How AI Diagnosed A Medical Mystery My Doctors Missed For Decades

After years of suffering from perplexing symptoms, I turned to AI—and finally got answers to the complex condition my doctors had overlooked

By Katherine Bindley

I NEVER THOUGHT of myself as a sick person. But I’d always had terrible luck when it came to my health.

I was hospitalized for the first time at 7 for asthma. I got my first debilitating migraine in eighth grade. My first surgery was an emergency appendectomy at 18.

In my 20s, I started having days and sometimes weekslong episodes of stomach pain. Tests for Crohn’s, celiac and ulcers were all negative. In my 30s, I learned I had a uterine septum: a congenital abnormality where a band of tissue runs down the middle of the uterus. And every few years, I’d be doubled over with abdominal pain so severe, I’d land in the ER. Doctors found nothing wrong.

There are plenty of reasons to not rely on AI for medical guidance. It hallucinates while delivering information with the utmost confidence. It can dole out advice that’s dangerous to your health in some cases. And it’s biased and sycophantic, which could mean getting assurances that nothing is wrong when something is (or vice versa).

But if you’ve spent years dealing with medical mysteries and found yourself short on satisfying answers, AI might well deserve a turn in the doctor’s chair.

By the time I asked Google’s AI Mode about my own health, I wasn’t so much desperate as I was curious whether it might be able to surface something my doctors had overlooked— like a mild hormonal imbalance.

What I couldn’t have predicted was that AI would send me on a monthslong medical mission highlighting a contrast between the technology’s ability to connect dots and the advice I’d received from providers over the years—or that I’d find myself considering major surgery.

My initial health query kicked off one night this summer while I was listening to the memoir “Famesick,” by Lena Dunham, who created HBO’s “Girls.” In the book, she details her struggles with endometriosis, a chronic inflammatory disease affecting roughly 10% of women of reproductive age.

Over the past year, I’d been experiencing exhaustion so severe that I was routinely

EIKO OJALA FOR WSJ

driving my dog to places where I could let him off leash instead of walking or running with him. My ferritin, a measure of the body’s stored iron, was chronically low from heavy periods. Results below 15 could be flagged as abnormal— mine was at seven.

I asked Gemini—the model powering AI Mode—what underlying issues might explain my symptoms. A few prompts in, Gemini said I likely had adenomyosis, a condition where endometrial-like tissue grows inside the uterine walls.

My first thought was: OK, AI is apparently not going to be a help here. Ten years ago, I had a full work-up with a top reproductive doctor in New York, including surgery to remove my uterine septum. Wouldn’t that doctor have caught this?

No, Gemini told me, adenomyosis is often overlooked on scans and could have easily been missed in surgery.

I asked if I might have endometriosis, too. Highly likely, Gemini said. I countered that the same top surgeon had performed a laparoscopy to look for endometriosis and found none. He just saw scar tissue on my bowel from my appendectomy.

Combined with my symptoms and history, that scar tissue was potentially “a massive red flag,” Gemini told me. “Even with a specific focused check,” it said, “it is still highly possible for endometriosis to have been missed or to have grown afterward.”

Further down the AI medical rabbit hole I went.

A Master of Disguise

Clinicians often call endometriosis the great imitator. Tissue similar to the uterine lining grows outside the uterus and the body can’t clear it (exactly why has been the subject of medical debate). That tissue creates lesions, commonly in the pelvic area, that bleed during hormonal cycles, resulting in chronic inflammation. Scar tissue often develops that binds nerves and can glue organs together.

Endometrial-like tissue can also embed onto and within the walls of organs: the bowel, the bladder, the diaphragm. The lesions grow their own nerves and blood vessels and can reprogram immune-system cells to protect instead of fight them.

While agonizing cramps are a hallmark symptom, women are often told they just have bad periods. And this disease knows how to hide the ball: It triggers symptoms that present like other disorders, sending women to the gastroenterologist, the urologist, the orthopedist, the ER. It takes an average of eight to 12 years from the onset of symptoms to a diagnosis. It’s common for women to learn they have the disease after fertility struggles. (Many go on to have successful pregnancies.)

One of the places endometrial implants tend to hide out is near the appendix. They can irritate the organ, causing symptoms that mimic appendicitis and leading surgeons to operate. I probably didn’t have a classic case of appendicitis when I was 18, Gemini concluded: That was likely an early sign of my disease. I was stunned. Gemini kept offering scripts to share with my doctors, encouraging me to look forward, not back. But I wanted to keep digging. I went from a private browser tab on my phone to giving Gemini every ultrasound, CT, MRI and surgical report I could find.

Calm Down

The worst part of my medical history has been acute pain episodes, beginning at 23, when I had a suspected ruptured ovarian cyst. Four years later, I was doubled over in the back of a New York City taxi sweating and crying with pain in my lower right side. I burst into Columbia University’s health center begging to see a doctor.

