Subscribe

Artificial Intelligence for Detecting Keratoconus: What the Latest Evidence Says and Why It Matters for India

Artificial Intelligence for Detecting Keratoconus: What the Latest Evidence Says and Why It Matters for India

Why Keratoconus Matters

Keratoconus is a progressive eye disease that usually starts between the ages of 10 and 40. The cornea gradually weakens and thins, bulging into a cone-like shape. Early on, glasses or contact lenses may work, but as the disease advances, vision worsens, sometimes requiring corneal transplantation.

The condition is especially relevant in India, where studies suggest keratoconus is far more common and often more severe compared to Western populations. Factors like hot climate, high rates of eye rubbing, allergic eye disease, and genetic predisposition contribute to this. Detecting keratoconus early can make the difference between preserving vision and facing lifelong visual disability.

What Did the Cochrane Review Look At?

The 2023 Cochrane Review by Vandevenne and colleagues evaluated how well artificial intelligence (AI) tools perform in diagnosing keratoconus using corneal imaging (topography and tomography).

  • 63 studies (over 56,000 eyes/images) were included, published between 1994 and 2022.
  • AI tools were compared against expert cornea specialists’ interpretation of scans.
  • The review looked at both manifest keratoconus (where signs are visible) and subclinical keratoconus (early or hidden disease that specialists may miss).

Key Findings

  1. For manifest keratoconus
    • AI performed almost perfectly.
    • Sensitivity: ~99% (ability to detect disease when present).
    • Specificity: ~98% (ability to rule it out when absent) .
  2. For subclinical keratoconus
    • Accuracy was still high but less consistent.
    • Sensitivity: ~90%.
    • Specificity: ~96% .
    • In simple terms: AI is good at flagging early disease, but it may miss some subtle cases.
  3. Limitations
    • Many studies had bias (case-control designs using pre-diagnosed patients rather than real-world screening).
    • Definitions of “subclinical” varied.
    • Evidence certainty was rated low to very low .

Authors’ Conclusions

  • AI appears to be a promising triage tool.
  • It is highly accurate for established keratoconus, less so for early detection.
  • More standardized, high-quality studies are needed before routine use.
  • A missed diagnosis can be dangerous—especially if a patient undergoes refractive surgery, which can worsen undetected keratoconus into ectasia.

What This Means for India

India faces a unique challenge:

  • High prevalence: Keratoconus is estimated to be several times more common than in Western countries.
  • Younger onset and faster progression: Many Indian patients present in their teens with advanced disease.
  • Limited corneal specialists: Most patients first visit optometrists or general ophthalmologists, who may not always have access to advanced tomography or the expertise to interpret it.

This is where AI could be transformative:

  • Early triage at primary care or optical centers: AI could flag suspicious corneal scans so patients are referred to tertiary centers before vision loss sets in.
  • Refractive surgery screening: Given India’s booming LASIK/SMILE market, AI can reduce the risk of missing subclinical keratoconus in surgery candidates.
  • Reducing burden on specialists: Automating the initial step of corneal scan interpretation could save valuable time in high-volume practices.

Challenges to Adoption in India

  • Evidence gap: Most AI studies so far are not from Indian populations, where corneal shape patterns may differ.
  • Infrastructure: AI tools need to be integrated with devices already in use in smaller clinics.
  • Validation: Local clinical validation is essential before deployment.
  • Ethical concerns: Over-reliance on AI without human oversight could lead to errors.

The Way Forward

For India, the potential of AI in keratoconus detection is significant but must be approached cautiously:

  • Research: Indian-specific datasets should be used to train and validate AI algorithms.
  • Integration: AI should complement, not replace, human judgment.
  • Policy: Government and professional bodies could support pilot programs in high-prevalence regions.
  • Awareness: Optometrists and general ophthalmologists should be trained to use AI-based triage responsibly.

Bottom Line

AI shows remarkable promise in detecting keratoconus, particularly when the disease is already visible. For early and subclinical disease, it can help, but accuracy is not yet guaranteed.

In India—where keratoconus is common, aggressive, and often detected late—AI could play a vital role in early screening and referral, provided it is validated locally and used alongside, not instead of, trained specialists.

📖 Reference:

Vandevenne MMS, Favuzza E, Veta M, Lucenteforte E, Berendschot TTJM, Mencucci R, Nuijts RMMA, Virgili G, Dickman MM. Artificial intelligence for detecting keratoconus. Cochrane Database of Systematic Reviews 2023, Issue 11. Art. No.: CD014911. DOI: 10.1002/14651858.CD014911.pub2.