Sign In

The need for AI in cancer care has never been clearer

Sci­en­tists have warned there could be thou­sands of excess deaths in the UK in the com­ing years due to delays in can­cer diag­no­sis and treat­ment dur­ing by the coro­n­avirus cri­sis.

The pan­dem­ic has meant rou­tine screen­ings, and urgent refer­rals and treat­ments, have been delayed or can­celled, lead­ing to a back­log of patients. Researchers at the Health Data Research Hub for Can­cer exam­ined data from eight hos­pi­tal trusts and found that, in a worst-case sce­nario, if delays con­tin­ue, there could be up to 35,000 addi­tion­al can­cer deaths with­in a year.

But arti­fi­cial intel­li­gence (AI) could be a solu­tion. Over the past decade, AI has emerged as a lead­ing tech­nol­o­gy with the poten­tial to aid the med­ical com­mu­ni­ty, from speed­ing up diag­nos­tics and improv­ing accu­ra­cy to improv­ing patient out­comes and hos­pi­tal effi­cien­cies.

Diagnosis benefits from AI in cancer care

One chal­lenge in oncol­o­gy is ear­ly diag­no­sis. The NHS is good at can­cer treat­ment, but the dis­ease is not always detect­ed ear­ly enough and this could be exac­er­bat­ed by COVID-19. AI in can­cer care could help find inno­v­a­tive ways to reach indi­vid­u­als faster and iden­ti­fy ear­ly signs of can­cer. AI cancer pullstats

“There’s inter­est in how we might be able to use AI along­side per­son­al data devices peo­ple have, such as Fit­bits,” says Dr Jodie Mof­fat, head of ear­ly diag­no­sis at Can­cer Research UK. “These might be able to sig­nal when things are chang­ing soon­er than might oth­er­wise have been detect­ed. That’s going to help you to iden­ti­fy peo­ple who might need some fol­low up or who are at risk in some way.”

AI can be used to help diag­nose dif­fer­ent types of can­cer, such as breast, prostate or skin can­cer. John Loder, invest­ment direc­tor for Nes­ta, an inno­va­tion foun­da­tion, cites Skin Ana­lyt­ics, a health­care com­pa­ny that is work­ing with clin­i­cians in der­ma­tol­ogy and using AI to deliv­er bet­ter patient out­comes. “Der­ma­tol­ogy is a lead­ing exam­ple because it’s an under­staffed area and a com­mon can­cer,” he says.

AI can diagnose more in less time

Data visu­al­i­sa­tion and machine-learn­ing tech­niques can also help extract clin­i­cal­ly use­ful knowl­edge from a huge amount of data. Researchers at the Insti­tute of Can­cer Research have devel­oped a large-scale AI data­base to aid drug dis­cov­ery research. It brings togeth­er data across biol­o­gy, chem­istry, phar­ma­col­o­gy, struc­tur­al biol­o­gy, cel­lu­lar net­works and clin­i­cal anno­ta­tions.

“With this we were able to start devel­op­ing AI tech­nol­o­gy that would allow us to answer key ques­tions. For exam­ple, to come up with a risk pro­file for dif­fer­ent patients,” says Pro­fes­sor Bis­san Al-Lazikani, head of data sci­ence at the Insti­tute of Can­cer Research.AI oncology pullstats

A study, pub­lished in Nature jour­nal in Jan­u­ary, found AI is more accu­rate than doc­tors in diag­nos­ing breast can­cer from mam­mo­grams. Cur­rent­ly in can­cer screen­ing in the NHS, two radi­ol­o­gists must analyse each wom­an’s X‑rays.

Researchers from Google Health and Impe­r­i­al Col­lege Lon­don test­ed a com­put­er mod­el on near­ly 29,000 women and showed it to be as effec­tive as human radi­ol­o­gists. The AI mod­el was as good as the cur­rent dou­ble-read­ing sys­tem of two doc­tors and was bet­ter at spot­ting can­cer than a sin­gle doc­tor, they found.

