The Blue Box

If dogs can smell cancer, why can't we?

Designed after a dog's olfactory system and olfactory neurons, The Blue Box is a 3D-printed box that contains 9 chemical sensors that are specifically sensitive to some breast cancer biomarkers – i.e. some molecules whose presence or absence in one's physiology is correlated to the condition of breast cancer. The Blue App is a smartphone application that accompanies the user throughout the screening process. It can be downloaded from Google Play [https://play.google.com/store/apps/dev?id=8916273439374129777].

 When a urine sample is introduced inside The Blue Box, the aforementioned sensors will perform a change in voltage depending on the nature of the chemical compounds present in the urine sample. During the next 30 minutes, the captured signal will be sent to the cloud via WiFi, where our AI algorithm is allocated. Once AI reaches a diagnostic, it will be sent back to the user's phone and displayed in the app.

Breast cancer (BC) is the most frequent tumor affecting women worldwide in low- and middle-income countries. The American Cancer Society estimated BC accounts for 30% of all US cancers. However, research dedicated to it is not proportional to its incidence. Actually, the NIH recognized women as underrepresented in medical research. 

Indeed, most national BC screening programs use the mammogram as their gold standard, which cannot identify tumors in women with fibrocystic breasts (healthy lumps). Shockingly, around 60% of women worldwide have fibrocystic breasts. Moreover, the CDC stated that only 65% women attend their screening, potentially resulting in 1/3 BCs being detected late. Reasons for women skipping it are multiple: pain (41% of interviewees), no healthcare insurance (20% of US women) according to the Journal of Women’s Health. 

Additionally, although mammogram's radiation dose is not substantial enough to be considered harmful, its cumulative dose has risks. The USPSTF recommends screening for ages 50-74 only. However, BC is the leading cause of death by cancer in women under 40 years of age, especially among African-American women. They present a denser mammary tissue, thus decreasing mammograms' sensitivity and specificity. In addition to that, their tissue is more sensitive to the mammogram's radiation dose. 

In the US, 451,936 BCs are detected yearly, 150,000 of which are detected late-stage because of mammogram absenteeism or inefficiency of current methods.

If BC is detected in an early stage, its metastasis risk is minimal as well as the likelihood of a subsequent metastasis, patient suffering and death. If detected in a later stage, however, patients will require treatment and potentially a mastectomy, which has a notorious impact on women's mental health. These interventions have an approximate cost of $25,000 per patient. It is for this reason that, for the US population that does not have health insurance (20%), the problem remains unsolved: They do not undergo the screening because they cannot afford it. They are therefore more likely to be diagnosed with advanced BC, needing a treatment that they cannot afford. If The Blue Box was placed in community health centers screenings were offered for $35 (5% of the price of a mammogram), the problem would end.

Furthermore, in 2017, the World Health Organization (WHO) published the “WHO Position paper on mammography screening.” stating an urgent need for a new radiation-free and sensitive BC screening solution. This solution can be reached by changing the focus: Focussing on metabolomics instead of imaging.

So far, we have conducted two pilot studies.

> Pilot Test 1: Ability to detect LATE-STAGE breast cancer:

Judit's bachelor thesis proved the main hypothesis that “the metabolite analysis performed by The Blue Box is sufficiently significant to enable class prediction among control subjects and metastatic breast cancer patients”. 90 urine samples were collected from control subjects and breast cancer patients at Sant Joan University Hospital in Spain. Class prediction was achieved with a sensitivity of 75%.

> Pilot Test 2: Ability to detect EARLY-STAGE breast cancer:

In July 2021, we started clinical studies in two hospitals. To date, we have used The Blue Box to analyze over 110 urine samples from >110 women who have received a suspicious mammogram diagnosis. We have also collected health and lifestyle-related data from them. Later on, their doctors will tell us their biopsy results (early-stage breast cancer or control), and we will use this data (the ground truth) to train our AI algorithm.

Our next goals (in 12 months vista) are:

- To gather enough (300) samples to train our AI-based classification algorithm.

- To test and validate our technology: To determine our technology’s classification rate, sensitivity and specificity.

Billy designed the Bluetooth protocol for each sensor to report their status to the Raspberry Pi in a master/slave structure. I coded the ESP32 (microprocessor) logic and sensing protocol. Each one of us had very different insights and areas of expertise: Billy could set up the whole connectivity workflow within minutes and fix any bug relatively easily. I, on the other hand, enjoyed designing the logics of operation and finding a potential real life application.

