The last few years have seen significant changes in the healthcare industry due to the COVID-19 pandemic. These changes have led to increased investment in artificial intelligence (AI)-based medical devices, such as those used in telemedicine, precision medicine (remote surgical procedures), and integrative algorithms for clinical, laboratory, and imaging contexts. According to data from the US Food and Drug Administration (FDA), the number of AI devices registered for medical use has increased from 106 in 2017/2018 to 178 in 2022, with 75.2% of these devices being used in radiology. However, there are still challenges to be addressed in the use of AI in medical imaging. These challenges include:
1: SAR-CoV-2 (COVID-19):
The heterogeneous characteristics in relation to laboratory and histopathological findings and images imposed notable challenges for the implementation of AI devices, however, advances have been registered with regard to the use of AI models [2,3]:
– a: Stratification of disease severity:
– b: Algorithms based on bio-molecular models for treatment optimization.
– c: Algorithms in medical imaging to predict the clinical course of the disease and the possibility of readmission.
However, advances have been made in the use of AI models to stratify disease severity, optimize treatment through bio-molecular models, and predict the clinical course of the disease and the possibility of readmission using medical imaging algorithms.
2: Robotic Surgery:
AI devices are now being tested in real-time for surgery. One such application refers to the integration of the Da Vinci console by AI, increasing the ability to locate and classify tumors in TRUS/MR in real time of the prostate, avoiding unnecessary manipulation [4, 6].
3: Digital pathology:
AI devices have been gaining strength for segmenting nodules and classifying them in areas such as urology, oncology, etc. However, the main challenges consist essentially in the intra-inter pathologist correlation aiming at increasing the accuracy of these models [4]
4: Database diversity
Considering the development in recent years, there are still several gaps with regard to demographic diversity, in this scenario, continuous efforts have been to develop strategies that make it possible to identify AI model implementation bias, bias related to the specialist, strategy, and automation of the data acquisition, in addition to environmental and occupational factors which make it impossible for vulnerable populations to access [7].
5: AI applied to Cancer
AI in oncology has been exponential, however, there is still considerable resistance in AI decision-making, with the main perspectives in this field being the creation of pragmatic and systematic measures in AI models in oncology, these effect models can encompass the choice of data, accuracy, and integrity of these. In addition to detailed reports about care and control by remote and wearable devices, seeking to complement the maximum amount of data collected from a single patient [8].
In general, there are still many challenges to be overcome, however, consistent advances have been observed in the integration between health professionals and data scientists, favoring the association of the clinical environment and its respective difficulties with effective forms of treatment and data recording.
Citations:
[1] Lin M. What’s needed to bridge the gap between US FDA clearance and real-world use of AI algorithms. Acad Radiol. 2022; 29, 567–568.
[2] Zhou Y, Wang F, Tang J, Nussinov R, Cheng F. Artificial intelligence in COVID-19 drug repurposing. Lancet Digit Health. 2020 Dec;2(12):e667-e676. doi: 10.1016/S2589-7500(20)30192-8.
[3] Huang S, Yang J, Fong S, Zhao Q. Artificial intelligence in the diagnosis of COVID-19: challenges and perspectives. Int J Biol Sci. 2021 Apr 10;17(6):1581-1587.
[4] Teber D, Guven S, Simpfendörfer T, Baumhauer M, Güven EO, Yencilek F, Gözen AS, Rassweiler J. Augmented reality: a new tool to improve surgical accuracy during laparoscopic partial nephrectomy? Preliminary in vitro and in vivo results. Eur Urol. 2009 Aug;56(2):332-8.
[5] Karimi D, Samei G, Kesch C, Nir G, Salcudean SE. Prostate segmentation in MRI using a convolutional neural network architecture and training strategy based on statistical shape models. Int J Comput Assist Radiol Surg. 2018 Aug;13(8):1211-1219.
[6] Goldenberg SL, Nir G, Salcudean SE. A new era: artificial intelligence and machine learning in prostate cancer. Nat Rev Urol. 2019 Jul;16(7):391-403.
