Keywords:
Liver, Artificial Intelligence, Oncology, CT, CT-Quantitative, Computer Applications-Detection, diagnosis, Computer Applications-General, Imaging sequences, Cancer, Metastases, Tissue characterisation
Authors:
M. Mottola1, A. Bevilacqua1, V. Vilgrain2; 1Bologna/IT, 2Clichy/FR
DOI:
10.26044/ecr2019/C-0522
Results
4 out of 276 couples of features have been selected because they showed the best performance in discriminating the group of patients (i.e.,
the highest number) who will develop liver metastases (hereafter,
mets). Moreover,
all the couples selected refer to one feature,
the skewn-HPI (range [-0.5÷0.6]),
as shown in Fig.
7 that,
when coupled with mean values of BF [118÷285]ml/min/100g (Fig.
1(a)),
BV [63÷66]ml/min/100g (Fig.
1(b)),
MTT [16÷27]s (Fig.
1(c)),
TTP [30÷53]s (Fig.
1(d)) allows detecting four (ID-39,
44,
53,
87) out of six patients developing liver metastases with no false positives.
In all the couples of Fig.
7,
mets can be linearly separated and the best separating line is drawn.
At least 2 mets are always detected,
patients 53 and 44 in Fig.
7(d) with skewn-HPI coupled with BV and patients 53 and 87 in Fig.
7(b) with HPI-skewn coupled with TTP.
All together,
patients 53,
44,
and 87 are separated with skewn-HPI and mean-MTT,
while one couple of features (Fig.
7(a)) allows including one more met,
patient 39,
with a total amount of four separated mets.
Moreover,
Table 1 shows that patient 53 is always discriminated in all the couples in which skewn-HPI is associated with all the features computed (23 possible combinations).
Instead,
the frequency rate of discrimination is 52%,
30%,
and 13%,
respectively for patients 44,
87,
and 39.
As expected,
perfusion parameters within the discrimination ranges highlights a higher degree of tissue perfusion with high mean BF (202ml/min/100g) and BV (64ml/min/100g) values and low MTT (21s).
In addition,
TTP mean value of discriminated patients is equal to 37s,
10s ahead of the mean TTP of the remaining patients.