Endpoint Specification

This section of the documentation provides an overview of the data schemas for the medical device's API endpoints. Instead of detailing every endpoint and its resources here, we embed the automatically generated API documentation within an iframe for a more dynamic and up-to-date experience.

Why Use an Iframe?

Embedding the API documentation offers several benefits:

• Real-Time Updates: The iframe reflects the latest changes and additions to the API codebase, ensuring you always have the most accurate information.
• Consistency: By linking directly to the source documentation, we eliminate the risk of discrepancies or outdated content.
• No Redundancy: Avoiding duplication keeps this documentation concise, focused, and perfectly aligned with the data models in the code.

Use the interactive interface provided in the iframe to browse available endpoints, review request and response structures, and explore examples. Any changes to our API are instantly reflected, so you can trust that you are always viewing the most current and reliable version of our endpoint specifications.

If you encounter issues accessing or interpreting the content, refer to the troubleshooting section or contact our support team for assistance.

Explanation

The device operates in a straightforward manner: you send images, and you receive a DiagnosticReport as defined in HL7's FHIR® specifications. The following chart illustrates the basic workflow:

Diagnostic Support

While the term diagnosis is commonly used to describe the device's output, it is important to note that the device provides a interpretative distribution representation of possible International Classification of Diseases (ICD) categories that might be represented in the pixels content of the image. Healthcare practitioners and organizations may use this output to inform a diagnosis, but the device itself does not generate a diagnosis. This distinction is clearly indicated in the output, which adheres to the FHIR standard and is labeled as a DiagnosticReport with a status of preliminary.

Severity Measure

Similarly, while the term severity measure is often used, the device outputs a quantifiable data on the intensity, count and extent of clinical signs such as erythema, desquamation, and induration, among others. This output helps healthcare practitioners assess the degree of a patient's condition, but it is not the severity itself. This is also indicated in the FHIR-compliant DiagnosticReport with a preliminary status.

Clinical Indicators

Clinical indicators are derived from the diagnostic support output of the device, which is a interpretative distribution representation of possible International Classification of Diseases (ICD) categories that might be represented in the pixels content of the image. This output is a probability distribution where each ICD-11 category is assigned a probability value between 0 and 1, with the total sum equaling 1.

Condition Confirmation

The device's output covers a wide range of ICD-11 categories, except for one: Non-specific lesion. This category is used when no condition is detected in the image. The likelihood of an image containing any condition (hasCondition) can be calculated as:

$$p_{condition}= 1 - p_{nsl}$$

Where $p_{nsl}$ is the probability assigned to the Non-specific lesion category. Since the total probability equals 1, subtracting $p_{nsl}$ gives the probability of a condition being present.

Weighted Sum Findings

Several clinical indicators (pigmentedLesion, urgentReferral, highPriorityReferral, and malignancy) are calculated using a weighted sum of the device's output. Each indicator uses binary weights (0 or 1) to determine which ICD-11 categories contribute to its value.

• Pigmented Lesion (pigmentedLesion): Positive weights for categories corresponding to pigmented lesions.
• Urgent Referral (urgentReferral): Positive weights for categories requiring referral within 0-48 hours.
• High Priority Referral (highPriorityReferral): Positive weights for categories requiring referral within 7-15 days.
• Malignancy (malignancy): Positive weights for categories related to malignancy (e.g., skin cancer).

The value of each finding ($f$) is computed as:

$$f = \sum_{i=1}^{N} w_i \cdot p_i$$

Where $N$ is the total number of predicted ICD-11 categories, and $w_i$ and $p_i$ are the weight and probability of the $i$-th category.

ICD-11 Categories Related to Malignancy

ICD-11 Code	Category	Malignant
2C30.3	Acral lentiginous melanoma (primary)	1
XH89V4#2C33	Adnexal carcinoma	1
2E63.00#XH9KY6#XH9NL4	Amelanotic malignant melanoma	1
XH6264#2B56.1	Angiosarcoma	1
XH2615#2C32	Basal cell carcinoma	1
XH63D2#2D41	Carcinoma	1
2C30	Cutaneous melanoma	1
2E08	Cutaneous metastasis	1
XH1951	Cutaneous T-cell lymphoma	1
2B53.Y#XH4QZ8	Dermatofibrosarcoma protuberans	1
XH36A5	Kaposi sarcoma	1
XH81N8	Merkel cell carcinoma	1
2E08	Metastatic malignant neoplasm involving skin	1
1G60.0	Mycetoma	1
2B01#XH8R56	Mycosis fungoides	1
-	Pleomorphic T cell lymphoma	1
XH0945#2C31	Squamous cell carcinoma	1
EB13	Stevens Johnson syndrome or toxic epidermal necrolysis	1
XH5NL6#2C32.2	Superficial basal cell carcinoma	1
1A6Z	Syphilis	1

Performance Indicators

Performance indicators provide insights into the device's skin disease recognition performance. These indicators are derived from the diagnostic support output, which is a interpretative distribution representation of possible International Classification of Diseases (ICD) categories that might be represented in the pixels content of the image.

Top-K Sensitivity and Specificity

To evaluate the device's performance for each ICD-11 category, we compute top-K sensitivity and specificity metrics on a hold-out test set. These metrics are defined as:

• Sensitivity (true positive rate): Probability of a positive output given a positive test case.
• Specificity (true negative rate): Probability of a negative output given a negative test case.

$$Sensitivity=\frac{TP}{TP + FN}$$ $$Specificity=\frac{TN}{TN + FP}$$

For each ICD-11 category $C$, we:

1. Treat $C$ as positive (1) if it matches the ground truth label, and negative (0) otherwise.
2. Consider $C$ as positive (1) if it appears in the top-K predictions, and negative (0) otherwise.

As the predictions and ground truth labels have been converted to a binary case (0/1), we can compute sensitivity and specificity using true positives, false negatives, false positives, and true negatives:

	Positive output	Negative output
Positive label	✔️ True positive (TP)	❌ False negative (FN)
Negative label	❌ False positive (FP)	✔️ True negative (TN)

We compute sensitivity and specificity for several values of $K$ (1,3, 5), resulting in the top-1, top-3, and top-5 sensitivity and specificity performance indicators.

Entropy

Entropy estimates the uncertainty of the device's output. Normalized entropy ($H$) is calculated as:

$$H=-\frac{\sum_{i=1}^{N}p_i \cdot \ln(p_i)}{\ln(N)}$$

Where $\ln$ is the natural logarithm, $N$ is the total number of ICD-11 categories, and $p_i$ is the probability of the $i$-th category.

Low entropy indicates high confidence, while high entropy suggests low confidence.