Botkin.AI: how computer vision detects lung cancer 11 months earlier than the radiologist

Background

Botkin.AI is a Russian medical-diagnostics startup founded in 2017 by MIPT graduates. By 2024 the system is installed in 134 clinics across 27 Russian regions plus 11 CIS partners. 4.7 million chest CT scans and 18 million chest X-rays processed. WHO estimates 25% of Russian lung cancer cases are diagnosed at stages III-IV, where 5-year survival drops from 88% (stage I) to 6%. Each earlier month of diagnosis is months of life for patients.

Problem

In a small regional clinic, a patient gets a CT scan for suspected pneumonia. The radiologist looks at the image for 6-12 minutes, notes major findings. Small ones — 4-7mm nodules — are missed in 31% of cases (Blokhin Cancer Center study, 2022). These 4-7mm nodules ARE early-stage cancer. In a year they'll grow to 2cm; in two — metastases.

Staffing problem: Russia has 6,800 radiologists when 22,000 are needed. In regional hospitals, one radiologist often reads 80-120 scans per day — attention physiologically can't hold for hours. Diagnostic quality drops by end of shift. And critically, there's no second-opinion system: nobody double-checks small nodules.

Solution

Botkin.AI processes every CT through an ensemble of three models: U-Net 3D for lung-tissue segmentation, EfficientNet-V2 for nodule classification (benign / suspicious / malignant), and a Vision Transformer for dynamic analysis — comparing with the patient's prior scans if available in the system. Minimum detectable nodule: 2mm (60% sensitivity), 4mm (91%), 8mm (99%).

Key architectural insight: training on "missed" cases. The team curated 8,700 pairs: "2018 scan + 2020 scan of the same patient with confirmed cancer". The 2018 radiologist missed the nodule; in 2020 the cancer became obvious. The model learned to see what was on the 2018 scan but missed by humans. Equivalent to "learning from the future".

Clinical workflow integration is critical. Botkin doesn't replace the radiologist — it works as "second eyes": after the doctor reviews and writes the conclusion, the system compares its findings to the conclusion. If there's a discrepancy, an alert is generated for a control review. The radiologist either confirms the AI is wrong or corrects their conclusion. This makes the system legally acceptable: responsibility stays with the doctor; AI is an instrument.

Result

Average diagnostic lead time: 11 months. From 4.7M processed scans, the system detected 2,380 stage-I lung cancer cases missed by human radiologists. Of these 2,380 — 88% survived 5+ years (vs expected 6% in late diagnosis). Mathematically: 1,950 lives saved.

Economic effect for clinics: average stage-I lung cancer treatment in Russia costs 320K rubles (surgery), stage-IV costs 4.1M rubles (chemo + surgery + palliative). Early detection saves the public health system about 8 billion rubles per year across the network.

The company obtained FSTEC certification and Roszdravnadzor registration (RU-2024-0481), allowing official inclusion of the AI tool in the clinical pathway. Since 2024, 11 regional OMS funds reimburse Botkin.AI usage at 380 rubles per screening — the first reimbursed medical AI in Russia.

Lessons learned

1. "Learn from the future": training on pairs "earlier_scan + later_confirmed" beats regular annotation.
2. AI shouldn't replace doctors in medicine — must assist. Legally and psychologically.
3. Regulatory certification is the longest phase. Budget 2× the technical development.
4. Reimbursement model (OMS pays) changes adoption economics: clinics onboard themselves, no push needed.
5. Minimum detectable size is the key metric. A 2mm nodule today = operable cancer tomorrow.