The Development and Performance of a Machine Learning Based Mobile Platform for Visually Determining the Etiology of Penile Pathology (2403.08417v1)

Abstract: Machine-learning algorithms can facilitate low-cost, user-guided visual diagnostic platforms for addressing disparities in access to sexual health services. We developed a clinical image dataset using original and augmented images for five penile diseases: herpes eruption, syphilitic chancres, penile candidiasis, penile cancer, and genital warts. We used a U-net architecture model for semantic pixel segmentation into background or subject image, the Inception-ResNet version 2 neural architecture to classify each pixel as diseased or non-diseased, and a salience map using GradCAM++. We trained the model on a random 91% sample of the image database using 150 epochs per image, and evaluated the model on the remaining 9% of images, assessing recall (or sensitivity), precision, specificity, and F1-score (accuracy). Of the 239 images in the validation dataset, 45 (18.8%) were of genital warts, 43 (18.0%) were of HSV infection, 29 (12.1%) were of penile cancer, 40 (16.7%) were of penile candidiasis, 37 (15.5%) were of syphilitic chancres, and 45 (18.8%) were of non-diseased penises. The overall accuracy of the model for correctly classifying the diseased image was 0.944. Between July 1st and October 1st 2023, there were 2,640 unique users of the mobile platform. Among a random sample of submissions (n=437), 271 (62.0%) were from the United States, 64 (14.6%) from Singapore, 41 (9.4%) from Candia, 40 (9.2%) from the United Kingdom, and 21 (4.8%) from Vietnam. The majority (n=277 [63.4%]) were between 18 and 30 years old. We report on the development of a machine-learning model for classifying five penile diseases, which demonstrated excellent performance on a validation dataset. That model is currently in use globally and has the potential to improve access to diagnostic services for penile diseases.

• The paper presents a machine-learning framework for rapid, image-based diagnosis of penile pathologies using semantic segmentation and pixel classification.
• It employs a U-net architecture for segmentation and Inception-ResNet v2 optimized by Adam, achieving 94.4% overall accuracy.
• The mobile platform showed significant global user engagement, indicating its potential to improve diagnostics in resource-limited settings.

Development and Evaluation of a Machine-Learning Mobile Platform for Penile Pathology Diagnosis

The paper presents the development and assessment of a machine-learning-based mobile platform aimed at diagnosing various penile pathologies through visual recognition. This paper addresses a significant barrier in sexual health services by leveraging machine-learning techniques to provide immediate and accessible diagnostic support for penile diseases, which include herpes eruptions, syphilitic chancres, penile candidiasis, penile cancer, and genital warts. This innovative approach offers a promising solution in contexts where access to sexual health diagnostics is limited by resource availability and social stigmas.

Methodology

The research introduces a sophisticated machine-learning framework that employs several methodological advances. A dataset of clinical images was curated from a combination of physician contributions and publicly available images, supplemented by innovative augmentation strategies. The systematic use of U-net architecture for semantic segmentation allowed for efficient delineation of diseased regions from background noise. Additionally, the Inception-ResNet version 2 architecture facilitated precise pixel classification, optimized through the Adam optimizer, enhancing the identification accuracy of penile pathologies. Salience mapping using GradCAM++ further provided visual explanations of the decision-making process, a critical feature for validating model predictions.

The model was trained using 91% of the dataset, comprising over 2,000 images, both original and augmented, across 150 epochs. Performance assessment was conducted on a validation dataset of 239 images, utilizing metrics such as recall, precision, specificity, and F1-score.

Results

The machine-learning model demonstrated robust performance, achieving an overall accuracy of 0.944 in classifying diseased images. Sensitivity and specificity analyses revealed that the model performed best in discerning non-diseased cases, followed by high efficacy in identifying genital warts, herpes eruption, and penile candidiasis. Notably, precision rates were above 90% for most disease categories. However, the model's recall was noted to be lowest for penile cancer, suggesting the need for further dataset expansion and algorithm refinement to address current limitations.

The mobile application utility was validated with 2,640 unique users between July and October 2023, showcasing a significant engagement predominantly from the United States, Singapore, and several other countries. This highlights the tool's expansive reach and potential impact in improving global sexual health accessibility.

Implications and Future Directions

The effectiveness of the machine-learning model in identifying penile pathologies signifies a substantial step forward in digital health technologies aimed at mitigating healthcare disparities. The potential for early-stage detection and intervention facilitated by this mobile platform could have considerable implications, reducing disease transmission and improving treatment outcomes.

Future research directions could involve expanding the model to encompass additional disease categories such as lichen sclerosis or vulvovaginal disorders. Moreover, comprehensive validation against gold-standard diagnostic procedures, alongside longitudinal impact evaluations on user health outcomes, are pivotal for enhancing credibility and utility. Additionally, integration with patient education and treatment linkage services could bolster the platform's overall effectiveness in a real-world context.

Conclusion

This investigation illustrates the capability of machine learning to revolutionize sexual health diagnostics, offering a scalable, user-friendly solution to pervasive access issues. While further model optimization and validation are necessary, the initial results affirm the platform's promising role in transforming health service delivery within resource-constrained settings.