As early as Roman times, medicine has paid special attention to human physiological odours, e.g., uncontrolled diabetes was associated with a sweet acetone odour. However, human odour studies did not meet the oncology field until April 1989 when Dr. Hywel Williams and Dr. Andres Pembroke from King’s College Hospital, London reported a case in The Lancet about a Collie-Doberman owner who attended their practice. She claimed her dog was showing increasing interest in a mole in her leg. The mole was then shown to be pathological and removed, saving the patient’s life. 

This proved that cancer produces metabolic changes in human physiology. Since then, an increasing number of publications have been seeking the definitive set of breast cancer biomarkers. Nonetheless, there does not exist a general consensus among authors yet.

In this project, a different approach to the challenge was followed – based on engineering rather than experimental medicine. The key is focussing on the proportion between VOCs (volatile organic compounds) rather than on VOCs themselves. Like the human nose, the implemented software will respond in concert to a given set of odours -a pattern or “smellprint”- which is analysed, compared with stored patterns, and recognized.

We are approaching this question as per our KPIs, which we have defined based on the key milestones to ensure that our product safely reaches the market. 

Our to MAIN KPIs are related to our 2 products: the Box and the App:

KPI 1: Confusion matrix of our classification algorithm, which defines the sensitivity and specificity of our classification algorithm.

By February 2022, over 150 patients should have participated in the study and we should therefore have some initial approximation of the classification capacity of the device when applied to early-stage breast cancer. To date, we have already used The Blue Box 110 times to record the smell of 110 patients.

KPI 2: Number of users interacting with the app and using it recurrently: number of downloads, new downloads, number of registered users, number of minutes spent per week in the app, number of users using the app weekly.

By March, we should have our first users.

Our SECONDARY KPIs focus on our virtual community, -i.e. our potential future users - and their degree of knowledge and satisfaction towards our product. 

KPI 3: Number of engaged users in our virtual community (overall Instagram, Twitter, LinkedIn and TikTok interactions plus website blog readers). We should have performed numerous iterations in both our business model and our product specifications because we have spent a lot of time talking to prospective users.

Objective 1: HARDWARE. To optimize our hardware. 

To assess and optimize how good the odor of a urine sample (the “smellprint”) is captured. This goal consists of the following tasks:

      1.1. To calculate each sensors’ detectability for each specific biomarker

        1.2. To optimize the sensing unit in collaboration with Z&P (sensor development company in Norway, elaborated below): To replace the two worse-performing sensors with two new sensors.

Objective 2: APP. To deliver an app MVP and have first users. Ongoing, app released in Feb 2022.

      2.1. To optimize data sending from hardware to app and from app to server (for both sensor and demographic data) [Update: already implemented].

      2.2. To develop the iOS version of the current Android app

The expected outcome is a new version of the application, both in Android and iOS that is attractive to the public even before The Blue Box reaches the market so that we can start having customers early on.

Objective 3: ALGORITHM. To keep optimizing our classification algorithm. 

     3.1. To keep gathering data from women with and without breast cancer – both their “smellprint” and demographic data

     3.3. To use sensor data and demographic data to train the classification algorithm

During Sept 2020 - June 2021, our technology was developed and optimized to a functioning prototype of TRL 5 in collaboration with UCI. From July 2021 onwards, our strategy has changed: the hardware is now at three hospitals in Spain collecting data from patients while we devote 70% of the startup’s resources to customer discovery and market access definition. In fact, we have been part of UC Launch (wining a 2nd position on demo day), a program that has guided us in customer discovery. 

However, we do spend 30% of our time continuing our collaboration with UCI and feeding the SW algorithm with data to train it as well as re-defining HW specifications as per the feedback received from customer discovery.

By late 2022, we will have finished our preclinical studies and be ready to start focusing on regulatory affairs. We will then apply for the US Food and Drug Administration (FDA) approval. In the US, the device is expected to fall under the category "class III medical device" because it is intended to provide an oncological diagnosis. 

Parallel to that or afterwards (depending on funds availability) we will apply for the Conformitè Europëenne (CE) mark and the European Union Medical Device Directive (MDR) clearance. The device is expected to fall under the in vitro Diagnostic Medical Devices under Regulation (EU) 2017/746, class (h) "Devices intended to be used in screening, diagnosis or staging of cancer" according to the Medical Device Coordination Groups Document, MDCG 2020-16. 

