ARTIFICIAL INTELLIGENCE
Pathologist results Pathologist results Parts without AI
with AI 49:11 Pathologist A 20:28 500 400 Pathologist B 22:09400 300 48:15 Pathologist C 21:18 200 200 Average 21:18
0 00
15 15
100
the AI-enabled digital workflow was evidenced when adding Ibex Prostate results to the workflow. For the reading times in the AI arm, the trend in reading time over time also shows limited influence by how many cases have been read in a similar workflow before by the pathologist. Overall, it is noticeable, however, that the reading times are more consistent, with a much smaller difference between cases. In discussing with the pathologists after reading had finished, they confirmed that the AI insights helped to harmonise the flow of reading, especially through prefilling of measurements such as core length and tumour length.
100 Digital viewer only
30 30
Minutes 0 0
Highly accurate AI Cancer detection by Ibex Prostate was highly accurate and in alignment with the accuracy of the pathologists’ diagnoses. Pathologists supported by AI demonstrated high performance (Table 1), as evidenced by the negative predictive value (NPV, 99.9%), positive predictive value (PPV, 99.1%), and very high accuracy for cancer detection with an AUC of 0.997 (95% CI:0.994-0.999). For Gleason grading, the pathologists using the Ibex AI technology showed high levels of agreement (90%) with the adjudicator. AI Arm ICC=0.945 (95% CI:0.934-0.955). The study demonstrated that Ibex Prostate can accurately analyse slides
and even helped pathologists find and report 5% more PNI (perineural invasion)- positive parts compared to the baseline DP workflow, as shown in Figure 3. This improvement indicates that AI resulted in fewer missed PNIs, reducing the number of patients with PNIs who are underdiagnosed and go without the appropriate treatment plan.
Driving laboratory efficiency Upon completion of the reviewer study, the pathologists were surveyed in order to assess the potential of deeply integrated AI in their routine workflow. Several areas of laboratory efficiency and pathologist experience improvement with AI were
45 45
With AI
60 60
51:28 300 56:57 500 Benign Malignant
Undetermined Benign
Malignant
Undetermined Benign
Malignant Undetermined NPV = 99.9%
Malignant Malignant Malignant Benign Benign Benign
Undetermined Undetermined Undetermined
0
437 12
1,354 1*
164 68 40
107 * Part determined as benign by the adjucator. PPV = 99.1%
Above: Table 1. Pathologists supported by AI demonstrate high performance in cancer detection with high NPV and PPV values. Thus, supports the viability of this solution for use in routine practice. Left: Fig 2. Average case reading times with and without AI in minutes.
WWW.PATHOLOGYINPRACTICE.COM AUGUST 2024
37
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52
