You might be already familiar with skin checking apps, we too have discussed them quite extensively recently. Skin checking applications allow users to take pictures of their suspicious skin lesions, upload these pictures to a server, the images are first evaluated by an A.I. algorithm and the results will be later validated by a dermatologist.
However, this time there is something new under the sun: AIP Labs and Semmelweis University (Hungary’s leading medical university) developed a skin checking application that offers several features we have not seen before. It was launched to work integrated with the national healthcare system and offer full-scope dermatology services including diagnostics, treatment plans and cloud-based drug subscriptions. According to currently available data, processing speeds have improved by 5-10 times compared to the traditional care model.
In this analysis, we called these apps an emerging trend, and still stand by this statement. If you are looking for a signature digital health solution, these are great examples, as they:
- address an existing demand
- rely on already existing infrastructure from the patients’ side (eg: their mobile phones)
- provide easy and cheap access to patients who otherwise might have difficulties with scheduling dermatology appointments for any reason, like physical distance, insurance problems, anxiety and so on
- allow fast and reliable evaluation of skin lesions, cutting down waiting times
- eliminate unnecessary medical visits, saving time for both patient and clinician
- assist dermatologists, who can check significantly more cases in a given amount of time
We could even conclude that if such a system provides follow-up and access to doctors and treatments should the need arise, these are close to the optimal setup – they filter out non-existing cases and let dermatologists focus on the real issues.
Unique model: launching a skin checking app integrated with the national healthcare system
While these apps are not exactly new, this is an exciting collaboration between the healthcare system, the country’s leading medical university and the developer of the A.I.-based solution, called AIP Derm.
The uniqueness of the model is that it offers some features we have not seen before, like:
- Free access to all patients with social security during the test period
- Diagnosing not just suspicious skin lesions for cancer risk, but hundreds of dermatology conditions
- Offering drug subscriptions through the country’s electronic medical system
In other words, Hungarian patients with any skin problem could turn to the app, upload their pictures, answer some questions, and receive a diagnosis in just a few days.
The model has four end scenarios
- No treatment is necessary
- The dermatologist suggests using over-the-counter medication
- The dermatologist suggests using subscription-only medication by providing an e-subscription via the central online medical system, allowing the patient to pick up the drug at the nearest pharmacy
- Suggests a personal appointment for the treatment
Covering some 95% of skin diseases
AIP Derm currently detects 700 pathologies, covering some 95% of skin diseases. The submitted images and the diagnosis created by the A.I. are always reviewed by a doctor. Once an image comes in, the system pre-evaluates it, tells the dermatologist team which pathologies it believes are likely to be involved, and then sorts them by chronological order or severity.
According to data collected in the first few months, doctors choose one of the first three options 90% of the time. Doctors see the cases triaged by urgency. The A.I. algorithm offers 3 possible diagnoses for dermoscopic images and 5 for macroscopic images. If a dermatologist can’t address a case within 20 seconds, they can ask the patient to submit additional photos or suggest an in-person consultation.
Over 80% of skin cancer cases were detected early
In total, 18.114 cases were submitted by patients during the test period between March and June of 2022. The three most common skin conditions were Common moles, Seborrheic Keratosis and Contact Dermatitis.
Just over six hundred cases of skin cancer were diagnosed during this period. Some were surprisingly advanced. The largest tumour detected was late-stage cancer, over 3 cm wide. But in general, most cases (over 80%) were in the early stages. This shows that A.I. is a very powerful accelerator for preventive healthcare. The detected cases were diagnosed under five categories: basal-cell-carcinoma(57%), melanoma (29%), squamous-cell-carcinoma (5%), squamous-cell-carcinoma-in-situ (4%) and invasive-cutaneous-squamous-cell-carcinoma (3%).
Processing speeds have improved by 5-10 times
In the pilot 10 doctors of the Semmelweis University Clinic of Dermatology, Venereology and Dermatooncology worked to evaluate the uploaded cases, they were able to go through 40-80 images per hour. These clinicians provided the treatment suggestions, issued the cloud-based drug subscriptions and suggested in-person consultation if necessary.
Unlike other skin-checking apps, this model was designed to work as a remote-only service – as opposed to being a telehealth solution with online consultation and follow-up. The reason for going this route was to examine whether it allows significant capacity increase. If doctors need to spend time with online consultations, this benefit is mostly lost. Participating clinicians’ reported that processing speeds have improved by 5-10 times compared to the traditional care model.
A variety of deep-learning models
AIP Derm has a variety of deep-learning algorithms that form both the visual engine and the Q&A engine, whose design and performance were optimised from the Hungary pilot and other European initiatives. It is built on a convolutional neural network, and it’s currently being replaced by A.I. transformator technology. Validation has been done through the Semmelweis University Dermatology Clinic and private dermatologists over a period of 18 months.
The model was trained on 2 million images, developed entirely on in-house infrastructure. The database is made up of skin images that were contributed by clinics and hospitals globally.
How is the algorithm trained?
It is a very complex process with individually sophisticated components that require careful clinical and evidence-based feedback to verify certain development hypotheses. In simple terms, developers are constantly collecting data as the platform is being used, which is the “raw” image and profile/Q&A data collected with each case, which allows them to build the internal training databases used for algorithm development, while also validating its performance on real-world data and use.
AIP uses several techniques when training new deep-learning models. To address class imbalances (when more examples exist for one class than the other) in the training data, additional weights are applied to mistakes made on under-represented classes, which encourages the model to better detect cases with low prevalence in the dataset.
Additional paid services after the test period
The business model is geared towards addressing one of the fundamental pain points in healthcare: too many patients and too few clinicians. AIP will serve as a one-stop shop health care delivery system over the next several years in the Hungarian market.
Accelerated and at-home digital diagnostics will be available in the last quarter of the year nationwide as an additional service on top of social security benefits for a fee – at a fraction of the market price. Patients will continue to receive a diagnosis, with an outpatient form, and a digital prescription that can be redeemed at any pharmacy.
The post How An A.I.-Based Skin Checking App Can Work With A National Healthcare System appeared first on The Medical Futurist.