Detecting Disease in Radiographs with Intuitive Confidence

Area of Science:

• Medical Imaging
• Computational Medicine
• Diagnostic Systems

Background:

• Traditional diagnostic methods often lack quantifiable confidence metrics.
• Existing computer-aided diagnosis (CAD) systems require intuitive and interpretable confidence measures.
• Chinese medicine concepts like Yin and Yang offer a framework for understanding balance and imbalance.

Purpose of the Study:

• To introduce a novel confidence measure for computer-aided diagnosis based on sine/cosine values.
• To demonstrate the interpretability of this confidence measure for physicians.
• To validate the application of this confidence measure in disease classification and medical image analysis.

Main Methods:

• Representing diagnostic confidence as points on a unit circle, using sine/cosine values of angles.
• Drawing parallels between sine/cosine values and the Yin/Yang balance in Chinese medicine.
• Applying theoretical results from neural signal processing regarding sine/cosine relationships in learning.
• Utilizing the proposed confidence values to identify tuberculosis in frontal chest X-rays.

Main Results:

• Sine/cosine values provide an intuitive representation of diagnostic confidence, linked to physiological balance.
• An angle of 45° (equal sine and cosine) signifies equilibrium, representing a non-dual state for classification.
• The method successfully highlighted tuberculosis manifestations in frontal chest X-rays.
• The proposed confidence measure is readily understandable by medical practitioners.

Conclusions:

• Sine/cosine confidence offers an intuitive and theoretically grounded approach for computer-aided diagnosis.
• This method enhances the interpretability of diagnostic systems, bridging traditional concepts with modern computation.
• The application in tuberculosis detection demonstrates the practical utility of this novel confidence measure in medical imaging.