A deep learning system for differential diagnosis of skin diseases

Abstract: Skin conditions affect an estimated 1.9 billion people worldwide. A shortage of dermatologists causes long wait times and leads patients to seek dermatologic care from general practitioners. However, the diagnostic accuracy of general practitioners has been reported to be only 0.24-0.70 (compared to 0.77-0.96 for dermatologists), resulting in referral errors, delays in care, and errors in diagnosis and treatment. In this paper, we developed a deep learning system (DLS) to provide a differential diagnosis of skin conditions for clinical cases (skin photographs and associated medical histories). The DLS distinguishes between 26 skin conditions that represent roughly 80% of the volume of skin conditions seen in primary care. The DLS was developed and validated using de-identified cases from a teledermatology practice serving 17 clinical sites via a temporal split: the first 14,021 cases for development and the last 3,756 cases for validation. On the validation set, where a panel of three board-certified dermatologists defined the reference standard for every case, the DLS achieved 0.71 and 0.93 top-1 and top-3 accuracies respectively. For a random subset of the validation set (n=963 cases), 18 clinicians reviewed the cases for comparison. On this subset, the DLS achieved a 0.67 top-1 accuracy, non-inferior to board-certified dermatologists (0.63, p<0.001), and higher than primary care physicians (PCPs, 0.45) and nurse practitioners (NPs, 0.41). The top-3 accuracy showed a similar trend: 0.90 DLS, 0.75 dermatologists, 0.60 PCPs, and 0.55 NPs. These results highlight the potential of the DLS to augment general practitioners to accurately diagnose skin conditions by suggesting differential diagnoses that may not have been considered. Future work will be needed to prospectively assess the clinical impact of using this tool in actual clinical workflows.

• The paper presents a deep learning system that attains 71% top-1 and 93% top-3 accuracy in diagnosing 26 common skin diseases.
• It employs multiple CNN modules and shallow metadata networks, mirroring clinical teledermatology practices.
• The system outperforms primary care physicians and nurse practitioners, offering decision support to enhance triage and treatment.

A Deep Learning System for Differential Diagnosis of Skin Diseases

The paper presents a deep learning system (DLS) designed to facilitate differential diagnosis of common skin diseases within a clinical setting. This system is an important development given the prevalence and strain skin conditions place on healthcare resources and the longstanding shortage of dermatologists. The proposed DLS targets 26 prevalent skin conditions, representing approximately 80% of primary care dermatological consultations.

Development and Validation

The DLS was developed using de-identified clinical data from a teledermatology service. This dataset was split temporally for development (14,021 cases) and validation (3,756 cases). Each clinical case comprised skin photographs and medical history data. A panel of three dermatologists established the reference standard for validation cases.

Performance Metrics

The DLS achieved a top-1 accuracy of 0.71 and a top-3 accuracy of 0.93 on the validation set. In comparison trials with 18 clinicians (including dermatologists, primary care physicians (PCPs), and nurse practitioners (NPs)), DLS demonstrated a top-1 accuracy of 0.67, which was statistically non-inferior to the board-certified dermatologists' performance (0.63) while exceeding PCPs (0.45) and NPs (0.41). The top-3 accuracy comparison also favored the DLS (0.90) over dermatologists (0.75), PCPs (0.60), and NPs (0.55).

Technical Implementation

The system integrates multiple convolutional neural network (CNN) modules for image analysis and a shallow network for processing metadata, such as demographic and medical history information. This multi-input approach aligns with dermatologists' practices in teledermatology, where multiple images and patient history inform diagnostic decisions.

Implications and Future Directions

The results indicate that the DLS can significantly enhance diagnostic accuracy and support non-specialist clinicians by suggesting possible differential diagnoses, which could streamline triage processes and reduce unnecessary delays in treatment initiation. Importantly, the DLS supports clinicians in challenging diagnostic scenarios where diseases present with similar symptoms.

The system's comprehensive differential diagnosis is particularly advantageous in dermatology, aiding decision-making processes in primary care settings where rapid and accurate diagnosis is critical. Moreover, the DLS can potentially optimize dermatologists' efforts in teledermatology by automating initial assessments, allowing specialists to concentrate on complex cases.

Considerations for Broader Impact

The paper provides a substantial step towards integrating AI-driven solutions in clinical workflows, highlighting the need for decision-support tools that address diagnostic ambiguities in dermatology. The potential of improving healthcare access by augmenting non-specialist capabilities is significant, especially for under-resourced areas with limited specialist availability.

Looking forward, expanding the system's validation across more diverse datasets and geographical regions will be crucial to ensure generalizability and efficacy in various clinical environments.

Overall, this study demonstrates a promising application of AI in healthcare, emphasizing the importance of integrating visual and contextual patient data for enhancing diagnostic processes in teledermatology and beyond.