Time-stage driven pathfinding framework for optimized medical treatments

Figures & data

Table 1. Features dataset example for a single patient in a stage before averaging

Timestep	Feature 1	Feature 2	$\cdots$	Feature 10
$t_1$	value 1	value 1	$\cdots$	value 1
$t_2$	value 2	value 2	$\cdots$	value 2
$⋮$	$⋮$	$⋮$	$⋮$	$⋮$
$t_{10}$	value 10	value 10	$\cdots$	value 10

Figure 1. DAG visualizations for the diagnoses.

Figure 2. Deterministic tree.

Figure 3. Non-deterministic tree.

Figure 4. Deterministic graph.

Table 2. Shortest path examples

Search	Diagnosis	Graph Type	Cost Method	Path
A Star	PNEUMONIA	DAG	1	‘start:0’ $\to$ ‘t1:5’$\to$
				‘t2:35’ $\to$ ‘t3:9’
A Star	CONGESTIVE HEART FAILURE	DAG	1	‘start:3’ $\to$ ‘t1:0’ $\to$
				‘t2:0’ $\to$ ‘t3:0’
A Star	DIABETIC KETOACIDOSIS	DAG	1	‘start:4’ $\to$‘t1:4’ $\to$
				‘t2:5’ $\to$ ‘t3:2’
Dijkstra	PNEUMONIA	Tree	3	no solution path
		Deterministic
Dijkstra	SEPSIS	Tree	3	‘start:0’ $\to$ ‘t2:2:0.84:cAV’ $\to$
		Non-Deterministic		‘t2:18’ $\to$ ‘t2:7’ $\to$ ‘t2:14:0.88:jKp’
Dijkstra	DIABETIC	Graph
	KETOACIDOSIS	Deterministic	2	‘start:1’ $\to$ ‘t1:5’ $\to$ ‘t2:5’

Table 3. Evaluation results

Diagnosis	Results
PNEUMONIA	72.22%
CONGESTIVE HEART FAILURE	61.11%
SEPSIS	61.11%
DIABETIC KETOACIDOSIS	80.0 %

Figure 5. DAG evaluation.

Figure 6. Tree evaluation.

Figure 7. Graph evaluation.

Figure 8. API-based serverless path recommendation pipeline.