-identified records from a partner clinic, under approval of its Helsinki (ethics) committee.

No identifiable data leaves the clinic.Models learn from de-identified records only.

06 · Team

Two.

A reproductive specialist and an engineer who builds AI systems.

Medicine

Dr. Kseniia Tikhaeva

MD, PhD (Candidate of Medical Sciences) · Founder and medical lead

  • Obstetrician-gynecologist, more than 12 years in medicine, more than 7 in IVF.
  • Former medical director of an IVF clinic.
  • Led the Reproductive Medicine research group at Volgograd State Medical University; associate professor of pathophysiology.
  • Co-led the development and rollout of a medical records system for IVF clinics (2016 to 2017).
  • Now based in Israel.
Selected publications and patent
  • Tikhaeva K.Yu. Efficacy of dosed transdermal 17β-estradiol gel in endometrial preparation for frozen-thawed embryo transfer. Problemy Reproduktsii (Russian Journal of Human Reproduction), 2016, 22(6): 84–88.
  • Tikhaeva K.Yu., Rogova L.N., Tkachenko L.V. Gelatinase activity in endometrial tissue of women with chronic endometritis and repeated implantation failure. Proceedings of the XXVI International Conference of the Russian Association of Human Reproduction, Moscow, 2016: 79–81.
  • Rogova L., Tikhaeva K., Tkachenko L. Role of gelatinase B and intraerythrocytic magnesium in the formation of chronic endometrial inflammation in rats. Medical News of North Caucasus, 2016, (2): 214–216.
  • Patent RU 2580986. Method for modelling experimental chronic endometrial inflammation in rats. Tikhaeva K.Yu., Rogova L.N., Tkachenko L.V. Published 10.04.2016.

Publications are in Russian.

Engineering

Konstantin Tikhaev

Information Systems Engineer · Co-founder and technology lead

  • Builds and runs AI systems in production, alone: A³, a real-time talking-avatar agent (US provisional patent), GoNoGo and Lifeline.
  • Runs an open, self-funded study of how AI models describe themselves: about 14,100 answers from 11 models, with a blind grader and hypotheses written before the run.
  • Founder and director of IS Multiproject, a construction supervision company in Russia, since 2014.
  • Seven years in retail operations: a group of 15 stores at Magnit, trade sectors across 15 hypermarkets at Leroy Merlin.
  • Engineer's degree in Information Systems and Technologies (five-year programme). Based in Haifa, Israel.
07 · Status

So far.

2016

Records system

Kseniia co-develops a medical records system for IVF clinics and sees how much data goes unused.

2020

IVF-DM designed

The decision points, the modelling approach and a three-stage research plan.

2026

Relaunch in Israel

Research prototype. This is where we are now.

08 · Contact

Let's talk.

Partner clinics, researchers and funders: we would like to hear from you.