Dr Hutan Ashrafi­an, one of the lead researchers and an AI expert at Impe­r­i­al Col­lege Lon­don, says AI can be par­tic­u­lar­ly help­ful in sup­port­ing diag­nos­tics. “It’s a low hang­ing fruit for AI at the moment, par­tic­u­lar­ly in sup­port­ing radi­o­log­i­cal diag­nos­tics because the data you get is dig­i­tal,” he says.

AI can improve efficiency in cancer centres

Before AI can be used in a clin­i­cal set­ting, it needs to be tri­alled and eval­u­at­ed prop­er­ly. At the moment, researchers at Google Health and Impe­r­i­al are organ­is­ing tri­als across the coun­try to see if they can repli­cate their results.

If the results are pos­i­tive, the AI algo­rithm could save a lot of time and sup­port health­care resources. “What we found was quite pow­er­ful,” says Ashrafi­an. “In one fell swoop, it could reduce the NHS work­load on a nation­al lev­el.”

AI is sup­port­ive and works along­side peo­ple. I don’t see it ever replac­ing doc­tors.

It might not be as glam­orous, but clin­i­cians say AI in can­cer care could be par­tic­u­lar­ly use­ful in improv­ing hos­pi­tal admin­is­tra­tive effi­cien­cy. For exam­ple, AI-based tools could help man­age appoint­ments, look after patient path­ways and track patients.

Dr Car­o­line Rubin, vice pres­i­dent for clin­i­cal radi­ol­o­gy at the Roy­al Col­lege of Radi­ol­o­gists, says: “It can real­ly help to make the sys­tem flow more effi­cient­ly. It’s huge­ly impor­tant that patients don’t get lost in sys­tems. It’s not some­thing that is cur­rent­ly huge­ly effi­cient. It’s also quite labour inten­sive.”

Obstacles to implementing AI in cancer care

Rubin says it’s not so much reluc­tance to adopt new tech­nolo­gies that may have slowed devel­op­ments down so far, but the need for prop­er test­ing, safe­ty and reg­u­la­tions to ensure tech­nol­o­gy is safe and as effec­tive and effi­cient as pos­si­ble.

It’s also essen­tial that any new tech­nol­o­gy fits eas­i­ly with­in a hos­pi­tal set­ting. “How they are embed­ded in work­flow is huge­ly impor­tant,” Rubin adds. “We can’t have some­thing that slows us down; it has to be effi­cient and sup­port­ive.” Patient con­sent must also be con­sid­ered, as well as ensur­ing data is kept safe­ly with­in the NHS.

The Care Qual­i­ty Com­mis­sion recent­ly released a report into deep-learn­ing in diag­nos­tic and screen­ing ser­vices. A key find­ing was that “more clar­i­ty [is need­ed] on how hos­pi­tals should imple­ment machine-learn­ing devices with­in clin­i­cal path­ways to ensure high-qual­i­ty care”.

There is a mas­sive accel­er­a­tion and we’re going to end up in a place where use of AI is a nat­ur­al part of med­i­cine

Where there is reluc­tance, it could be that imple­ment­ing AI tech­nolo­gies can be expen­sive, requir­ing fund­ing and invest­ment. From the point of view of pub­lic per­cep­tion, peo­ple trust doc­tors and some may be wary of machines. “But AI is sup­port­ive and works along­side peo­ple,” says Rubin. “I don’t see it ever replac­ing doc­tors.”

Covid-19 and cancer care

Ashrafi­an adds: “The pan­dem­ic has cat­a­pult­ed the need for dig­i­tal.” Tech can facil­i­tate remote prac­tice as with AI tools there’s no need for peo­ple to vis­it hos­pi­tals where they could be exposed to COVID-19.

“The way I see it, we are absolute­ly at the inflec­tion point,” says Al-Lazikani at the Insti­tute of Can­cer Research. “There is a mas­sive accel­er­a­tion in this and we’re going to end up in a place where use of AI is a nat­ur­al part of med­i­cine.”