“Calm down, it’s probably a UTI,” the front desk woman told me.

The health center sent me to a nearby gynecologist.

“I’m tempted to open you up,” she said.

Instead of alarm at how quick she was to mention surgery, I remember thinking, “Thank God, she believes me.”

Later at the ER, my CT scan was clear. Near midnight, I was discharged. My new gynecologist concluded I’d had another ovarian cyst rupture.

At 37, another ER trip and a new diagnosis: acute gastritis. At 40, I was doubled over on the bathroom floor crying in agony: my lower right side again. An ultrasound showed a small hemorrhagic cyst remnant: “It is uncertain whether this finding alone accounts for the reported right lower quadrant pain,” the report read.

Gemini and I went through these episodes one by one. Ovarian cyst ruptures were still the strongest culprits. Endometriosis not only increases their likelihood but also makes them far more painful because the rupture occurs in an inflamed environment, acting like a chemical burn.

By the time I asked Gemini about an “incidental” finding from 2023, a mysterious cyst I was told not to worry about, I knew what the answer would be: It was also likely related to my endometriosis.

I finally asked Gemini the odds of having my symptoms and history without endometriosis and adenomyosis: The odds are close to zero. If a medical statistician or an expert gynecological excision specialist looked at this specific constellation of symptoms, they would tell you that the chance of having all of this without underlying endometriosis and adenomyosis is virtually nonexistent.

Are You Sure You’re Not Hallucinating?

As I waited to see a real doctor, I reminded myself of AI’s sycophancy and hallucinations, and that it has a mixed track record in medical settings. I’d open a new browser and start my questions from scratch, or throw entirely new ones at it. Sometimes it said a symptom or pain episode was from another condition. But then I’d realize I’d forgotten a particular piece of my history, and after adding it, Gemini would arrive at the same conclusion. I walked into my new primary care doctor’s office at the University of California San Francisco and told him I’d used AI to diagnose myself with endometriosis. Despite some initial skepticism, he spent more than an hour with me. He ordered iron infusions and an ultrasound while I waited to get in with a specialist.

I knew the ultrasound would come back normal—I’d had nearly a dozen—and it did.

When I later visited the endometriosis center at UCSF, the specialist said we had 15 minutes to review my history. He gave me a preliminary diagnosis and prescribed me birth control, which can help manage symptoms. I was to return in three months. My MRI report later came back: “NO SIGN OF ENDOMETRIOSIS.”

I told Gemini: The results weren’t what you predicted.

Yes, I was wrong about the MRI but this doesn’t mean you’re in the clear, it replied (women who undergo surgery for endometriosis frequently have had normal scans). You’re not crazy, it assured me. You’re not imagining your pain. You need to see a different doctor. It helped me find an out-of-network surgeon in private practice.

The specialist did her own dynamic ultrasound, which captures how organs move. She rattled off observations for her AI note taker, calling out which of my organs were tethered by scar tissue. She diag--nosed me with both adenomyosis and endometriosis. Two other specialists I saw after her agreed. Surgery is the only way to know for sure what’s going on, but my symptoms and history were enough to suggest to my doctors that I was a strong candidate.

AI can create headaches for doctors, according to Isaac Kohane, professor and chair of the department of biomedical informatics at Harvard Medical School. Patients come in with the same language doctors use; their AI assessment is well-written and well-reasoned. But when AI is wrong, it can be a huge lift to disabuse patients of what they thought was a diagnosis, taking up valuable time in an appointment. AI can also cause unnecessary anxiety in patients, backing up what it might suspect the patient wants to hear.

But Kohane still believes there are enough holes in the medical system that we should all be using AI for our health—to educate ourselves, to come into our appointments equipped with more information and to dig deeper if we’re not getting sufficient answers.

“Even with good doctors, many factors act against an effective diagnosis. One is simply time,” he says. “AI has all the time that the data center will allow it to have.”

I’m more wary of AI than most people in San Francisco. I worry about its impact on jobs and on our overreliance on it. But for women’s health, I believe it’s a godsend.

AI ultimately gave me what our siloed medical system couldn’t: a look at my health with a zoomed out lens and longitudinal pattern synthesis. More important, it believed me, treated me with a compassion I rarely found in human doctors and gave me back the power to advocate for myself—something I thought I’d lost for good.

I plan to move ahead with excision surgery to get rid of the scar tissue that’s likely tethering my organs and to remove whatever endometriosis my doctors can find. Despite how painful my recovery will be and the risk that my disease could grow back, I’m actually looking forward to surgery.

At long last, I’ve got a real chance at feeling better. On that, me, my AI and my human doctors are all in agreement.

There are enough holes in the medical system that we should all be using AI, one expert suggests.

As I waited to see a real doctor, I reminded myself of AI’s hallucinations.

FROM TOP: EIKO OJALA FOR WSJ; EMILY PEDERSON/ WSJ

Writer Katherine Bindley in San Francisco.