The Era Where AI Prepares Medications and Treatment Plans — What's Happening in the Field of Cancer Medicine

The Era Where AI Prepares Medications and Treatment Plans — What's Happening in the Field of Cancer Medicine

When you hear the word AI, what comes to mind? Convenience, efficiency, a sense of the future—and at the same time, an inexplicable fear and the risk of shaking society. In fact, since generative AI rapidly became widespread, public opinion has been persistently more "anxious than hopeful." Weaponization, misinformation, privacy violations, job displacement. These concerns are by no means exaggerated and have been accumulating as real incidents.


However, the same AI is working as an "accelerator to help people" in a completely different field. The stage is cancer medicine. Rather than flashy demos or advertisements, AI is beginning to absorb the cries of doctors and researchers in the gritty medical field who lament, "There's not enough time," "There's not enough manpower," and "There are too many manual tasks."


1) At the entrance of drug discovery, AI "finds"

The path to new cancer drugs, especially antibody drugs, is long until you "hit the jackpot." Finding targets, creating candidate molecules, verifying efficacy, and advancing to clinical trials require enormous trial and error.


The article introduces a case where GV20 Therapeutics used AI to find an antibody drug candidate, GV20-0251, targeting the immune checkpoint IGSF8, and administered it in a Phase I trial for patients with advanced solid tumors who had exhausted standard treatments. There were patients who showed tumor shrinkage or disease stabilization, indicating "initial positive responses."


What's important here is that this is not a story of "AI designing a magical drug from scratch." The explanation in the article is quite the opposite; AI analyzes tumor data from patients and plays the role of finding antibodies possessed by "resistant immune cells" that were actually generated in the body. In other words, AI is depicted as an excellent scout that finds traces of natural "battles" and organizes them into a form that is easy to reuse.


In the world of drug discovery, this flow of "exploration with AI → clinical application" is symbolic. Research on the target IGSF8 itself and trial information on GV20-0251 are reported externally, and the article's perspective of "AI advancing clinical trials" is not merely speculative.


2) In radiation therapy, AI "prepares"

Another pillar of cancer treatment is radiation therapy. Here too, AI enters in a form that is inconspicuous but highly effective.


Radiation therapy is a "precision task" of delivering the maximum dose to the tumor while avoiding exposure to surrounding normal organs. The first step supporting this precision is the task of drawing contours to distinguish tumors and organs based on CT images and creating a 3D anatomical map (auto-segmentation). Traditionally, medical professionals have drawn lines by hand on hundreds of image slices for each organ. It takes time and can vary between physicians. AI aims to assist and automate contour creation, reducing the burden and standardizing the process.


Hearing "AI saves time" might sound trivial. However, in medicine, that "saved time" is redirected to patient explanations, decision-making support, and safety checks. It is time directly linked to the quality and safety of treatment.


3) AI "drafts" to save from the reply nightmare

Cancer care doesn't end with the treatment itself. Inquiries from patients, consultations about symptoms, medication instructions, anxiety over test results... Medical professionals can burn out just from handling messages.


The article describes a situation at a large facility where the radiation oncology department receives a large volume of patient messages daily, taking 24 to 48 hours to respond. They introduced a system where AI drafts replies, and doctors review and send them. Even if each response is shortened by a few minutes, the cost and burden significantly decrease when the volume is large.


This kind of system is not about "replacing doctors with AI," but more about "AI filling in the blank page before doctors write." It's faster to safely revise a draft than to create a text from scratch. The faster it gets, the sooner patients can feel reassured.


4) Towards "finding" medicine: AI picks up complications

Medical records are voluminous. A single patient is linked to an enormous amount of clinical records, and complex complications or side effects cannot be organized with just one checkbox. In the past, extracting patients with specific complications could involve doctors reading thousands of pages of records and taking a year to create "correct data." The article touches on the possibility that specialized AI can handle similar tasks in a short time and may even be more accurate than traditional manual data.


What emerges here is that AI's role is shifting from being an "entity that makes diagnoses" to an "entity that reduces oversights and increases discoveries." It unearths information before humans make "judgments." For the medical field, this is powerful.


5) The remaining barriers: "black box" and trust

Of course, the entry of AI into medicine doesn't solve everything. The article repeatedly emphasizes that medicine is heavily regulated, with high procedural barriers like FDA approval, and above all, the "black box nature" of not being able to explain "why that conclusion was reached" can undermine trust.


AI learns through reward design and sometimes optimizes in the direction of "looking good." In medicine, that "looking good" can be fatal. That's why a realistic compromise is for medical professionals to verify AI outputs and collaborate in a way where they take responsibility.


Reactions on SNS (Summary of tendencies seen in actual posts)

Regarding topics close to the theme of this article (RadOnc-GPT, AI in radiation therapy, GV20-0251/IGSF8, etc.), the following reactions are notably divided on SNS.

 


A. "This is the AI we've been waiting for" group (Expecting reduced burden on medical professionals)

Posts touching on LLM and automation in the field of radiation therapy often carry the tone of "If accuracy and speed can coexist, the field will be saved." Especially, the idea of applying AI to "time-consuming chores" like drafting patient messages or extracting outcomes from records is generally well-received.


B. "Amazing but scary" group (Explainability, responsibility, medical safety)

On the other hand, cautious opinions like "It's amazing when it's right, but who takes responsibility when it's wrong?" "Is the handling of medical data really okay?" and "It's troubling if the basis for conclusions can't be traced" are also strong. This connects to a broader distrust of AI in general, not just in medicine.


C. Drug discovery group is "waiting for clinical data" (Evaluating calmly)

Regarding topics like AI-utilized drug discovery such as GV20-0251, while there is excitement over company and researcher announcements (ASCO presentation notices, trial progress, etc.) with comments like "Interesting" and "New targets are hot," the investment and research communities often take a more measured view, noting "Phase I focuses on safety. The true value is yet to come."


D. Community close to the field is "implementation is tough" group (Operation, data, responsibility demarcation)

In technical and medical communities on platforms like Reddit, while auto-segmentation is already a "significant area," there are many "realistic voices" highlighting the challenges of actual operation, such as data preparation, differences between facilities, quality assurance (QA), and vendor selection.


Conclusion: AI is not the "main character," but a collaborator in medicine

What makes this article interesting is that it does not depict AI as an all-powerful hero but rather as an entity quietly taking over the "tedious and time-consuming preliminary processes." New drug exploration, contour creation, drafting replies, unearthing records—all are enormous tasks that precede "human judgment."


The future of medicine will likely progress more towards "AI reclaiming doctors' time" rather than "AI replacing doctors." What is needed for that is explainability, regulatory compliance, and a "design where humans can take responsibility"—in other words, creating a form of collaboration.


Source URL

Related Articles