Rough set theory in diagnostics: mathematics instead of intuition

# How Mathematics Optimizes Diagnostics: Applying Rough Set Theory in Medicine

Doctors daily face the task of selecting the minimally necessary tests for making a diagnosis. Instead of intuitive decisions, rough set theory provides a formal method for identifying critically important symptoms. This approach, based on Zdzisław Pawlak's work from 1982, allows reducing redundant examinations without sacrificing diagnostic accuracy. Implementation requires only a binary data matrix and basic set theory operations—no neural networks or massive datasets needed.

Basics of Rough Set Theory

Rough set theory addresses uncertainty and incomplete data without relying on probabilistic models or fuzzy logic. The key principle is acknowledging the existence of objects that are indistinguishable given a specific set of attributes. In a medical context, this means patients with the same symptoms should receive the same diagnosis.

Indiscernibility Relation and Set Approximations

For any subset of symptoms, an indiscernibility relation is formed: two patients are considered equivalent if their symptom profiles match within the selected set. This partitions the data into equivalence classes. The lower approximation of a diagnosis set includes patients definitively belonging to that category, while the upper approximation includes those potentially in it. The difference between them defines the uncertainty zone.

Consistency Criteria for Decision Tables

A data table is consistent if there are no pairs of patients with identical symptoms but different diagnoses. For example:

• Patient A: fever=1, cough=1 → Flu
• Patient B: fever=1, cough=1 → ARI

Such a table is inconsistent—new symptoms must be added to distinguish the cases. Consistency is a prerequisite for finding reducts.

Reducts: Minimal Sets of Diagnostic Attributes

Superreducts and Minimal Reducts

A superreduct is any set of symptoms that preserves the table's consistency. A reduct, however, is a minimal superreduct: removing any attribute from it breaks consistency. A single table can have multiple reducts, reflecting alternative diagnostic paths.

Attribute Significance Vector

To assess each symptom's importance, a significance vector is calculated:

function calculate_significance(reducts, total_features)
    significance = zeros(total_features)
    for reduct in reducts
        for feature in reduct
            significance[feature] += 1
        end
    end
    return significance / length(reducts)
end

Each vector element shows the proportion of reducts including that attribute. A value of 1.0 means the attribute is essential for all diagnoses, 0.5 means it's replaceable, and near 0 means it's redundant.

Practical Implementation in Julia

Problem Setup

Source data: a table of 12 patients with 7 symptoms (fever, cough, shortness of breath, sore throat, weakness, runny nose, headache) and 5 diagnoses. Goal: find all minimal symptom sets that uniquely determine the diagnosis.

Exhaustive Search Algorithm

• Generate all possible attribute subsets (2^7 - 2 = 126 combinations)
• For each combination:

- Form the submatrix

- Check consistency by comparing unique rows

- Save consistent sets as superreducts

• Extract minimal reducts from superreducts
• Calculate the significance vector

Results Analysis Example

For the test table, 5 reducts were found. Key observations:

• Fever appears in all reducts (significance 1.0)
• Cough and sore throat have significance 0.8
• Runny nose and headache—0.4
• Shortness of breath—0.6

This means fever is a mandatory marker for all diagnoses, while other symptoms can substitute for each other depending on the clinical picture.

Approach Advantages

• Minimizing redundant tests—up to 40% reduction without losing accuracy
• Interpretability—clear rules instead of a "black box"
• Works with small datasets—effective on samples as small as 10-20 observations
• No hyperparameters—no need to tune thresholds or weights

Key Takeaways

• Rough set theory doesn't replace clinical expertise but structures it into formal rules
• The method applies beyond medicine—in industrial diagnostics and data analysis
• Source table consistency is critical—always verify data before analysis
• The significance vector helps identify both essential and replaceable attributes
• Implementation takes just 50-70 lines of code in Julia or Python

Limitations and Prospects

The method has exponential complexity O(2^n), limiting it to fewer than 20 attributes. For larger cases, heuristics or hybrid approaches with decision trees are used. Promising directions:

• Integration with ontologies to account for symptom hierarchies
• Combination with active learning for dynamic test selection
• Deployment on resource-constrained IoT devices

In industrial diagnostics, similar methods already cut equipment downtime by 15-20% through optimized maintenance checklists. In financial analytics, the approach identifies minimal indicator sets for default prediction.

Conclusion

Rough set theory shows how fundamental mathematics solves real-world problems. Its strength lies in balancing rigor with implementation simplicity. For medical system developers, the key takeaway: complex ML models aren't always needed. Sometimes it's enough to frame the problem right and apply the right mathematical tools. The code in this article can be adapted to any domain—from industrial sensors to user surveys—in a single workday.

— Editorial Team