As one enjoys the summer sun, one might notice new moles on the body due to sun exposure. Medically, these moles, patches, or blemishes are classified as lesions. Over the past few years though these lesions have been increasingly presenting themselves as deadly malignant melanoma.
To identify which mole might be cancerous, one needs to undergo a mole scanning procedure. This involves a dermatologist placing a dermoscope on top of a mole to enlarge it and record a picture. The doctor then does the procedure all over the patient’s body. These time-consuming scans need the availability of a specialist and are seldom conducted, resulting in neglect and excessive healthcare expenses.
As Europe’s sixth most frequent cancer, about 144,000 new cases are detected each year across the continent. The survival rate is 95% if the melanoma is detected while it is less than 1mm deep on the skin surface. Melanoma becomes metastatic if found later, and just 23% of individuals survive for more than 5 years. We don’t know how to treat it, but early detection can save a life.
Rafael Garcia leads iTobos (short for Intelligent Total Body Scanner) which is a pan-European project which aims at automatically scanning the human body for cancerous lesions using AI.
How does this work?
Garcia explains the procedure as “a patient will enter a room and lay on a bed. The scanner above them will move to acquire images of their body surface, and that’s it! The scanner will look like a series of cameras in an arch that moves.”
To collect data on a balanced mix of healthy and at-risk patients, iTobos proposes to undertake its first trials with one-half of at-risk patients. Early users will lie comfortably on a bed and switch sides after the initial scan, which will take only a few minutes. The patient’s individually identifying characteristics are not photographed, and a mole map of the patient is created with the assistance of a single technician.
iTobos will capture a greater resolution of each mole and incorporate various sources of information to estimate the amount of risk of melanoma for each lesion, unlike prior technologies that fully mapped the surface of a human’s skin for mole categorization.
The goal of the project
The goal is to develop a machine that captures the same image resolution as a dermoscope that dermatologists are used to analyzing but does so automatically using robotic camera movement and artificial intelligence to identify and crop each lesion across the body while masking personally identifiable features.
The Real Challenge
Melanoma detection is difficult, but an early diagnosis can save lives.
Let’s take the above image into account. The first two are normal moles while the other two are dangerous enough to lead to death. Even the most experienced eye can be fooled by the differences in appearance that skin lesions have between age groups and races; what may appear worrisome to an east-asian 30-year-old may appear normal to a 50-year-old Southern European.
The benefit of AI is that it can condense the expertise of dermatologists all around the world into a single software.
This information is referred to as training data, and it is made up of photos and comments that indicate a doctor’s assessment of the mole. When confronted with a scenario that has comparable characteristics to a previous example, AI learns to condense this combined information into neural networks, which might evoke memories of the training data.
Significance of Accurate Training Data
It’s no easy task to curate training data for any medical application. If there are too many benign lesions, the AI will become biased in their favour.
If you incorrectly tag your demographics, the AI may dismiss information it can’t see or analyze, such as the patient’s age or ethnicity. V7’s technology is being utilized to accurately capture both demographics and 16 lesion kinds by applying tags to every patient and lesion. V7 is a London-based AI startup founded in 2018 that develops an online training data platform to automate the process of AI data annotations. V7 helps iTobos in developing AI software and data to detect moles.
People from sunny regions of the world see dermatologists more frequently, hence the populations that go for mole scans are quite uneven. As a result, it’s critical to train the technology on representative samples of these folks.
V7’s AI will semi-automatically spot every mole across the 3,000 images captured by iTobos per patient, generate a cropped version, and human labellers will ensure that the segmentation around each mole is perfect and does not include “distractors” such as clothing, tattoos, or jewellery that could bias the AI towards the wrong result.
The Future of iTobos and V7
Aberto Rizzoli, the CEO of V7 states that “V7 is thrilled to be part of a project that will change many lives and advance our progress towards eradicating cancer. To support the project we are developing new technologies to combine the consensus of multiple dermatologists with that of multiple AI models, and enabling doctors to visualize the massive scale of data produced by iTobos across multiple timeframes to assess the changes appearing in lesions between scans”.