We have had an initial conversation with our advisor at Wayfinder (our incubator at UCI) and expert in medical device regulatory and clinical affairs, who could serve us as a mentor to optimize our regulatory affairs journey. Approximately, FDA approval will last 12-15 months and cost about $150k.

I still remember the moment I (Judit) realised that Billy was the perfect person to found a startup with. Back in California, during our masters, we were given a kit of Raspberry Pi's, various microprocessors and several sensors. In one summer afternoon we designed a "Smart Bed" with it.

Billy designed the Bluetooth protocol for each sensor to report their status to the Raspberry Pi in a master/slave structure. I coded the ESP32 (microprocessor) logic and sensing protocol. Each one of us had very different insights and areas of expertise: Billy could set up the whole connectivity workflow within minutes and fix any bug relatively easily. I, on the other hand, enjoyed designing the logics of operation and finding a potential real life application.

Two temperature sensors were set to be attached to the pillow, measuring the patient's temperature in real time. Other sensors were meant to be inside a mattress, sensing patient sleeping pose, to prevent ulceration upon continuous posture. 

I learn a lot from Billy, both at a professional and personal level. I think of us as a team who potentiates each other and who has each other's back. In fact, we typically ask for each other's advice for personal matters outside of work as well.

Unfortunately, due to a medical condition, Billy was forced to lower his implication to the startup when the minimal viable prototype was finished. We then welcomed Joan to the team. He was a CTO in a previous startup, and he carried out the design, production, medical regulations, and industrialization of a medical device to detect sepsis. Thanks to his expertise, we are now a more multidisciplinary team.

*NOTE: Judit was 24 when applications opened and 25 by the deadline. We hope it is not inconvenient. 

I still remember the moment I (Judit) realised that Billy was the perfect person to found a startup with. Back in California, during our masters, we were given a kit of Raspberry Pi's, various microprocessors and several sensors. In one summer afternoon we designed a "Smart Bed" with it.

Two temperature sensors were set to be attached to the pillow, measuring the patient's temperature in real time. Other sensors were meant to be inside a mattress, sensing patient sleeping pose, to prevent ulceration upon continuous posture. 

Now, I learn a lot from Joan, both at a professional and personal level. I think of us as a team who potentiates each other and who has each other's back. In fact, we typically ask for each other's advice for personal matters outside of work as well.

In April 2015 my high school teacher explained how the human body turns a piece of bread into 265 kcal with slightly no energy loss. In fact, the 2.5μm-sized  mitochondrion is able to do so by working at an efficiency that has never nor is it ever likely to be reached by any other human-designed machine. This thought from back in my high school years awakened some kind of fascination in me that still continues to amaze me: I will never aim at reaching the degree of perfection and design of biology, but biology is what I will observe and learn from every time I want to engineer a solution.

I am someone who believes that in order to achieve great design, one must first observe nature and model it through hardware and software. And hence my core passion for this project. I want to diagnose cancer mimicking the olfactory cortex of the dog – learning from biology to create better answers. 

Since then, I started paying attention to what women were complaining about. I have always worked very close to both oncologists and gynaecologists who could benefit from a Blue Box for their patients, but I have also talked a lot to radiologists who are so used to mammography that present themselves reluctant to such a change of approach. Right now, I have a head of gynaecology and two heads of breast cancer prevention as local PI of my study and I work with them weekly.

As a biomedical engineer, I have always been fascinated about developing my skills to act as a "medical translator": I like to observe in what ways medicine can advance and translate a need into a software and hardware-based innovation.

Some years ago, I met a therapist specialised in childhood autism. I was curious to know her strategy to try and develop the kid's ability to recognise emotions in their peers and empathise with them. She was using some paper-based photos of daily life situations that the kids had to match with emotion names. However, she was frustrated because her tiny patients had a hard time concentrating and only wanted to play computer games. 

In fact, her frustration was an opportunity: I asked her to attend a therapy session the next week: There was something all her patients had in common: They had trouble keeping focus on some task for a long time, they got frustrated because of that, and then had an ever harder time keeping their focus. After some meetings, I convinced the therapist to let me translate her image-based game into an app. I asked my sister and cousins to model for me while simulation daily live situation and a variety of emotions. I then created an app-based game with a punctuation system for the kids to engage in the task. I also developed 2 more games within the app and a ranking page to compete with other users.